Nico Weber
Nico Weber
Rebased again after #25112, which contained most of the plumbing, and after #25111, which made butt caps look nicer and added support for square caps.
I updated the test to look like Tests/LibPDF/paths.pdf. I'm pretty happy with this now. I'm going to merge it once CI is done.
+1, just removing the FIXME seems like the thing to do here 🙂
@alimpfard
Do you know what caused the regression?
A while ago, I tried taking the current ResolutionLevelPositionComponentLayerProgressionIterator approach but micro-optimizing it a bit: https://github.com/nico/serenity/commits/jpeg2000-prog2-memmo That brings decoding time from ~3.4s to ~1.08s. That's a pretty decent win, but...
@LucasChollet FYI, in case you're looking for interesting things to look at :)
https://github.com/LaurenzV/hayro/blob/225691a9638a347bda861a544b121ea5fbd43378/hayro-jpeg2000/src/progression.rs#L188-L258 could be a good approach: precompute precinct bounds across all component tiles and resolutions, and then just vary what they're sorted by. Credit to the hayro folks for this...
I'm guessing that's because WEB_API forces emission of inlines. If you aren't already, you could consider using https://blog.llvm.org/2018/11/30-faster-windows-builds-with-clang-cl_14.html to make the win build more like the other platforms (makes the...
linkify d458471e094f1c9ae3407eb2c6248278ae37cac3