This is a major overhaul of the hardware abstraction layer.
A few notes:
General platform distinction happens in
frontend/device.lua
which will delegate everything else to
frontend/device/<platform_name>/device.lua
which should extend
frontend/device/generic/device.lua
Screen handling is implemented in
frontend/device/screen.lua
which includes the *functionality* to support device specifics.
Actually setting up the device specific functionality, however,
is done in the device specific setup code in the relevant
device.lua file.
The same goes for input handling.
so that when document is modified the persistent cache will
be invalidated automatically because the cache key will not
be matched. There is no perfermance overhead here at all. We
even don't need to check the modification time of the cache item
on disk, because the name of the on disk cache is a md5sum of the
cacheitem key, now the filename of the cache files contains the
modification time information.
If the document is modified since one rendered page is cached to disk,
the cache key won't match the cache file. And the cache file will
be discarded without the need to open the cache file or to check
the modification time of the cache file itself.
In regional zoom mode double tap will zoom to the tapped
region(paragraph or column, etc., detected optically via libk2pdfopt).
As the first demo, this feature is only turned on in flipping mode by
tapping the top-left corner of the screen. Eventually we may incorporate
this feature in "free" zoom mode.
that has a text layer underneath.
For multi-column page reflowing usually fits individual column to
page width and no finner word boxes are available to map reflowed
word box onto native word box. Now the relative position of a reflowed
word center is calculated as portion of the width and height of a larger
block it belongs. This relative position is also considered in
reflow-to-native position transform so that dictionary lookup now works
on multi-column pages in reflowing mode.
This should fix#376.
This should finish the work to make all globals to local variables.
That allows LuaJIT to properly compile things by interning the
references to the relevant parts (rather than looking up globals
all the time which stops a trace).
If getAutoBBox is fed with a blank page the returned bbox will be
much smaller than the native page dimensions. The problem is that
in zoom to 'content*' mode document renderer will try to render
the whole page at a zoom level that can cover the screen with the
bbox region of the page. That's if the blank page size is 3000*4000
and the returned autobbox is 300*400, the screen size is 600*800,
then the page will be rendered in 3000*(600/300)*4000*(800/400)
which is so large that the cache manager will not accept.
This patch just check if the returned autobbox is considerablely
much smaller than the native page dimensions. If so it will return
the manual bbox.
This patch should fix#293. And it's tested with the case provided
by @Markismus in #291.
and there is no need to OCR word if we can extract text from
original page in reflowing mode thanks to the rect maps between
normal page and reflowed page.
Semi-automatic cropping is implemented with a initial
manual cropping that exclude the obvious decorating line or
the spine part in scanned books and then an automatic cropping
ie. background/foreground pixel detection restricted within
the manual cropping area.
This PR is a candidate for the solution of issue #141.
