search

googlefonts / fontc

Where in we pursue oxidizing (context: https://github.com/googlefonts/oxidize) fontmake.
Apache License 2.0
73 stars 12 forks source link

compiling GoogleSans with fontc gets stuck in an infinite loop #365

Closed anthrotype closed 1 year ago

anthrotype commented 1 year ago

Running cargo run -- --source ../googlesans/source/GoogleSans/GoogleSans.designspace off latest main branch never stops. A thread panics because of https://github.com/googlefonts/fontc/issues/349 'no open subpath (missing MoveTo)', but I don't think that's the problem because even if I fix the latter issue in a WIP branch of mine (where kurbo no longer panics because all BezPaths start with valid MoveTo), fontc continues forever until I CTRL-C.

I suspect it might have to do something with #359, because if I revert that PR it no longer hangs -- in my WIP branch at least.

anthrotype commented 1 year ago

even if I fix the latter issue in a WIP branch of mine (where kurbo no longer panics because all BezPaths start with valid MoveTo), fontc continues forever until I CTRL-C.

correction, in my WIP branch (where the thread does not panic) fontc does complete eventually. But it takes more than 2 minutes and is almost 4 times slower than when I also revert #359.

cmyr commented 1 year ago

will take a look at this in a little bit. :)

cmyr commented 1 year ago

Okay I dug in a bit: my best guess right now is that we aren't really handling panics at all, so we're getting into a deadlock where some job depends on another job that can't finish.

We have a few options for fixing this. The easiest in the short-term is probably just to track the time of the last completed job, and then timeout if we don't make progress after some time. We should probably do this anyway, as a backup. More specifically, though, I think we need some way to signal a panic has occurred, probably by using rayon's ThreadPoolBuilder to set a custom panic handler, and maybe we can set a flag in the panic handler and check that in the main loop.

anthrotype commented 1 year ago

is almost 4 times slower than when I also revert https://github.com/googlefonts/fontc/pull/359

maybe this warrants its own issue. Something is going on with #359 that has slowed down things noticeably. Building GoogleSans.designspace takes 27 seconds after #359, and only 5 seconds if I revert it (timings are with fontc built in release mode and with --emit-ir=false, debug or emit-ir=true make it even slower). I am not able to figure out what in #359 would produce such a major (5x) slow-down.