The parsing step is considerably slower in Python, but if parsing
is moved out of the test function, Python wins at 45ms to Rust's 67ms
on 10,000 rounds, presumably due to the overhead of serializing to
Protobuf. Not enough of a difference to justify the inclusion of extra
dependencies and duplicating the lookup code in any case.
- translation files are now stored in a separate repo, and
use a layout compatible with Pontoon
- normalize the language code in aqt, so that old config settings
and command line arguments are correctly handled
- store Qt and gettext translations in separate subfolders
- remove Crowdin scripts
- Adding files inside Anki now updates the media DB, so a full file
scan at sync time is no longer required if no other changes have been
made.
- Use a protobuf message for backend initialization, and return a
string error if initialization fails.
It often arrives that I want to know when a card is going to be
flushed and in this case change it.
This could be the case if I want to change the scheduler without
implementing a whole scheduler. It simply reads the card history and
change interval and due date.
It's also the case for the "'trigger -> action' rules", which apply
some coded actions when some event occurs. E.g. suspend/unsuspend a
sibling when card become mature/is forgotten.
I created multiple add-ons which want to transform a note before it is
being saved. For example, one add-on trim it, and remove useless line
break which arrived by accident.
Another add-on want to compile LaTeX as soon as the note is done, and
warn the user if LaTeX can't be compiled.
Having a hook in pre-flush would be useful here
This change would help me simplify the code of
https://ankiweb.net/shared/info/1983204951
It would allow my exporter to use the code of AnkiExporter while
selecting which decks I export, instead of exporting all decks as I
was doing until today. (I never realized I was exporting all deck
actually, until someone sent me a bug report)
- _renderQA() has moved to template.py:render_card()
- dropped QAData in favour of a properly typed dict
- render_card() returns a TemplateRenderOutput struct instead of a dict
- card_did_render now takes that output as the first arg, and can
mutate it
- TemplateRenderContext now stores the original card, so it can return
a card even in the add screen case
The old mungeFields and mungeQA hook have been removed as part of this
change. mungeQA can be replaced with the card_did_render hook.
In the mungeFields case, please switch to using field_filter instead.
We can now show replay buttons for the audio contained in {{FrontSide}}
without having to play it again when the answer is shown.
The template code now always defers FrontSide rendering, as it wasn't
a big saving, and meant the logic had to be implemented twice.
Forces the Fusion theme when running night mode, so we don't need
to work around platform themes that don't respond to the defined
palette.
Feedback/suggestions on the chosen colours welcome - _vars.scss is the
file to change if you want to experiment with adjustments.
- the new API operates on AVTags so it can support TTS as well as
audio files
- added a simple "run for each file" implementation for mpv and mplayer.
- will need to test handling of unicode filenames on Windows
- the old mpv and mplayer code is currently not active
This simply wraps the field in extra text that the frontend will
deal with. Also added some helpers for extracting and stripping
audio and TTS tags from the rendered text.
- the context exists for the lifecycle of one card's render,
and caches calls to things like .card() to avoid add-ons needing to
do their own cache management.
- add-ons can optionally add extra data to the context if they need
it across multiple filters
- removed card_will_render. the legacy hook is still available for
now
- card_did_render is now called only once, with both front and back
text
This commit solve a problem I actually have for some strange
reason. Here is how to reproduce the problem
1. Create a note of type Basic (and reversed card) in a deck D with front "First"
and no back
2. Wait a day (or install an add-on which show hour and minutes of
creation time in browser)
3. create a second note, with front "Last" and no back
4. In first note, add in back field "First" and delete the front field
5. Use "Empty card".
6. In the setting of the deck D, set in random order, and then back in
the "order added"
7. Review deck D. You'll see card 1 of the second note. (Don't review
it)
8. Open the browser. Show the column "due" and created. You'll see
that the first card (first) have due value 2, while it was created
before according to the created date. The card "last" have due value
1.
This is due to the fact that the value "created" is linked to note
creation, while the order is linked to the card creation time, and
card may be created after other note
The type arg is no longer used, as neither type 0 nor 1 appears to
have been used in the codebase.
By using the existing card ids, it allows add-ons that gather
information about a card to work properly in the card template screen
without extra hacks.
This allows us to add a docstring to .append() so users can see
the names of the arguments that are being passed, and means we
don't have to remember to prepend run_ when calling a hook.
these builds are now available on
https://github.com/ankitects/anki-typecheck
The add-on docs will be updated shortly to provide info on
using the new hooks and type checking.
Still todo:
- Add separate module for GUI hooks
- Update the remaining runHook/runFilter() calls
- Document the changes, including defensive registration
- The front and back are rendered in one call now. If the front
side contains no custom filters, we can bake {{FrontSide}} into the
rear side. If it did contain custom filters, we return the partially
complete rear template instead, and the calling code can inject
the FrontSide in after it has been fully rendered.
- Instead of modifying "cloze" into something like "cq-2", the card
ordinal and whether we're rendering the question or answer are now
passed in to the rendering filters as context.
- The Rust code doesn't need to support filter names split on '-'
anymore.
- Drop the "Show" part of hint descriptions so i18n support can be
deferred.
- Ignore blank filter names caused by user using two colons instead
of one.
- Fixed hint field and text transposition.
This is paving the way to move the standard filters into Rust.
Non-empty fields are now determined in Rust, using a single regex
instead of the overkill stripHTMLMedia(). The old implementation
has been moved into the Pystache code for now.
This extends the existing Rust code to handle conditional
replacement. The replacement of field names and filters to text
remains in Python, so that add-ons can still define their own
field modifiers.
The code is currently running the old Pystache rendering and the
new implementation in parallel, and will print a message to the
console if they don't match. If you notice any problems, please
let me know.
In the process of factoring out the field filtering, the "extra"
and "fullname" args are just passed in as a blank string now.
Extra was functionality that allowed a field modifier to be defined
as "filtername(arg1,arg2):field", and fullname was the name of the
field including any provided field modifiers. From grepping through
the add-ons on AnkiWeb, neither appears to have been used.
pytest will show what differs in simple assert statements
concurrent mode is supported with a plugin, but like nose2, concurrent
mode hides the cause of import errors, so I've left it off for now.
