Split the html page download out of createEpub, so that createFromDescription
can pass in its own html for epub packaging (with optional image download).
Still pass in a url, so that relative urls can be made absolute.
- Popup footnotes: workaround strange behaviour of MuPDF
which is adding some unwanted margin above elements
with an "id=" attribute.
- In-page footnotes: also remove any left and right margin,
to keep footnotes aligned with the main text.
Also add 2 classnames (witnessed in quite a few recent
french books) to the list of classic classnames.
- Also bump crengine: CSS: fix parsing of multi classnames
and attributes selectors
Bump crengine:
- Ensure text decoration (underline) is continued over word gaps
- Invalidate TOC page numbers on rendering change
- Ensure reproducible cache files when same rendering settings
- LVBlockWriteStream: workaround to exclude fatal error
- Use utf8proc for string uppercase/lowercase/capitalize
Bump base:
Thirdparty: adds utf8proc 2.3.0 (libutf8proc.so.2)
For use by crengine, for now mostly for more complete and
accurate text-transform: uppercase/lowercase/capitalize.
Also revert test tweak (in 27ddd6f) to workaround an issue with
cre cache that should be solved by this crengine bump.
Fixes#5005
Results from re-running "Download news" should be as before, except when the
article content has changed between runs, but the title has not and an updated
date has not been given in the feed. User can force it to re-download an
article by manually deleting the file then running "Download news".
Either at all, or because it would impose reduced functionality.
Meaning:
* On 4bpp fb, because those are completely unsupported by the C BB
* On 24bpp fb, because it has limited support for that target format (no alpha blending).
* On 8bpp Linux eInk fb with an inverted palette (that's basically only the K4 ;)).
* On Linux eInk devices w/ no HW inversion support, as that would render NightMode inoperable.
That last point means that it's currently *always* blacklisted on PocketBook, because I don't have a PB device, so I don't know how safe HW inversion would be, meaning none of them are marked as being HW invert capable.
(I was a tiny bit hasty with https://github.com/koreader/koreader-base/pull/912 ;).)
Without having to resort to weird custom defaults.
* Split the current margins setting in three:
* Horizontal margins (because you generally want those two to be balanced).
* Top margin & Bottom margin (because you may want to tweak those separately to deal with quirky status bar/final line shenanigans).
* Also, add a "Reclaim bar height from bottom margin" toggle to the status bar menu, to optionally make sure the status bar won't eat into the bottom margin.
* Includes a free fix to diacritics popup refresh handling in the keyboard ;).
Wrap the whole of loadConfigAndProcessFeeds with Trapper:wrap so that feeds are downloaded sequentially, otherwise they become interleaved and the UI progress updates get confusing. Use Trapper:info in place of manually constructed InfoMessages.
Lift the Trapper:wrap call out of the individual article processing code,
so that articles are processed one by one, in order to:
* Avoid concurrent progress updates fighting over the UI dialog
* Avoid trying to download many images at the same time
Allow the user to specify whether to download images for each individual
feed specified in feed_config.lua. Default to false to stay closest to
existing behaviour.
It should be a path to a directory, not a file.
https://circleci.com/docs/2.0/configuration-reference/#store_test_results
> Path (absolute, or relative to your working_directory) to directory containing subdirectories of JUnit XML or Cucumber JSON test metadata files
Perhaps the file needs to be named results.xml. We'll see.
Some combination of Travis and/or older Busted necessitated insufficient insulation.
Follow-up to https://github.com/koreader/koreader/pull/4970
Space condensing carried over; reset at the start of readertoc.
