* Run grep against /proc/modules directly instead of forking to lsmod
* Revamp the CPUFreq governor choice logic
I finally figured out why the scaling on the H2O was so wonky:
Because it relies on an obscure i.MX hardware feature called DVFS,
and it gets flipped for mysterious workaround-y reasons depending on
Wi-Fi state...
* Start playing with conservative, because just staying pegged at max
clock is entirely stupid.
And DVFS is extremely conservative, it needs somewhat sustained load to
clock up...
* Take care of the DVFS switcheroo in the Wi-Fi scripts
Update shellcheck and shfmt to the latest version.
Fixes <https://github.com/koreader/koreader/issues/5152>.
Btw, you can apply shellcheck suggestions with a command like:
```
shellcheck --include=SC2250 -f diff *.sh | git apply
```
Parsing and rasterisation provided by nanosvg (some svg tags like text,
span and images are not supported).
Only svg image included as files in the .epub are supported
(embedded svg tags in html are not).
Holding on svg image shows it 4x zoomed in ImageViewer.
* Travis: speed up by caching base and running luacheck earlier
* ignore bin and install for git status change detection
* skip coverage except on official master branch. It adds 3 whole minutes and does nothing to prevent regressions
* also cache ~/.luarocks. It evens out but would generally prevent remote timeout shenenigans
* remove base cache dir before caching with verbose remove to see what's going on
* more inclusive shell code quality analysis
* fixed more shellcheck issues
* better shellcheck/shfmt debugging info
and have more confidence with the unit testing framework.
Now `make testfront` won't retry on failure and testing files are
ordered in each run so that it's possible to reproduce testing failure.
And this patch also fix flush settings not working before suspend issue:
at some point the `FlushSettings` event is sent to `UIManager` instead
of `ReaderUI`, but `UIManager` only delegated events to active widgets
and `ReaderUI` is actually not an active widgets thus will miss the event.
This patch also add a verbose debug mode with "-v" as a switch to turn
on this mode. With verbose mode on, event handling will be logged.
1. It turns out that `device_id` is not defined in unit test environment
which makes the `assert(self.kosync_device_id)` fail and somehow makes `luajit` segmentation fault.
2. In the device spec, stubbing `io.open` will make subsquent `require`
all return boolean value and segmentation fault `luajit`.
This patch is a minimum viable change that makes CI work on both my
local box and Travis.
