2
0
mirror of https://github.com/koreader/koreader synced 2024-11-18 03:25:46 +00:00
koreader/frontend/device/input.lua

1479 lines
60 KiB
Lua
Raw Normal View History

--[[--
An interface to get input events.
--]]
local DataStorage = require("datastorage")
local DEBUG = require("dbg")
local Event = require("ui/event")
local GestureDetector = require("device/gesturedetector")
local Key = require("device/key")
local framebuffer = require("ffi/framebuffer")
local input = require("ffi/input")
local logger = require("logger")
local time = require("ui/time")
local _ = require("gettext")
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- We're going to need a few <linux/input.h> constants...
local ffi = require("ffi")
local C = ffi.C
require("ffi/posix_h")
require("ffi/linux_input_h")
2016-02-16 07:10:07 +00:00
-- luacheck: push
-- luacheck: ignore
-- key press event values (KEY.value)
local EVENT_VALUE_KEY_PRESS = 1
local EVENT_VALUE_KEY_REPEAT = 2
local EVENT_VALUE_KEY_RELEASE = 0
-- For Kindle Oasis orientation events (ABS.code)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- the ABS code of orientation event will be adjusted to -24 from 24 (C.ABS_PRESSURE)
-- as C.ABS_PRESSURE is also used to detect touch input in KOBO devices.
local ABS_OASIS_ORIENTATION = -24
local DEVICE_ORIENTATION_PORTRAIT_LEFT = 15
local DEVICE_ORIENTATION_PORTRAIT_RIGHT = 17
local DEVICE_ORIENTATION_PORTRAIT = 19
local DEVICE_ORIENTATION_PORTRAIT_ROTATED_LEFT = 16
2016-08-08 08:42:01 +00:00
local DEVICE_ORIENTATION_PORTRAIT_ROTATED_RIGHT = 18
local DEVICE_ORIENTATION_PORTRAIT_ROTATED = 20
local DEVICE_ORIENTATION_LANDSCAPE = 21
local DEVICE_ORIENTATION_LANDSCAPE_ROTATED = 22
-- Kindle Oasis 2 & 3 variant
-- c.f., drivers/input/misc/accel/bma2x2.c
local UPWARD_PORTRAIT_UP_INTERRUPT_HAPPENED = 15
local UPWARD_PORTRAIT_DOWN_INTERRUPT_HAPPENED = 16
local UPWARD_LANDSCAPE_LEFT_INTERRUPT_HAPPENED = 17
local UPWARD_LANDSCAPE_RIGHT_INTERRUPT_HAPPENED = 18
-- For the events of the Forma & Libra accelerometers (MSC.value)
-- c.f., drivers/hwmon/mma8x5x.c
local MSC_RAW_GSENSOR_PORTRAIT_DOWN = 0x17
local MSC_RAW_GSENSOR_PORTRAIT_UP = 0x18
local MSC_RAW_GSENSOR_LANDSCAPE_RIGHT = 0x19
local MSC_RAW_GSENSOR_LANDSCAPE_LEFT = 0x1a
-- Not that we care about those, but they are reported, and accurate ;).
local MSC_RAW_GSENSOR_BACK = 0x1b
local MSC_RAW_GSENSOR_FRONT = 0x1c
-- Based on ABS_MT_TOOL_TYPE values on Elan panels
local TOOL_TYPE_FINGER = 0
local TOOL_TYPE_PEN = 1
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- For debug logging of ev.type
local linux_evdev_type_map = {
[C.EV_SYN] = "EV_SYN",
[C.EV_KEY] = "EV_KEY",
[C.EV_REL] = "EV_REL",
[C.EV_ABS] = "EV_ABS",
[C.EV_MSC] = "EV_MSC",
[C.EV_SW] = "EV_SW",
[C.EV_LED] = "EV_LED",
[C.EV_SND] = "EV_SND",
[C.EV_REP] = "EV_REP",
[C.EV_FF] = "EV_FF",
[C.EV_PWR] = "EV_PWR",
[C.EV_FF_STATUS] = "EV_FF_STATUS",
[C.EV_MAX] = "EV_MAX",
[C.EV_SDL] = "EV_SDL",
}
-- For debug logging of ev.code
local linux_evdev_syn_code_map = {
[C.SYN_REPORT] = "SYN_REPORT",
[C.SYN_CONFIG] = "SYN_CONFIG",
[C.SYN_MT_REPORT] = "SYN_MT_REPORT",
[C.SYN_DROPPED] = "SYN_DROPPED",
}
local linux_evdev_key_code_map = {
[C.KEY_BATTERY] = "KEY_BATTERY",
[C.BTN_TOOL_PEN] = "BTN_TOOL_PEN",
[C.BTN_TOOL_FINGER] = "BTN_TOOL_FINGER",
[C.BTN_TOOL_RUBBER] = "BTN_TOOL_RUBBER",
[C.BTN_TOUCH] = "BTN_TOUCH",
[C.BTN_STYLUS] = "BTN_STYLUS",
[C.BTN_STYLUS2] = "BTN_STYLUS2",
}
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
local linux_evdev_abs_code_map = {
[C.ABS_X] = "ABS_X",
[C.ABS_Y] = "ABS_Y",
[C.ABS_PRESSURE] = "ABS_PRESSURE",
[C.ABS_DISTANCE] = "ABS_DISTANCE",
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
[C.ABS_MT_SLOT] = "ABS_MT_SLOT",
[C.ABS_MT_TOUCH_MAJOR] = "ABS_MT_TOUCH_MAJOR",
[C.ABS_MT_TOUCH_MINOR] = "ABS_MT_TOUCH_MINOR",
[C.ABS_MT_WIDTH_MAJOR] = "ABS_MT_WIDTH_MAJOR",
[C.ABS_MT_WIDTH_MINOR] = "ABS_MT_WIDTH_MINOR",
[C.ABS_MT_ORIENTATION] = "ABS_MT_ORIENTATION",
[C.ABS_MT_POSITION_X] = "ABS_MT_POSITION_X",
[C.ABS_MT_POSITION_Y] = "ABS_MT_POSITION_Y",
[C.ABS_MT_TOOL_TYPE] = "ABS_MT_TOOL_TYPE",
[C.ABS_MT_BLOB_ID] = "ABS_MT_BLOB_ID",
[C.ABS_MT_TRACKING_ID] = "ABS_MT_TRACKING_ID",
[C.ABS_MT_PRESSURE] = "ABS_MT_PRESSURE",
[C.ABS_TILT_X] = "ABS_TILT_X",
[C.ABS_TILT_Y] = "ABS_TILT_Y",
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
[C.ABS_MT_DISTANCE] = "ABS_MT_DISTANCE",
[C.ABS_MT_TOOL_X] = "ABS_MT_TOOL_X",
[C.ABS_MT_TOOL_Y] = "ABS_MT_TOOL_Y",
}
local linux_evdev_msc_code_map = {
[C.MSC_RAW] = "MSC_RAW",
}
local linux_evdev_rep_code_map = {
[C.REP_DELAY] = "REP_DELAY",
[C.REP_PERIOD] = "REP_PERIOD",
}
2016-02-16 07:10:07 +00:00
-- luacheck: pop
local _internal_clipboard_text = nil -- holds the last copied text
local Input = {
-- must point to the device implementation when instantiating
device = nil,
-- this depends on keyboard layout and should be overridden:
event_map = {},
-- adapters are post processing functions that transform a given event to another event
event_map_adapter = {},
2021-07-21 20:20:08 +00:00
-- EV_ABS event to honor for pressure event (if any)
pressure_event = nil,
group = {
Cursor = { "Up", "Down", "Left", "Right" },
PgFwd = { "RPgFwd", "LPgFwd" },
PgBack = { "RPgBack", "LPgBack" },
2022-01-23 17:40:37 +00:00
Back = { "Back" },
Alphabet = {
"A", "B", "C", "D", "E", "F", "G", "H", "I", "J", "K", "L", "M",
"N", "O", "P", "Q", "R", "S", "T", "U", "V", "W", "X", "Y", "Z"
},
AlphaNumeric = {
"0", "1", "2", "3", "4", "5", "6", "7", "8", "9",
"A", "B", "C", "D", "E", "F", "G", "H", "I", "J", "K", "L", "M",
"N", "O", "P", "Q", "R", "S", "T", "U", "V", "W", "X", "Y", "Z"
},
Numeric = {
"0", "1", "2", "3", "4", "5", "6", "7", "8", "9"
},
Text = {
" ", ".", "/",
"0", "1", "2", "3", "4", "5", "6", "7", "8", "9",
"A", "B", "C", "D", "E", "F", "G", "H", "I", "J", "K", "L", "M",
"N", "O", "P", "Q", "R", "S", "T", "U", "V", "W", "X", "Y", "Z"
},
Any = {
" ", ".", "/",
"0", "1", "2", "3", "4", "5", "6", "7", "8", "9",
"A", "B", "C", "D", "E", "F", "G", "H", "I", "J", "K", "L", "M",
"N", "O", "P", "Q", "R", "S", "T", "U", "V", "W", "X", "Y", "Z",
"Up", "Down", "Left", "Right", "Press", "Backspace", "End",
"Back", "Sym", "AA", "Menu", "Home", "Del",
"LPgBack", "RPgBack", "LPgFwd", "RPgFwd"
},
},
-- NOTE: When looking at the device in Portrait mode, that's assuming PgBack is on TOP, and PgFwd on the BOTTOM
rotation_map = {
2016-08-08 08:42:01 +00:00
[framebuffer.ORIENTATION_PORTRAIT] = {},
[framebuffer.ORIENTATION_LANDSCAPE] = { Up = "Right", Right = "Down", Down = "Left", Left = "Up", LPgBack = "LPgFwd", LPgFwd = "LPgBack", RPgBack = "RPgFwd", RPgFwd = "RPgBack" },
2016-08-08 08:42:01 +00:00
[framebuffer.ORIENTATION_PORTRAIT_ROTATED] = { Up = "Down", Right = "Left", Down = "Up", Left = "Right", LPgFwd = "LPgBack", LPgBack = "LPgFwd", RPgFwd = "RPgBack", RPgBack = "RPgFwd" },
[framebuffer.ORIENTATION_LANDSCAPE_ROTATED] = { Up = "Left", Right = "Up", Down = "Right", Left = "Down" }
},
timer_callbacks = nil, -- instance-specific table, because the object may get destroyed & recreated at runtime
disable_double_tap = true,
tap_interval_override = nil,
-- keyboard state:
modifiers = {
Alt = false,
Ctrl = false,
Shift = false,
Sym = false,
},
-- repeat state:
repeat_count = 0,
-- touch state:
main_finger_slot = 0,
cur_slot = 0,
MTSlots = nil, -- table, object may be replaced at runtime
active_slots = nil, -- ditto
ev_slots = nil, -- table
gesture_detector = nil,
-- simple internal clipboard implementation, can be overidden to use system clipboard
hasClipboardText = function()
return _internal_clipboard_text ~= nil and _internal_clipboard_text ~= ""
end,
getClipboardText = function()
return _internal_clipboard_text
end,
setClipboardText = function(text)
_internal_clipboard_text = text
end,
}
function Input:new(o)
2016-02-16 07:10:07 +00:00
o = o or {}
setmetatable(o, self)
self.__index = self
if o.init then o:init() end
return o
end
function Input:init()
-- Initialize instance-specific tables
2022-08-16 22:25:33 +00:00
-- NOTE: All of these arrays may be destroyed & recreated at runtime, so we don't want a parent/class object for those.
self.timer_callbacks = {}
self.MTSlots = {}
self.active_slots = {}
-- Handle default finger slot
2021-07-21 20:20:08 +00:00
self.cur_slot = self.main_finger_slot
self.ev_slots = {
[self.main_finger_slot] = {
slot = self.main_finger_slot,
},
}
self.gesture_detector = GestureDetector:new{
screen = self.device.screen,
input = self,
}
-- set up fake event map
self.event_map[10000] = "IntoSS" -- go into screen saver
self.event_map[10001] = "OutOfSS" -- go out of screen saver
self.event_map[10010] = "UsbPlugIn"
self.event_map[10011] = "UsbPlugOut"
self.event_map[10020] = "Charging"
self.event_map[10021] = "NotCharging"
self.event_map[10030] = "WakeupFromSuspend"
self.event_map[10031] = "ReadyToSuspend"
-- user custom event map
local custom_event_map_location = string.format(
"%s/%s", DataStorage:getSettingsDir(), "event_map.lua")
local ok, custom_event_map = pcall(dofile, custom_event_map_location)
if ok then
for key, value in pairs(custom_event_map) do
self.event_map[key] = value
end
logger.info("loaded custom event map", custom_event_map)
end
2022-01-23 17:40:37 +00:00
if G_reader_settings:isTrue("backspace_as_back") then
table.insert(self.group.Back, "Backspace")
end
-- setup inhibitInputUntil scheduling function
self._inhibitInputUntil_func = function() self:inhibitInputUntil() end
end
--[[--
Wrapper for FFI input open.
2014-11-04 08:56:04 +00:00
Note that we adhere to the "." syntax here for compatibility.
@todo Clean up separation FFI/this.
2014-11-04 08:56:04 +00:00
--]]
2015-01-12 16:18:59 +00:00
function Input.open(device, is_emu_events)
return input.open(device, is_emu_events and 1 or 0)
2014-11-04 08:56:04 +00:00
end
--[[--
Different device models can implement their own hooks
and register them.
--]]
function Input:registerEventAdjustHook(hook, hook_params)
local old = self.eventAdjustHook
2016-04-03 04:52:30 +00:00
self.eventAdjustHook = function(this, ev)
old(this, ev)
hook(this, ev, hook_params)
end
end
2016-04-03 04:52:30 +00:00
function Input:registerGestureAdjustHook(hook, hook_params)
local old = self.gestureAdjustHook
self.gestureAdjustHook = function(this, ges)
old(this, ges)
hook(this, ges, hook_params)
end
end
function Input:eventAdjustHook(ev)
-- do nothing by default
end
2016-04-03 04:52:30 +00:00
function Input:gestureAdjustHook(ges)
-- do nothing by default
end
--- Catalog of predefined hooks.
function Input:adjustTouchSwitchXY(ev)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if ev.type == C.EV_ABS then
if ev.code == C.ABS_X then
ev.code = C.ABS_Y
elseif ev.code == C.ABS_Y then
ev.code = C.ABS_X
elseif ev.code == C.ABS_MT_POSITION_X then
ev.code = C.ABS_MT_POSITION_Y
elseif ev.code == C.ABS_MT_POSITION_Y then
ev.code = C.ABS_MT_POSITION_X
end
end
end
2016-04-03 04:52:30 +00:00
function Input:adjustTouchScale(ev, by)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if ev.type == C.EV_ABS then
if ev.code == C.ABS_X or ev.code == C.ABS_MT_POSITION_X then
ev.value = by.x * ev.value
end
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if ev.code == C.ABS_Y or ev.code == C.ABS_MT_POSITION_Y then
ev.value = by.y * ev.value
end
end
end
2016-04-03 04:52:30 +00:00
function Input:adjustTouchMirrorX(ev, max_x)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if ev.type == C.EV_ABS
and (ev.code == C.ABS_X or ev.code == C.ABS_MT_POSITION_X) then
ev.value = max_x - ev.value
end
end
2016-04-03 04:52:30 +00:00
function Input:adjustTouchMirrorY(ev, max_y)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if ev.type == C.EV_ABS
and (ev.code == C.ABS_Y or ev.code == C.ABS_MT_POSITION_Y) then
ev.value = max_y - ev.value
end
end
2016-04-03 04:52:30 +00:00
function Input:adjustTouchTranslate(ev, by)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if ev.type == C.EV_ABS then
if ev.code == C.ABS_X or ev.code == C.ABS_MT_POSITION_X then
ev.value = by.x + ev.value
end
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if ev.code == C.ABS_Y or ev.code == C.ABS_MT_POSITION_Y then
ev.value = by.y + ev.value
end
end
end
2016-04-03 04:52:30 +00:00
function Input:adjustKindleOasisOrientation(ev)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if ev.type == C.EV_ABS and ev.code == C.ABS_PRESSURE then
ev.code = ABS_OASIS_ORIENTATION
end
end
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
function Input:setTimeout(slot, ges, cb, origin, delay)
local item = {
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
slot = slot,
gesture = ges,
callback = cb,
}
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- We're going to need the clock source id for these events from GestureDetector
local clock_id = self.gesture_detector:getClockSource()
local deadline
-- If we're on a platform with the timerfd backend, handle that
local timerfd
if input.setTimer then
-- If GestureDetector's clock source probing was inconclusive, do this on the UI timescale instead.
if clock_id == -1 then
deadline = time.now() + delay
clock_id = C.CLOCK_MONOTONIC
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
else
deadline = origin + delay
end
-- What this does is essentially to ask the kernel to wake us up when the timer expires,
-- instead of ensuring that ourselves via a polling timeout.
-- This ensures perfect accuracy, and allows it to be computed in the event's own timescale.
local sec, usec = time.split_s_us(deadline)
timerfd = input.setTimer(clock_id, sec, usec)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
end
if timerfd then
-- It worked, tweak the table a bit to make it clear the deadline will be handled by the kernel
item.timerfd = timerfd
-- We basically only need this for the sorting ;).
item.deadline = deadline
else
-- No timerfd, we'll compute a poll timeout ourselves.
if clock_id == C.CLOCK_MONOTONIC then
-- If the event's clocksource is monotonic, we can use it directly.
deadline = origin + delay
else
-- Otherwise, fudge it by using a current timestamp in the UI's timescale (MONOTONIC).
-- This isn't the end of the world in practice (c.f., #7415).
deadline = time.now() + delay
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
end
item.deadline = deadline
end
table.insert(self.timer_callbacks, item)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- NOTE: While the timescale is monotonic, we may interleave timers based on different delays, so we still need to sort...
table.sort(self.timer_callbacks, function(v1, v2)
return v1.deadline < v2.deadline
end)
end
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- Clear all timeouts for a specific slot (and a specific gesture, if ges is set)
function Input:clearTimeout(slot, ges)
for i = #self.timer_callbacks, 1, -1 do
local item = self.timer_callbacks[i]
if item.slot == slot and (not ges or item.gesture == ges) then
-- If the timerfd backend is in use, close the fd and free the list's node, too.
if item.timerfd then
input.clearTimer(item.timerfd)
end
table.remove(self.timer_callbacks, i)
end
end
end
function Input:clearTimeouts()
-- If the timerfd backend is in use, close the fds, too
if input.setTimer then
for _, item in ipairs(self.timer_callbacks) do
if item.timerfd then
input.clearTimer(item.timerfd)
end
end
end
self.timer_callbacks = {}
end
-- Reset the gesture parsing state to a blank slate
function Input:resetState()
if self.gesture_detector then
self.gesture_detector:dropContacts()
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- Resets the clock source probe
self.gesture_detector:resetClockSource()
end
self:clearTimeouts()
end
function Input:handleKeyBoardEv(ev)
-- Detect loss of contact for the "snow" protocol...
-- NOTE: Some ST devices may also behave similarly, but we handle those via ABS_PRESSURE
if self.snow_protocol then
if ev.code == C.BTN_TOUCH and ev.value == 0 then
-- Kernel sends it after loss of contact for *all* slots,
-- only once the final contact point has been lifted.
if #self.MTSlots == 0 then
-- Likely, since this is usually in its own event stream,
-- meaning self.MTSlots has *just* been cleared by our last EV_SYN:SYN_REPORT handler...
-- So, poke at the actual data to find the slots that are currently active (i.e., in the down state),
-- and re-populate a minimal self.MTSlots array that simply switches them to the up state ;).
for _, slot in pairs(self.ev_slots) do
if slot.id ~= -1 then
table.insert(self.MTSlots, slot)
slot.id = -1
end
end
else
-- Unlikely, given what we mentioned above...
-- Note that, funnily enough, its EV_KEY:BTN_TOUCH:1 counterpart
-- *can* be in the same initial event stream as the EV_ABS batch...
for _, MTSlot in ipairs(self.MTSlots) do
self:setMtSlot(MTSlot.slot, "id", -1)
end
end
return
end
elseif self.wacom_protocol then
if ev.code == C.BTN_TOOL_PEN then
-- Always send pen data to slot 0
self:setupSlotData(0)
if ev.value == 1 then
self:setCurrentMtSlot("tool", TOOL_TYPE_PEN)
else
self:setCurrentMtSlot("tool", TOOL_TYPE_FINGER)
end
elseif ev.code == C.BTN_TOUCH then
-- Much like on snow, use this to detect contact down & lift,
-- as ABS_PRESSURE may be entirely omitted from hover events,
-- and ABS_DISTANCE is not very clear cut...
self:setupSlotData(0)
if ev.value == 1 then
self:setCurrentMtSlot("id", 0)
else
self:setCurrentMtSlot("id", -1)
end
end
end
local keycode = self.event_map[ev.code]
if not keycode then
-- do not handle keypress for keys we don't know
return
end
if self.event_map_adapter[keycode] then
return self.event_map_adapter[keycode](ev)
end
-- take device rotation into account
if self.rotation_map[self.device.screen:getRotationMode()][keycode] then
keycode = self.rotation_map[self.device.screen:getRotationMode()][keycode]
end
-- fake events
if keycode == "IntoSS" or keycode == "OutOfSS"
or keycode == "UsbPlugIn" or keycode == "UsbPlugOut"
or keycode == "Charging" or keycode == "NotCharging"
or keycode == "WakeupFromSuspend" or keycode == "ReadyToSuspend" then
return keycode
end
-- The hardware camera button is used in Android to toggle the touchscreen
if keycode == "Camera" and ev.value == EVENT_VALUE_KEY_RELEASE
and G_reader_settings:isTrue("camera_key_toggles_touchscreen") then
local isAndroid, android = pcall(require, "android")
if isAndroid then
-- toggle touchscreen behaviour
android.toggleTouchscreenIgnored()
-- show a toast with the new behaviour
if android.isTouchscreenIgnored() then
android.notification(_("Touchscreen disabled"))
else
android.notification(_("Touchscreen enabled"))
end
end
return
end
if keycode == "Power" then
-- Kobo generates Power keycode only, we need to decide whether it's
-- power-on or power-off ourselves.
if ev.value == EVENT_VALUE_KEY_PRESS then
return "PowerPress"
elseif ev.value == EVENT_VALUE_KEY_RELEASE then
return "PowerRelease"
end
end
-- toggle fullscreen on F11
if self:isEvKeyPress(ev) and keycode == "F11" and not self.device:isAlwaysFullscreen() then
local UIManager = require("ui/uimanager")
UIManager:broadcastEvent(Event:new("ToggleFullscreen"))
end
-- quit on Alt + F4
-- this is also emitted by the close event in SDL
if self:isEvKeyPress(ev) and self.modifiers["Alt"] and keycode == "F4" then
local UIManager = require("ui/uimanager")
UIManager:broadcastEvent(Event:new("Close")) -- Tell all widgets to close.
UIManager:nextTick(function() UIManager:quit() end) -- Ensure the program closes in case of some lingering dialog.
end
-- handle modifier keys
if self.modifiers[keycode] ~= nil then
if ev.value == EVENT_VALUE_KEY_PRESS then
self.modifiers[keycode] = true
elseif ev.value == EVENT_VALUE_KEY_RELEASE then
self.modifiers[keycode] = false
end
return
end
local key = Key:new(keycode, self.modifiers)
if ev.value == EVENT_VALUE_KEY_PRESS then
return Event:new("KeyPress", key)
elseif ev.value == EVENT_VALUE_KEY_REPEAT then
-- NOTE: We only care about repeat events from the pageturn buttons...
-- And we *definitely* don't want to flood the Event queue with useless SleepCover repeats!
if keycode == "LPgBack"
or keycode == "RPgBack"
or keycode == "LPgFwd"
or keycode == "RPgFwd" then
--- @fixme Crappy event staggering!
--
-- The Forma & co repeats every 80ms after a 400ms delay, and 500ms roughly corresponds to a flashing update,
-- so stuff is usually in sync when you release the key.
-- Obvious downside is that this ends up slower than just mashing the key.
--
-- A better approach would be an onKeyRelease handler that flushes the Event queue...
self.repeat_count = self.repeat_count + 1
if self.repeat_count == 1 then
return Event:new("KeyRepeat", key)
elseif self.repeat_count >= 6 then
self.repeat_count = 0
end
end
elseif ev.value == EVENT_VALUE_KEY_RELEASE then
self.repeat_count = 0
return Event:new("KeyRelease", key)
end
end
A host of low power states related tweaks (#9036) * Disable all non power management related input during suspend. (This prevents wonky touch events from being tripped when closing a sleep cover on an already-in-suspend device, among other things). * Kobo: Use our WakeupMgr instance, not the class. * WakupMgr: split `removeTask` in two: * `removeTask`, which *only* takes a queue index as input, and only removes a single task. Greatly simplifies the function (i.e., it's just a `table.remove`). * `removeTasks`, which takes an epoch or a cb ref, and removes *every* task that matches. * Both of these will also *always* re-schedule the next task (if any) on exit, since we can have multiple WakeupMgr tasks queued, but we can only have a single RTC wake alarm set ;). * `wakeupAction` now takes a `proximity` argument, which it passes on to its `validateWakeupAlarmByProximity` call, allowing call sites to avoir having to duplicate that call themselves when they want to use a custom proximity window. * `wakeupAction` now re-schedules the next task (if any) on exit. * Simplify `Kobo:checkUnexpectedWakeup`, by removing the duplicate `WakerupMgr:validateWakeupAlarmByProximity` call, now that we can pass a proximity window to `WakeuoMgr:wakeupAction`. * The various network activity timeouts are now halved when autostandby is enabled. * Autostandby: get rid of the dummy deadline_guard task, as it's no longer necessary since #9009. * UIManager: The previous change allows us to simplify `getNextTaskTimes` into a simpler `getNextTaskTime` variant, getting rid of a table & a loop. * ReaderFooter & ReaderHeader: Make sure we only perform a single refresh when exiting standby. * Kobo: Rewrite sysfs writes to use ANSI C via FFI instead of stdio via Lua, as it obscured some common error cases (e.g., EBUSY on /sys/power/state). * Kobo: Simplify `suspend`, now that we have sane error handling in sysfs writes. * Kobo.powerd: Change `isCharging` & `isAuxCharging` behavior to match the behavior of the NTX ioctl (i.e., Charging == Plugged-in). This has the added benefit of making the AutoSuspend checks behave sensibly in the "fully-charged but still plugged in" scenario (because being plugged in is enough to break PM on `!canPowerSaveWhileCharging` devices). * AutoSuspend: Disable our `AllowStandby` handler when auto standby is disabled, so as to not interfere with other modules using `UIManager:allowStandby` (fix #9038). * PowerD: Allow platforms to implement `isCharged`, indicating that the battery is full while still plugged in to a power source (battery icon becomes a power plug icon). * Kobo.powerd: Implement `isCharged`, and kill charging LEDs once battery is full. * Kindle.powerd: Implement `isCharged` on post-Wario devices. (`isCharging` is still true in that state, as it ought to).
2022-05-01 21:41:08 +00:00
-- Mangled variant of handleKeyBoardEv that will only handle power management related keys.
-- (Used when blocking input during suspend via sleep cover).
function Input:handlePowerManagementOnlyEv(ev)
local keycode = self.event_map[ev.code]
if not keycode then
-- Do not handle keypress for keys we don't know
return
end
-- We'll need to parse the synthetic event map, because SleepCover* events are synthetic.
if self.event_map_adapter[keycode] then
keycode = self.event_map_adapter[keycode](ev)
end
-- Power management synthetic events
if keycode == "SleepCoverClosed" or keycode == "SleepCoverOpened"
or keycode == "Suspend" or keycode == "Resume" then
return keycode
end
-- Fake events
if keycode == "IntoSS" or keycode == "OutOfSS"
or keycode == "UsbPlugIn" or keycode == "UsbPlugOut"
or keycode == "Charging" or keycode == "NotCharging"
or keycode == "WakeupFromSuspend" or keycode == "ReadyToSuspend" then
A host of low power states related tweaks (#9036) * Disable all non power management related input during suspend. (This prevents wonky touch events from being tripped when closing a sleep cover on an already-in-suspend device, among other things). * Kobo: Use our WakeupMgr instance, not the class. * WakupMgr: split `removeTask` in two: * `removeTask`, which *only* takes a queue index as input, and only removes a single task. Greatly simplifies the function (i.e., it's just a `table.remove`). * `removeTasks`, which takes an epoch or a cb ref, and removes *every* task that matches. * Both of these will also *always* re-schedule the next task (if any) on exit, since we can have multiple WakeupMgr tasks queued, but we can only have a single RTC wake alarm set ;). * `wakeupAction` now takes a `proximity` argument, which it passes on to its `validateWakeupAlarmByProximity` call, allowing call sites to avoir having to duplicate that call themselves when they want to use a custom proximity window. * `wakeupAction` now re-schedules the next task (if any) on exit. * Simplify `Kobo:checkUnexpectedWakeup`, by removing the duplicate `WakerupMgr:validateWakeupAlarmByProximity` call, now that we can pass a proximity window to `WakeuoMgr:wakeupAction`. * The various network activity timeouts are now halved when autostandby is enabled. * Autostandby: get rid of the dummy deadline_guard task, as it's no longer necessary since #9009. * UIManager: The previous change allows us to simplify `getNextTaskTimes` into a simpler `getNextTaskTime` variant, getting rid of a table & a loop. * ReaderFooter & ReaderHeader: Make sure we only perform a single refresh when exiting standby. * Kobo: Rewrite sysfs writes to use ANSI C via FFI instead of stdio via Lua, as it obscured some common error cases (e.g., EBUSY on /sys/power/state). * Kobo: Simplify `suspend`, now that we have sane error handling in sysfs writes. * Kobo.powerd: Change `isCharging` & `isAuxCharging` behavior to match the behavior of the NTX ioctl (i.e., Charging == Plugged-in). This has the added benefit of making the AutoSuspend checks behave sensibly in the "fully-charged but still plugged in" scenario (because being plugged in is enough to break PM on `!canPowerSaveWhileCharging` devices). * AutoSuspend: Disable our `AllowStandby` handler when auto standby is disabled, so as to not interfere with other modules using `UIManager:allowStandby` (fix #9038). * PowerD: Allow platforms to implement `isCharged`, indicating that the battery is full while still plugged in to a power source (battery icon becomes a power plug icon). * Kobo.powerd: Implement `isCharged`, and kill charging LEDs once battery is full. * Kindle.powerd: Implement `isCharged` on post-Wario devices. (`isCharging` is still true in that state, as it ought to).
2022-05-01 21:41:08 +00:00
return keycode
end
if keycode == "Power" then
-- Kobo generates Power keycode only, we need to decide whether it's
-- power-on or power-off ourselves.
if ev.value == EVENT_VALUE_KEY_PRESS then
return "PowerPress"
elseif ev.value == EVENT_VALUE_KEY_RELEASE then
return "PowerRelease"
end
end
-- Nothing to see, move along!
return
end
-- Empty event handler used to send input to the void
function Input:voidEv(ev)
return
end
-- Generic event handler for unhandled input events
function Input:handleGenericEv(ev)
return Event:new("GenericInput", ev)
end
function Input:handleMiscEv(ev)
-- should be handled by a misc event protocol plugin
end
function Input:handleSdlEv(ev)
-- overwritten by device implementation
end
--[[--
Parse each touch ev from kernel and build up tev.
tev will be sent to GestureDetector:feedEvent
Events for a single tap motion from Linux kernel (MT protocol B):
MT_TRACK_ID: 0
MT_X: 222
MT_Y: 207
SYN REPORT
MT_TRACK_ID: -1
SYN REPORT
Notice that each line is a single event.
From kernel document:
For type B devices, the kernel driver should associate a slot with each
identified contact, and use that slot to propagate changes for the contact.
Creation, replacement and destruction of contacts is achieved by modifying
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
the C.ABS_MT_TRACKING_ID of the associated slot. A non-negative tracking id
is interpreted as a contact, and the value -1 denotes an unused slot. A
tracking id not previously present is considered new, and a tracking id no
longer present is considered removed. Since only changes are propagated,
the full state of each initiated contact has to reside in the receiving
end. Upon receiving an MT event, one simply updates the appropriate
attribute of the current slot.
--]]
function Input:handleTouchEv(ev)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if ev.type == C.EV_ABS then
-- NOTE: Ideally, an input frame starts with either ABS_MT_SLOT or ABS_MT_TRACKING_ID,
-- but they *both* may be omitted if the last contact point just moved without lift.
-- The use of setCurrentMtSlotChecked instead of setCurrentMtSlot ensures
-- we actually setup the slot data storage and/or reference for the current slot in this case,
-- as the reference list is empty at the beginning of an input frame (c.f., Input:newFrame).
-- The most common platforms where you'll see this happen are:
-- * PocketBook, because of our InkView EVT_POINTERMOVE translation
-- (c.f., translateEvent @ ffi/input_pocketbook.lua).
-- * SDL, because of our SDL_MOUSEMOTION/SDL_FINGERMOTION translation
-- (c.f., waitForEvent @ ffi/SDL2_0.lua).
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if ev.code == C.ABS_MT_SLOT then
self:setupSlotData(ev.value)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
elseif ev.code == C.ABS_MT_TRACKING_ID then
if self.snow_protocol then
-- NOTE: We'll never get an ABS_MT_SLOT event, instead we have a slot-like ABS_MT_TRACKING_ID value...
-- This also means this may never be set to -1 on contact lift,
-- which is why we instead rely on EV_KEY:BTN_TOUCH:0 for that (c.f., handleKeyBoardEv).
self:setupSlotData(ev.value)
else
-- The Elan driver needlessly repeats unchanged ABS_MT_TRACKING_ID values,
-- which allows us to do this here instead of relying more aggressively on setCurrentMtSlotChecked.
if #self.MTSlots == 0 then
self:addSlot(self.cur_slot)
end
end
self:setCurrentMtSlot("id", ev.value)
elseif ev.code == C.ABS_MT_TOOL_TYPE then
-- NOTE: On the Elipsa: Finger == 0; Pen == 1
self:setCurrentMtSlot("tool", ev.value)
elseif ev.code == C.ABS_MT_POSITION_X or ev.code == C.ABS_X then
self:setCurrentMtSlotChecked("x", ev.value)
elseif ev.code == C.ABS_MT_POSITION_Y or ev.code == C.ABS_Y then
self:setCurrentMtSlotChecked("y", ev.value)
2021-07-21 20:20:08 +00:00
elseif self.pressure_event and ev.code == self.pressure_event and ev.value == 0 then
-- Drop hovering *pen* events
local tool = self:getCurrentMtSlotData("tool")
if tool and tool == TOOL_TYPE_PEN then
self:setCurrentMtSlot("id", -1)
end
end
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
elseif ev.type == C.EV_SYN then
if ev.code == C.SYN_REPORT then
for _, MTSlot in ipairs(self.MTSlots) do
self:setMtSlot(MTSlot.slot, "timev", time.timeval(ev.time))
end
-- feed ev in all slots to state machine
local touch_gestures = self.gesture_detector:feedEvent(self.MTSlots)
self:newFrame()
local ges_evs = {}
for _, touch_ges in ipairs(touch_gestures) do
self:gestureAdjustHook(touch_ges)
table.insert(ges_evs, Event:new("Gesture", self.gesture_detector:adjustGesCoordinate(touch_ges)))
end
return ges_evs
end
end
end
function Input:handleTouchEvPhoenix(ev)
-- Hack on handleTouchEV for the Kobo Aura
-- It seems to be using a custom protocol:
-- finger 0 down:
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- input_report_abs(elan_touch_data.input, C.ABS_MT_TRACKING_ID, 0);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_TOUCH_MAJOR, 1);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_WIDTH_MAJOR, 1);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_POSITION_X, x1);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_POSITION_Y, y1);
-- input_mt_sync (elan_touch_data.input);
-- finger 1 down:
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- input_report_abs(elan_touch_data.input, C.ABS_MT_TRACKING_ID, 1);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_TOUCH_MAJOR, 1);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_WIDTH_MAJOR, 1);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_POSITION_X, x2);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_POSITION_Y, y2);
-- input_mt_sync (elan_touch_data.input);
-- finger 0 up:
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- input_report_abs(elan_touch_data.input, C.ABS_MT_TRACKING_ID, 0);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_TOUCH_MAJOR, 0);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_WIDTH_MAJOR, 0);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_POSITION_X, last_x);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_POSITION_Y, last_y);
-- input_mt_sync (elan_touch_data.input);
-- finger 1 up:
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- input_report_abs(elan_touch_data.input, C.ABS_MT_TRACKING_ID, 1);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_TOUCH_MAJOR, 0);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_WIDTH_MAJOR, 0);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_POSITION_X, last_x2);
-- input_report_abs(elan_touch_data.input, C.ABS_MT_POSITION_Y, last_y2);
-- input_mt_sync (elan_touch_data.input);
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if ev.type == C.EV_ABS then
if ev.code == C.ABS_MT_TRACKING_ID then
self:setupSlotData(ev.value)
self:setCurrentMtSlot("id", ev.value)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
elseif ev.code == C.ABS_MT_TOUCH_MAJOR and ev.value == 0 then
self:setCurrentMtSlot("id", -1)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
elseif ev.code == C.ABS_MT_POSITION_X then
self:setCurrentMtSlot("x", ev.value)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
elseif ev.code == C.ABS_MT_POSITION_Y then
self:setCurrentMtSlot("y", ev.value)
end
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
elseif ev.type == C.EV_SYN then
if ev.code == C.SYN_REPORT then
for _, MTSlot in ipairs(self.MTSlots) do
self:setMtSlot(MTSlot.slot, "timev", time.timeval(ev.time))
end
-- feed ev in all slots to state machine
local touch_gestures = self.gesture_detector:feedEvent(self.MTSlots)
self:newFrame()
local ges_evs = {}
for _, touch_ges in ipairs(touch_gestures) do
self:gestureAdjustHook(touch_ges)
table.insert(ges_evs, Event:new("Gesture", self.gesture_detector:adjustGesCoordinate(touch_ges)))
end
return ges_evs
end
end
end
function Input:handleTouchEvLegacy(ev)
-- Single Touch Protocol.
-- Some devices emit both singletouch and multitouch events,
-- on those devices, the 'handleTouchEv' function may not behave as expected. Use this one instead.
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if ev.type == C.EV_ABS then
if ev.code == C.ABS_X then
self:setCurrentMtSlotChecked("x", ev.value)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
elseif ev.code == C.ABS_Y then
self:setCurrentMtSlotChecked("y", ev.value)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
elseif ev.code == C.ABS_PRESSURE then
if ev.value ~= 0 then
self:setCurrentMtSlotChecked("id", 1)
else
self:setCurrentMtSlotChecked("id", -1)
end
end
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
elseif ev.type == C.EV_SYN then
if ev.code == C.SYN_REPORT then
for _, MTSlot in ipairs(self.MTSlots) do
self:setMtSlot(MTSlot.slot, "timev", time.timeval(ev.time))
end
-- On Kobo Mk. 3 devices, the frame that reports a contact lift *actually* does the coordinates transform for us...
-- Unfortunately, our own transforms are not stateful, so, just revert 'em here,
-- since we can't simply avoid not doing 'em for that frame...
-- c.f., https://github.com/koreader/koreader/issues/2128#issuecomment-1236289909 for logs on a Touch B
if self.touch_kobo_mk3_protocol then
if self:getCurrentMtSlotData("id") == -1 then
-- Technically, it's the frame where ABS_PRESSURE is set to 0 ;).
local y = 599 - self:getCurrentMtSlotData("x") -- Mk. 3 devices are all 600x800, so just hard-code it here.
local x = self:getCurrentMtSlotData("y")
self:setCurrentMtSlot("x", x)
self:setCurrentMtSlot("y", y)
end
end
-- feed ev in all slots to state machine
local touch_gestures = self.gesture_detector:feedEvent(self.MTSlots)
self:newFrame()
local ges_evs = {}
for _, touch_ges in ipairs(touch_gestures) do
self:gestureAdjustHook(touch_ges)
table.insert(ges_evs, Event:new("Gesture", self.gesture_detector:adjustGesCoordinate(touch_ges)))
end
return ges_evs
end
end
end
function Input:handleOasisOrientationEv(ev)
local rotation_mode, screen_mode
if self.device:isZelda() then
if ev.value == UPWARD_PORTRAIT_UP_INTERRUPT_HAPPENED then
-- i.e., UR
rotation_mode = framebuffer.ORIENTATION_PORTRAIT
screen_mode = 'portrait'
elseif ev.value == UPWARD_LANDSCAPE_LEFT_INTERRUPT_HAPPENED then
-- i.e., CW
rotation_mode = framebuffer.ORIENTATION_LANDSCAPE
screen_mode = 'landscape'
elseif ev.value == UPWARD_PORTRAIT_DOWN_INTERRUPT_HAPPENED then
-- i.e., UD
rotation_mode = framebuffer.ORIENTATION_PORTRAIT_ROTATED
screen_mode = 'portrait'
elseif ev.value == UPWARD_LANDSCAPE_RIGHT_INTERRUPT_HAPPENED then
-- i.e., CCW
rotation_mode = framebuffer.ORIENTATION_LANDSCAPE_ROTATED
screen_mode = 'landscape'
end
else
if ev.value == DEVICE_ORIENTATION_PORTRAIT
or ev.value == DEVICE_ORIENTATION_PORTRAIT_LEFT
or ev.value == DEVICE_ORIENTATION_PORTRAIT_RIGHT then
-- i.e., UR
rotation_mode = framebuffer.ORIENTATION_PORTRAIT
screen_mode = 'portrait'
elseif ev.value == DEVICE_ORIENTATION_LANDSCAPE then
-- i.e., CW
rotation_mode = framebuffer.ORIENTATION_LANDSCAPE
screen_mode = 'landscape'
elseif ev.value == DEVICE_ORIENTATION_PORTRAIT_ROTATED
or ev.value == DEVICE_ORIENTATION_PORTRAIT_ROTATED_LEFT
or ev.value == DEVICE_ORIENTATION_PORTRAIT_ROTATED_RIGHT then
-- i.e., UD
rotation_mode = framebuffer.ORIENTATION_PORTRAIT_ROTATED
screen_mode = 'portrait'
elseif ev.value == DEVICE_ORIENTATION_LANDSCAPE_ROTATED then
-- i.e., CCW
rotation_mode = framebuffer.ORIENTATION_LANDSCAPE_ROTATED
screen_mode = 'landscape'
end
end
2016-08-02 09:21:27 +00:00
local old_rotation_mode = self.device.screen:getRotationMode()
if self.device:isGSensorLocked() then
local old_screen_mode = self.device.screen:getScreenMode()
if rotation_mode ~= old_rotation_mode and screen_mode == old_screen_mode then
-- Cheaper than a full SetRotationMode event, as we don't need to re-layout anything.
self.device.screen:setRotationMode(rotation_mode)
local UIManager = require("ui/uimanager")
UIManager:onRotation()
end
else
if rotation_mode ~= old_rotation_mode then
return Event:new("SetRotationMode", rotation_mode)
end
2016-08-02 09:21:27 +00:00
end
end
--- Accelerometer on the Forma/Libra
function Input:handleMiscEvNTX(ev)
local rotation_mode, screen_mode
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if ev.code == C.MSC_RAW then
if ev.value == MSC_RAW_GSENSOR_PORTRAIT_UP then
-- i.e., UR
rotation_mode = framebuffer.ORIENTATION_PORTRAIT
screen_mode = 'portrait'
elseif ev.value == MSC_RAW_GSENSOR_LANDSCAPE_RIGHT then
-- i.e., CW
rotation_mode = framebuffer.ORIENTATION_LANDSCAPE
screen_mode = 'landscape'
elseif ev.value == MSC_RAW_GSENSOR_PORTRAIT_DOWN then
-- i.e., UD
rotation_mode = framebuffer.ORIENTATION_PORTRAIT_ROTATED
screen_mode = 'portrait'
elseif ev.value == MSC_RAW_GSENSOR_LANDSCAPE_LEFT then
-- i.e., CCW
rotation_mode = framebuffer.ORIENTATION_LANDSCAPE_ROTATED
screen_mode = 'landscape'
else
-- Discard FRONT/BACK
return
end
else
-- Discard unhandled event codes, just to future-proof this ;).
return
end
local old_rotation_mode = self.device.screen:getRotationMode()
if self.device:isGSensorLocked() then
local old_screen_mode = self.device.screen:getScreenMode()
if rotation_mode and rotation_mode ~= old_rotation_mode and screen_mode == old_screen_mode then
-- Cheaper than a full SetRotationMode event, as we don't need to re-layout anything.
self.device.screen:setRotationMode(rotation_mode)
local UIManager = require("ui/uimanager")
UIManager:onRotation()
end
else
if rotation_mode and rotation_mode ~= old_rotation_mode then
return Event:new("SetRotationMode", rotation_mode)
end
end
end
--- Allow toggling the accelerometer at runtime.
function Input:toggleMiscEvNTX(toggle)
if toggle == true then
-- Honor Gyro events
if not self.isNTXAccelHooked then
self.handleMiscEv = self.handleMiscEvNTX
self.isNTXAccelHooked = true
end
elseif toggle == false then
-- Ignore Gyro events
if self.isNTXAccelHooked then
A host of low power states related tweaks (#9036) * Disable all non power management related input during suspend. (This prevents wonky touch events from being tripped when closing a sleep cover on an already-in-suspend device, among other things). * Kobo: Use our WakeupMgr instance, not the class. * WakupMgr: split `removeTask` in two: * `removeTask`, which *only* takes a queue index as input, and only removes a single task. Greatly simplifies the function (i.e., it's just a `table.remove`). * `removeTasks`, which takes an epoch or a cb ref, and removes *every* task that matches. * Both of these will also *always* re-schedule the next task (if any) on exit, since we can have multiple WakeupMgr tasks queued, but we can only have a single RTC wake alarm set ;). * `wakeupAction` now takes a `proximity` argument, which it passes on to its `validateWakeupAlarmByProximity` call, allowing call sites to avoir having to duplicate that call themselves when they want to use a custom proximity window. * `wakeupAction` now re-schedules the next task (if any) on exit. * Simplify `Kobo:checkUnexpectedWakeup`, by removing the duplicate `WakerupMgr:validateWakeupAlarmByProximity` call, now that we can pass a proximity window to `WakeuoMgr:wakeupAction`. * The various network activity timeouts are now halved when autostandby is enabled. * Autostandby: get rid of the dummy deadline_guard task, as it's no longer necessary since #9009. * UIManager: The previous change allows us to simplify `getNextTaskTimes` into a simpler `getNextTaskTime` variant, getting rid of a table & a loop. * ReaderFooter & ReaderHeader: Make sure we only perform a single refresh when exiting standby. * Kobo: Rewrite sysfs writes to use ANSI C via FFI instead of stdio via Lua, as it obscured some common error cases (e.g., EBUSY on /sys/power/state). * Kobo: Simplify `suspend`, now that we have sane error handling in sysfs writes. * Kobo.powerd: Change `isCharging` & `isAuxCharging` behavior to match the behavior of the NTX ioctl (i.e., Charging == Plugged-in). This has the added benefit of making the AutoSuspend checks behave sensibly in the "fully-charged but still plugged in" scenario (because being plugged in is enough to break PM on `!canPowerSaveWhileCharging` devices). * AutoSuspend: Disable our `AllowStandby` handler when auto standby is disabled, so as to not interfere with other modules using `UIManager:allowStandby` (fix #9038). * PowerD: Allow platforms to implement `isCharged`, indicating that the battery is full while still plugged in to a power source (battery icon becomes a power plug icon). * Kobo.powerd: Implement `isCharged`, and kill charging LEDs once battery is full. * Kindle.powerd: Implement `isCharged` on post-Wario devices. (`isCharging` is still true in that state, as it ought to).
2022-05-01 21:41:08 +00:00
self.handleMiscEv = self.voidEv
self.isNTXAccelHooked = false
end
else
-- Toggle it
if self.isNTXAccelHooked then
A host of low power states related tweaks (#9036) * Disable all non power management related input during suspend. (This prevents wonky touch events from being tripped when closing a sleep cover on an already-in-suspend device, among other things). * Kobo: Use our WakeupMgr instance, not the class. * WakupMgr: split `removeTask` in two: * `removeTask`, which *only* takes a queue index as input, and only removes a single task. Greatly simplifies the function (i.e., it's just a `table.remove`). * `removeTasks`, which takes an epoch or a cb ref, and removes *every* task that matches. * Both of these will also *always* re-schedule the next task (if any) on exit, since we can have multiple WakeupMgr tasks queued, but we can only have a single RTC wake alarm set ;). * `wakeupAction` now takes a `proximity` argument, which it passes on to its `validateWakeupAlarmByProximity` call, allowing call sites to avoir having to duplicate that call themselves when they want to use a custom proximity window. * `wakeupAction` now re-schedules the next task (if any) on exit. * Simplify `Kobo:checkUnexpectedWakeup`, by removing the duplicate `WakerupMgr:validateWakeupAlarmByProximity` call, now that we can pass a proximity window to `WakeuoMgr:wakeupAction`. * The various network activity timeouts are now halved when autostandby is enabled. * Autostandby: get rid of the dummy deadline_guard task, as it's no longer necessary since #9009. * UIManager: The previous change allows us to simplify `getNextTaskTimes` into a simpler `getNextTaskTime` variant, getting rid of a table & a loop. * ReaderFooter & ReaderHeader: Make sure we only perform a single refresh when exiting standby. * Kobo: Rewrite sysfs writes to use ANSI C via FFI instead of stdio via Lua, as it obscured some common error cases (e.g., EBUSY on /sys/power/state). * Kobo: Simplify `suspend`, now that we have sane error handling in sysfs writes. * Kobo.powerd: Change `isCharging` & `isAuxCharging` behavior to match the behavior of the NTX ioctl (i.e., Charging == Plugged-in). This has the added benefit of making the AutoSuspend checks behave sensibly in the "fully-charged but still plugged in" scenario (because being plugged in is enough to break PM on `!canPowerSaveWhileCharging` devices). * AutoSuspend: Disable our `AllowStandby` handler when auto standby is disabled, so as to not interfere with other modules using `UIManager:allowStandby` (fix #9038). * PowerD: Allow platforms to implement `isCharged`, indicating that the battery is full while still plugged in to a power source (battery icon becomes a power plug icon). * Kobo.powerd: Implement `isCharged`, and kill charging LEDs once battery is full. * Kindle.powerd: Implement `isCharged` on post-Wario devices. (`isCharging` is still true in that state, as it ought to).
2022-05-01 21:41:08 +00:00
self.handleMiscEv = self.voidEv
else
self.handleMiscEv = self.handleMiscEvNTX
end
self.isNTXAccelHooked = not self.isNTXAccelHooked
end
end
-- helpers for touch event data management:
function Input:initMtSlot(slot)
if not self.ev_slots[slot] then
self.ev_slots[slot] = {
slot = slot
}
end
end
function Input:setMtSlot(slot, key, val)
self.ev_slots[slot][key] = val
end
function Input:setCurrentMtSlot(key, val)
self:setMtSlot(self.cur_slot, key, val)
end
-- Same as above, but ensures the current slot actually has a live ref first
function Input:setCurrentMtSlotChecked(key, val)
if #self.MTSlots == 0 then
self:addSlot(self.cur_slot)
end
self:setMtSlot(self.cur_slot, key, val)
end
function Input:getMtSlot(slot)
return self.ev_slots[slot]
end
function Input:getCurrentMtSlot()
return self:getMtSlot(self.cur_slot)
end
function Input:getCurrentMtSlotData(key)
local slot = self:getCurrentMtSlot()
if slot then
return slot[key]
end
return nil
end
function Input:newFrame()
-- Array of references to the data for each slot seen in this input frame
-- (Points to self.ev_slots, c.f., getMtSlot)
self.MTSlots = {}
-- Simple hash to keep track of which references we've inserted into self.MTSlots (keys are slot numbers)
self.active_slots = {}
end
function Input:addSlot(value)
self:initMtSlot(value)
table.insert(self.MTSlots, self:getMtSlot(value))
self.active_slots[value] = true
self.cur_slot = value
end
function Input:addSlotIfChanged(value)
if self.cur_slot ~= value then
-- We've already seen that slot in this frame, don't insert a duplicate reference!
if self.active_slots[value] then
self.cur_slot = value
else
self:addSlot(value)
end
end
end
function Input:setupSlotData(value)
if #self.MTSlots == 0 then
self:addSlot(value)
else
self:addSlotIfChanged(value)
end
end
function Input:isEvKeyPress(ev)
return ev.value == EVENT_VALUE_KEY_PRESS
end
function Input:isEvKeyRepeat(ev)
return ev.value == EVENT_VALUE_KEY_REPEAT
end
function Input:isEvKeyRelease(ev)
return ev.value == EVENT_VALUE_KEY_RELEASE
end
--- Main event handling.
-- `now` corresponds to UIManager:getTime() (an fts time), and it's just been updated by UIManager.
-- `deadline` (an fts time) is the absolute deadline imposed by UIManager:handleInput() (a.k.a., our main event loop ^^):
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- it's either nil (meaning block forever waiting for input), or the earliest UIManager deadline (in most cases, that's the next scheduled task,
-- in much less common cases, that's the earliest of UIManager.INPUT_TIMEOUT (currently, only KOSync ever sets it) or UIManager.ZMQ_TIMEOUT if there are pending ZMQs).
function Input:waitEvent(now, deadline)
-- On the first iteration of the loop, we don't need to update now, we're following closely (a couple ms at most) behind UIManager.
local ok, ev
-- Wrapper around the platform-specific input.waitForEvent (which itself is generally poll-like, and supposed to poll *once*).
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- Speaking of input.waitForEvent, it can return:
-- * true, ev: When a batch of input events was read.
-- ev is an array of event tables, themselves mapped after the input_event <linux/input.h> struct.
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- * false, errno, timerfd: When no input event was read, possibly for benign reasons.
-- One such common case is after a polling timeout, in which case errno is C.ETIME.
-- If the timerfd backend is in use, and the early return was caused by a timerfd expiring,
-- it returns false, C.ETIME, timerfd; where timerfd is a C pointer (i.e., light userdata)
-- to the timerfd node that expired (so as to be able to free it later, c.f., input/timerfd-callbacks.h).
-- Otherwise, errno is the actual error code from the backend (e.g., select's errno for the C backend).
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- * nil: When something terrible happened (e.g., fatal poll/read failure). We abort in such cases.
while true do
if self.timer_callbacks[1] then
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- If we have timers set, we need to honor them once we're done draining the input events.
while self.timer_callbacks[1] do
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- Choose the earliest deadline between the next timer deadline, and our full timeout deadline.
local deadline_is_timer = false
local with_timerfd = false
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
local poll_deadline
-- If the timer's deadline is handled via timerfd, that's easy
if self.timer_callbacks[1].timerfd then
-- We use the ultimate deadline, as the kernel will just signal us when the timer expires during polling.
poll_deadline = deadline
with_timerfd = true
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
else
if not deadline then
-- If we don't actually have a full timeout deadline, just honor the timer's.
poll_deadline = self.timer_callbacks[1].deadline
deadline_is_timer = true
else
if self.timer_callbacks[1].deadline < deadline then
poll_deadline = self.timer_callbacks[1].deadline
deadline_is_timer = true
else
poll_deadline = deadline
end
end
end
local poll_timeout
-- With the timerfd backend, poll_deadline is set to deadline, which might be nil, in which case,
-- we can happily block forever, like in the no timer_callbacks branch below ;).
if poll_deadline then
-- If we haven't hit that deadline yet, poll until it expires, otherwise,
-- have select return immediately so that we trip a timeout.
now = now or time.now()
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if poll_deadline > now then
-- Deadline hasn't been blown yet, honor it.
poll_timeout = poll_deadline - now
else
-- We've already blown the deadline: make select return immediately (most likely straight to timeout).
-- NOTE: With the timerfd backend, this is sometimes a tad optimistic,
-- as we may in fact retry for a few iterations while waiting for the timerfd to actually expire.
poll_timeout = 0
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
end
end
local timerfd
local sec, usec = time.split_s_us(poll_timeout)
ok, ev, timerfd = input.waitForEvent(sec, usec)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- We got an actual input event, go and process it
if ok then break end
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- If we've drained all pending input events, causing waitForEvent to time out, check our timers
if ok == false and ev == C.ETIME then
-- Check whether the earliest timer to finalize a Gesture detection is up.
local consume_callback = false
if timerfd then
-- If we were woken up by a timerfd, that means the timerfd backend is in use, of course,
-- and it also means that we're guaranteed to have reached its deadline.
consume_callback = true
elseif not with_timerfd then
-- On systems where the timerfd backend is *NOT* in use, we have a few more cases to handle...
if deadline_is_timer then
-- We're only guaranteed to have blown the timer's deadline
-- when our actual select deadline *was* the timer's!
consume_callback = true
elseif time.now() >= self.timer_callbacks[1].deadline then
-- But if it was a task deadline instead, we to have to check the timer's against the current time,
-- to double-check whether we blew it or not.
consume_callback = true
end
end
if consume_callback then
local touch_ges
local timer_idx = 1
if timerfd then
-- If there's a deadline collision, make sure we call the callback that matches the timerfd returned.
-- We'll handle the next one on the next iteration, as an expired timerfd will ensure
-- that select will return immediately.
for i, item in ipairs(self.timer_callbacks) do
if item.timerfd == timerfd then
-- In the vast majority of cases, we should find our match on the first entry ;).
timer_idx = i
touch_ges = item.callback()
break
end
end
else
-- If there's a deadline collision, we'll just handle the next one on the next iteration,
-- because the blown deadline means we'll have asked waitForEvent to return immediately.
touch_ges = self.timer_callbacks[1].callback()
end
-- Cleanup after the timer callback.
-- GestureDetector has guards in place to avoid double frees in case the callback itself
-- affected the timerfd or timer_callbacks list (e.g., by dropping a contact).
if timerfd then
input.clearTimer(timerfd)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
end
table.remove(self.timer_callbacks, timer_idx)
if touch_ges then
self:gestureAdjustHook(touch_ges)
return {
Event:new("Gesture", self.gesture_detector:adjustGesCoordinate(touch_ges))
}
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
end -- if touch_ges
end -- if poll_deadline reached
else
-- Something went wrong, jump to error handling *now*
break
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
end -- if poll returned ETIME
-- Refresh now on the next iteration (e.g., when we have multiple timers to check, and we've just timed out)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
now = nil
end -- while #timer_callbacks > 0
else
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- If there aren't any timers, just block for the requested amount of time.
-- deadline may be nil, in which case waitForEvent blocks indefinitely (i.e., until the next input event ;)).
local poll_timeout
-- If UIManager put us on deadline, enforce it, otherwise, block forever.
if deadline then
-- Convert that absolute deadline to value relative to *now*, as we may loop multiple times between UI ticks.
now = now or time.now()
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
if deadline > now then
-- Deadline hasn't been blown yet, honor it.
poll_timeout = deadline - now
else
-- Deadline has been blown: make select return immediately.
poll_timeout = 0
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
end
end
local sec, usec = time.split_s_us(poll_timeout)
ok, ev = input.waitForEvent(sec, usec)
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
end -- if #timer_callbacks > 0
-- Handle errors
if ok then
-- We're good, process the event and go back to UIManager.
break
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
elseif ok == false then
if ev == C.ETIME then
-- Don't report an error on ETIME, and go back to UIManager
ev = nil
break
elseif ev == C.EINTR then -- luacheck: ignore
-- Retry on EINTR
else
-- Warn, report, and go back to UIManager
logger.warn("Polling for input events returned an error:", ev, "->", ffi.string(C.strerror(ev)))
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
break
end
elseif ok == nil then
-- Something went horribly wrong, abort.
logger.err("Polling for input events failed catastrophically")
local UIManager = require("ui/uimanager")
UIManager:abort()
break
end
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
-- We'll need to refresh now on the next iteration, if there is one.
now = nil
end
if ok and ev then
local handled = {}
-- We're guaranteed that ev is an array of event tables. Might be an array of *one* event, but an array nonetheless ;).
for __, event in ipairs(ev) do
if DEBUG.is_on then
-- NOTE: This is rather spammy and computationally intensive,
-- and we can't conditionally prevent evalutation of function arguments,
-- so, just hide the whole thing behind a branch ;).
DEBUG:logEv(event)
if event.type == C.EV_KEY then
logger.dbg(string.format(
"key event => code: %d (%s), value: %s, time: %d.%06d",
event.code, self.event_map[event.code] or linux_evdev_key_code_map[event.code], event.value,
event.time.sec, event.time.usec))
elseif event.type == C.EV_SYN then
logger.dbg(string.format(
"input event => type: %d (%s), code: %d (%s), value: %s, time: %d.%06d",
event.type, linux_evdev_type_map[event.type], event.code, linux_evdev_syn_code_map[event.code], event.value,
event.time.sec, event.time.usec))
elseif event.type == C.EV_ABS then
logger.dbg(string.format(
"input event => type: %d (%s), code: %d (%s), value: %s, time: %d.%06d",
event.type, linux_evdev_type_map[event.type], event.code, linux_evdev_abs_code_map[event.code], event.value,
event.time.sec, event.time.usec))
elseif event.type == C.EV_MSC then
logger.dbg(string.format(
"input event => type: %d (%s), code: %d (%s), value: %s, time: %d.%06d",
event.type, linux_evdev_type_map[event.type], event.code, linux_evdev_msc_code_map[event.code], event.value,
event.time.sec, event.time.usec))
elseif event.type == C.EV_REP then
logger.dbg(string.format(
"input event => type: %d (%s), code: %d (%s), value: %s, time: %d.%06d",
event.type, linux_evdev_type_map[event.type], event.code, linux_evdev_rep_code_map[event.code], event.value,
event.time.sec, event.time.usec))
else
logger.dbg(string.format(
"input event => type: %d (%s), code: %d, value: %s, time: %d.%06d",
event.type, linux_evdev_type_map[event.type], event.code, event.value,
event.time.sec, event.time.usec))
end
end
self:eventAdjustHook(event)
if event.type == C.EV_KEY then
local handled_ev = self:handleKeyBoardEv(event)
if handled_ev then
table.insert(handled, handled_ev)
end
elseif event.type == C.EV_ABS and event.code == ABS_OASIS_ORIENTATION then
local handled_ev = self:handleOasisOrientationEv(event)
if handled_ev then
table.insert(handled, handled_ev)
end
elseif event.type == C.EV_ABS or event.type == C.EV_SYN then
local handled_evs = self:handleTouchEv(event)
-- handleTouchEv only returns an array of Events once it gets a SYN_REPORT,
-- so more often than not, we just get a nil here ;).
if handled_evs then
for _, handled_ev in ipairs(handled_evs) do
table.insert(handled, handled_ev)
end
end
elseif event.type == C.EV_MSC then
local handled_ev = self:handleMiscEv(event)
if handled_ev then
table.insert(handled, handled_ev)
end
elseif event.type == C.EV_SDL then
local handled_ev = self:handleSdlEv(event)
if handled_ev then
table.insert(handled, handled_ev)
end
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
else
-- Received some other kind of event that we do not know how to specifically handle yet
A host of low power states related tweaks (#9036) * Disable all non power management related input during suspend. (This prevents wonky touch events from being tripped when closing a sleep cover on an already-in-suspend device, among other things). * Kobo: Use our WakeupMgr instance, not the class. * WakupMgr: split `removeTask` in two: * `removeTask`, which *only* takes a queue index as input, and only removes a single task. Greatly simplifies the function (i.e., it's just a `table.remove`). * `removeTasks`, which takes an epoch or a cb ref, and removes *every* task that matches. * Both of these will also *always* re-schedule the next task (if any) on exit, since we can have multiple WakeupMgr tasks queued, but we can only have a single RTC wake alarm set ;). * `wakeupAction` now takes a `proximity` argument, which it passes on to its `validateWakeupAlarmByProximity` call, allowing call sites to avoir having to duplicate that call themselves when they want to use a custom proximity window. * `wakeupAction` now re-schedules the next task (if any) on exit. * Simplify `Kobo:checkUnexpectedWakeup`, by removing the duplicate `WakerupMgr:validateWakeupAlarmByProximity` call, now that we can pass a proximity window to `WakeuoMgr:wakeupAction`. * The various network activity timeouts are now halved when autostandby is enabled. * Autostandby: get rid of the dummy deadline_guard task, as it's no longer necessary since #9009. * UIManager: The previous change allows us to simplify `getNextTaskTimes` into a simpler `getNextTaskTime` variant, getting rid of a table & a loop. * ReaderFooter & ReaderHeader: Make sure we only perform a single refresh when exiting standby. * Kobo: Rewrite sysfs writes to use ANSI C via FFI instead of stdio via Lua, as it obscured some common error cases (e.g., EBUSY on /sys/power/state). * Kobo: Simplify `suspend`, now that we have sane error handling in sysfs writes. * Kobo.powerd: Change `isCharging` & `isAuxCharging` behavior to match the behavior of the NTX ioctl (i.e., Charging == Plugged-in). This has the added benefit of making the AutoSuspend checks behave sensibly in the "fully-charged but still plugged in" scenario (because being plugged in is enough to break PM on `!canPowerSaveWhileCharging` devices). * AutoSuspend: Disable our `AllowStandby` handler when auto standby is disabled, so as to not interfere with other modules using `UIManager:allowStandby` (fix #9038). * PowerD: Allow platforms to implement `isCharged`, indicating that the battery is full while still plugged in to a power source (battery icon becomes a power plug icon). * Kobo.powerd: Implement `isCharged`, and kill charging LEDs once battery is full. * Kindle.powerd: Implement `isCharged` on post-Wario devices. (`isCharging` is still true in that state, as it ought to).
2022-05-01 21:41:08 +00:00
local handled_ev = self:handleGenericEv(event)
if handled_ev then
table.insert(handled, handled_ev)
end
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
end
end
return handled
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
elseif ok == false and ev then
return {
Event:new("InputError", ev)
}
The great Input/GestureDetector/TimeVal spring cleanup (a.k.a., a saner main loop) (#7415) * ReaderDictionary: Port delay computations to TimeVal * ReaderHighlight: Port delay computations to TimeVal * ReaderView: Port delay computations to TimeVal * Android: Reset gesture detection state on APP_CMD_TERM_WINDOW. This prevents potentially being stuck in bogus gesture states when switching apps. * GestureDetector: * Port delay computations to TimeVal * Fixed delay computations to handle time warps (large and negative deltas). * Simplified timed callback handling to invalidate timers much earlier, preventing accumulating useless timers that no longer have any chance of ever detecting a gesture. * Fixed state clearing to handle the actual effective slots, instead of hard-coding slot 0 & slot 1. * Simplified timed callback handling in general, and added support for a timerfd backend for better performance and accuracy. * The improved timed callback handling allows us to detect and honor (as much as possible) the three possible clock sources usable by Linux evdev events. The only case where synthetic timestamps are used (and that only to handle timed callbacks) is limited to non-timerfd platforms where input events use a clock source that is *NOT* MONOTONIC. AFAICT, that's pretty much... PocketBook, and that's it? * Input: * Use the <linux/input.h> FFI module instead of re-declaring every constant * Fixed (verbose) debug logging of input events to actually translate said constants properly. * Completely reset gesture detection state on suspend. This should prevent bogus gesture detection on resume. * Refactored the waitEvent loop to make it easier to comprehend (hopefully) and much more efficient. Of specific note, it no longer does a crazy select spam every 100µs, instead computing and relying on sane timeouts, as afforded by switching the UI event/input loop to the MONOTONIC time base, and the refactored timed callbacks in GestureDetector. * reMarkable: Stopped enforcing synthetic timestamps on input events, as it should no longer be necessary. * TimeVal: * Refactored and simplified, especially as far as metamethods are concerned (based on <bsd/sys/time.h>). * Added a host of new methods to query the various POSIX clock sources, and made :now default to MONOTONIC. * Removed the debug guard in __sub, as time going backwards can be a perfectly normal occurrence. * New methods: * Clock sources: :realtime, :monotonic, :monotonic_coarse, :realtime_coarse, :boottime * Utility: :tonumber, :tousecs, :tomsecs, :fromnumber, :isPositive, :isZero * UIManager: * Ported event loop & scheduling to TimeVal, and switched to the MONOTONIC time base. This ensures reliable and consistent scheduling, as time is ensured never to go backwards. * Added a :getTime() method, that returns a cached TimeVal:now(), updated at the top of every UI frame. It's used throughout the codebase to cadge a syscall in circumstances where we are guaranteed that a syscall would return a mostly identical value, because very few time has passed. The only code left that does live syscalls does it because it's actually necessary for accuracy, and the only code left that does that in a REALTIME time base is code that *actually* deals with calendar time (e.g., Statistics). * DictQuickLookup: Port delay computations to TimeVal * FootNoteWidget: Port delay computations to TimeVal * HTMLBoxWidget: Port delay computations to TimeVal * Notification: Port delay computations to TimeVal * TextBoxWidget: Port delay computations to TimeVal * AutoSuspend: Port to TimeVal * AutoTurn: * Fix it so that settings are actually honored. * Port to TimeVal * BackgroundRunner: Port to TimeVal * Calibre: Port benchmarking code to TimeVal * BookInfoManager: Removed unnecessary yield in the metadata extraction subprocess now that subprocesses get scheduled properly. * All in all, these changes reduced the CPU cost of a single tap by a factor of ten (!), and got rid of an insane amount of weird poll/wakeup cycles that must have been hell on CPU schedulers and batteries..
2021-03-30 00:57:59 +00:00
elseif ok == nil then
-- No ok and no ev? Hu oh...
return {
Event:new("InputError", "Catastrophic")
}
end
end
A host of low power states related tweaks (#9036) * Disable all non power management related input during suspend. (This prevents wonky touch events from being tripped when closing a sleep cover on an already-in-suspend device, among other things). * Kobo: Use our WakeupMgr instance, not the class. * WakupMgr: split `removeTask` in two: * `removeTask`, which *only* takes a queue index as input, and only removes a single task. Greatly simplifies the function (i.e., it's just a `table.remove`). * `removeTasks`, which takes an epoch or a cb ref, and removes *every* task that matches. * Both of these will also *always* re-schedule the next task (if any) on exit, since we can have multiple WakeupMgr tasks queued, but we can only have a single RTC wake alarm set ;). * `wakeupAction` now takes a `proximity` argument, which it passes on to its `validateWakeupAlarmByProximity` call, allowing call sites to avoir having to duplicate that call themselves when they want to use a custom proximity window. * `wakeupAction` now re-schedules the next task (if any) on exit. * Simplify `Kobo:checkUnexpectedWakeup`, by removing the duplicate `WakerupMgr:validateWakeupAlarmByProximity` call, now that we can pass a proximity window to `WakeuoMgr:wakeupAction`. * The various network activity timeouts are now halved when autostandby is enabled. * Autostandby: get rid of the dummy deadline_guard task, as it's no longer necessary since #9009. * UIManager: The previous change allows us to simplify `getNextTaskTimes` into a simpler `getNextTaskTime` variant, getting rid of a table & a loop. * ReaderFooter & ReaderHeader: Make sure we only perform a single refresh when exiting standby. * Kobo: Rewrite sysfs writes to use ANSI C via FFI instead of stdio via Lua, as it obscured some common error cases (e.g., EBUSY on /sys/power/state). * Kobo: Simplify `suspend`, now that we have sane error handling in sysfs writes. * Kobo.powerd: Change `isCharging` & `isAuxCharging` behavior to match the behavior of the NTX ioctl (i.e., Charging == Plugged-in). This has the added benefit of making the AutoSuspend checks behave sensibly in the "fully-charged but still plugged in" scenario (because being plugged in is enough to break PM on `!canPowerSaveWhileCharging` devices). * AutoSuspend: Disable our `AllowStandby` handler when auto standby is disabled, so as to not interfere with other modules using `UIManager:allowStandby` (fix #9038). * PowerD: Allow platforms to implement `isCharged`, indicating that the battery is full while still plugged in to a power source (battery icon becomes a power plug icon). * Kobo.powerd: Implement `isCharged`, and kill charging LEDs once battery is full. * Kindle.powerd: Implement `isCharged` on post-Wario devices. (`isCharging` is still true in that state, as it ought to).
2022-05-01 21:41:08 +00:00
-- Allow toggling the handling of most every kind of input, except for power management related events.
function Input:inhibitInput(toggle)
if toggle then
-- Only handle power management events
if not self._key_ev_handler then
logger.info("Inhibiting user input")
self._key_ev_handler = self.handleKeyBoardEv
self.handleKeyBoardEv = self.handlePowerManagementOnlyEv
end
-- And send everything else to the void
if not self._oasis_ev_handler then
self._oasis_ev_handler = self.handleOasisOrientationEv
self.handleOasisOrientationEv = self.voidEv
end
if not self._abs_ev_handler then
self._abs_ev_handler = self.handleTouchEv
self.handleTouchEv = self.voidEv
end
if not self._msc_ev_handler then
if not self.device:isPocketBook() and not self.device:isAndroid() then
-- NOTE: PocketBook is a special snowflake, synthetic Power events are sent as EV_MSC.
-- Thankfully, that's all that EV_MSC is used for on that platform.
-- NOTE: Android, on the other hand, handles a *lot* of critical stuff over EV_MSC,
-- as it's used to communicate between Android and Lua land ;).
self._msc_ev_handler = self.handleMiscEv
self.handleMiscEv = self.voidEv
end
A host of low power states related tweaks (#9036) * Disable all non power management related input during suspend. (This prevents wonky touch events from being tripped when closing a sleep cover on an already-in-suspend device, among other things). * Kobo: Use our WakeupMgr instance, not the class. * WakupMgr: split `removeTask` in two: * `removeTask`, which *only* takes a queue index as input, and only removes a single task. Greatly simplifies the function (i.e., it's just a `table.remove`). * `removeTasks`, which takes an epoch or a cb ref, and removes *every* task that matches. * Both of these will also *always* re-schedule the next task (if any) on exit, since we can have multiple WakeupMgr tasks queued, but we can only have a single RTC wake alarm set ;). * `wakeupAction` now takes a `proximity` argument, which it passes on to its `validateWakeupAlarmByProximity` call, allowing call sites to avoir having to duplicate that call themselves when they want to use a custom proximity window. * `wakeupAction` now re-schedules the next task (if any) on exit. * Simplify `Kobo:checkUnexpectedWakeup`, by removing the duplicate `WakerupMgr:validateWakeupAlarmByProximity` call, now that we can pass a proximity window to `WakeuoMgr:wakeupAction`. * The various network activity timeouts are now halved when autostandby is enabled. * Autostandby: get rid of the dummy deadline_guard task, as it's no longer necessary since #9009. * UIManager: The previous change allows us to simplify `getNextTaskTimes` into a simpler `getNextTaskTime` variant, getting rid of a table & a loop. * ReaderFooter & ReaderHeader: Make sure we only perform a single refresh when exiting standby. * Kobo: Rewrite sysfs writes to use ANSI C via FFI instead of stdio via Lua, as it obscured some common error cases (e.g., EBUSY on /sys/power/state). * Kobo: Simplify `suspend`, now that we have sane error handling in sysfs writes. * Kobo.powerd: Change `isCharging` & `isAuxCharging` behavior to match the behavior of the NTX ioctl (i.e., Charging == Plugged-in). This has the added benefit of making the AutoSuspend checks behave sensibly in the "fully-charged but still plugged in" scenario (because being plugged in is enough to break PM on `!canPowerSaveWhileCharging` devices). * AutoSuspend: Disable our `AllowStandby` handler when auto standby is disabled, so as to not interfere with other modules using `UIManager:allowStandby` (fix #9038). * PowerD: Allow platforms to implement `isCharged`, indicating that the battery is full while still plugged in to a power source (battery icon becomes a power plug icon). * Kobo.powerd: Implement `isCharged`, and kill charging LEDs once battery is full. * Kindle.powerd: Implement `isCharged` on post-Wario devices. (`isCharging` is still true in that state, as it ought to).
2022-05-01 21:41:08 +00:00
end
if not self._sdl_ev_handler then
self._sdl_ev_handler = self.handleSdlEv
self.handleSdlEv = self.voidEv
end
if not self._generic_ev_handler then
self._generic_ev_handler = self.handleGenericEv
self.handleGenericEv = self.voidEv
end
-- Reset gesture detection state to a blank slate, to avoid bogus gesture detection on restore.
self:resetState()
A host of low power states related tweaks (#9036) * Disable all non power management related input during suspend. (This prevents wonky touch events from being tripped when closing a sleep cover on an already-in-suspend device, among other things). * Kobo: Use our WakeupMgr instance, not the class. * WakupMgr: split `removeTask` in two: * `removeTask`, which *only* takes a queue index as input, and only removes a single task. Greatly simplifies the function (i.e., it's just a `table.remove`). * `removeTasks`, which takes an epoch or a cb ref, and removes *every* task that matches. * Both of these will also *always* re-schedule the next task (if any) on exit, since we can have multiple WakeupMgr tasks queued, but we can only have a single RTC wake alarm set ;). * `wakeupAction` now takes a `proximity` argument, which it passes on to its `validateWakeupAlarmByProximity` call, allowing call sites to avoir having to duplicate that call themselves when they want to use a custom proximity window. * `wakeupAction` now re-schedules the next task (if any) on exit. * Simplify `Kobo:checkUnexpectedWakeup`, by removing the duplicate `WakerupMgr:validateWakeupAlarmByProximity` call, now that we can pass a proximity window to `WakeuoMgr:wakeupAction`. * The various network activity timeouts are now halved when autostandby is enabled. * Autostandby: get rid of the dummy deadline_guard task, as it's no longer necessary since #9009. * UIManager: The previous change allows us to simplify `getNextTaskTimes` into a simpler `getNextTaskTime` variant, getting rid of a table & a loop. * ReaderFooter & ReaderHeader: Make sure we only perform a single refresh when exiting standby. * Kobo: Rewrite sysfs writes to use ANSI C via FFI instead of stdio via Lua, as it obscured some common error cases (e.g., EBUSY on /sys/power/state). * Kobo: Simplify `suspend`, now that we have sane error handling in sysfs writes. * Kobo.powerd: Change `isCharging` & `isAuxCharging` behavior to match the behavior of the NTX ioctl (i.e., Charging == Plugged-in). This has the added benefit of making the AutoSuspend checks behave sensibly in the "fully-charged but still plugged in" scenario (because being plugged in is enough to break PM on `!canPowerSaveWhileCharging` devices). * AutoSuspend: Disable our `AllowStandby` handler when auto standby is disabled, so as to not interfere with other modules using `UIManager:allowStandby` (fix #9038). * PowerD: Allow platforms to implement `isCharged`, indicating that the battery is full while still plugged in to a power source (battery icon becomes a power plug icon). * Kobo.powerd: Implement `isCharged`, and kill charging LEDs once battery is full. * Kindle.powerd: Implement `isCharged` on post-Wario devices. (`isCharging` is still true in that state, as it ought to).
2022-05-01 21:41:08 +00:00
else
-- Restore event handlers, if any
if self._key_ev_handler then
logger.info("Restoring user input handling")
self.handleKeyBoardEv = self._key_ev_handler
self._key_ev_handler = nil
end
if self._oasis_ev_handler then
self.handleOasisOrientationEv = self._oasis_ev_handler
self._oasis_ev_handler = nil
end
if self._abs_ev_handler then
self.handleTouchEv = self._abs_ev_handler
self._abs_ev_handler = nil
end
if self._msc_ev_handler then
self.handleMiscEv = self._msc_ev_handler
self._msc_ev_handler = nil
end
if self._sdl_ev_handler then
self.handleSdlEv = self._sdl_ev_handler
self._sdl_ev_handler = nil
end
if self._generic_ev_handler then
self.handleGenericEv = self._generic_ev_handler
self._generic_ev_handler = nil
end
end
end
--[[--
Request all input events to be ignored for some duration.
@param set_or_seconds either `true`, in which case a platform-specific delay is chosen, or a duration in seconds (***int***).
]]
function Input:inhibitInputUntil(set_or_seconds)
local UIManager = require("ui/uimanager")
UIManager:unschedule(self._inhibitInputUntil_func)
if not set_or_seconds then -- remove any previously set
self:inhibitInput(false)
return
end
local delay_s
if set_or_seconds == true then
-- Use an adequate delay to account for device refresh duration
-- so any events happening in this delay (ie. before a widget
-- is really painted on screen) are discarded.
if self.device:hasEinkScreen() then
-- A screen refresh can take a few 100ms,
-- sometimes > 500ms on some devices/temperatures.
-- So, block for 400ms (to have it displayed) + 400ms
-- for user reaction to it
delay_s = 0.8
else
-- On non-eInk screen, display is usually instantaneous
delay_s = 0.4
end
else -- we expect a number
delay_s = set_or_seconds
end
UIManager:scheduleIn(delay_s, self._inhibitInputUntil_func)
self:inhibitInput(true)
end
return Input