Open chrissimpkins opened 2 years ago
"Failed to build compreffor pyclipper"
Seems like Cython has problems with 3.11. Not sure how easily this is fixed.
what's the error? maybe file a bug
It's this one: https://github.com/aio-libs/aiohttp/issues/6600
Haven't looked into if this could be fixed by a version bump somewhere.
compreffor issue with cpython 3.11 should now be fixed with https://pypi.org/project/compreffor/0.5.2 please try again
also fixed the same issue in pyclipper https://pypi.org/project/pyclipper/1.3.0.post3/
Installs now, kthx.
Results for the cut-instances
branch, which excludes Telugu:
Benchmark 1: make -j (CPython 3.8.10)
Time (mean ± σ): 483.297 s ± 0.042 s [User: 3020.954 s, System: 18.076 s]
Range (min … max): 483.267 s … 483.327 s 2 runs
Benchmark 1: cd ../googlesans-cp311 ; make -j (CPython 3.11b3)
Time (mean ± σ): 404.991 s ± 1.003 s [User: 2513.287 s, System: 12.531 s]
Range (min … max): 404.281 s … 405.700 s 2 runs
Benchmark 1: cd ../googlesans-pypy ; make -j (Pypy 7.3.9)
Time (mean ± σ): 345.172 s ± 1.359 s [User: 1673.498 s, System: 17.142 s]
Range (min … max): 344.211 s … 346.133 s 2 runs
So, 3.11b3 is 16% faster than 3.8 and Pypy is 29% faster than 3.8. Note: I upgarded all deps to their newest version for 3.11 but saw no obvious unfairness.
Beauty.
Maybe report back upstream. 👏
@madig The results in https://github.com/googlefonts/oxidize/issues/26#issuecomment-1154412465 are from GS compiles?
How is memory use in the cPy 3.11 vs. cPy < 3.11 comparison? Any significant changes there?
Related mem issues with pypy3 compile time performance improvements documented in #28
@madig The results in #26 (comment) are from GS compiles?
Yes, from before Telugu landed.
I have not measured memory usage.
cPython 3.11 betas are documented to be ~25% faster than cPy 3.10 with improvements in interpreter startup and run times.
Do the cPy 3.11 changes improve font compile times with the fontmake-based Python compiler pipeline?