Use FT_CONFIG_OPTION_USE_HARFBUZZ

[radarhere]

Is this a viable simplification for python-pillow#9913?

[akx]

AIUI, FreeType defaults to loading HarfBuzz dynamically unless it's forced as a dependency and found at build time. (And as described in the original issue, that dynamic linkage won't work due to auditwheel's renames.)

But I'll take a look at the libraries that get built here...

[akx]

... I have to keep re-running the macOS workflow because test wheel is flaking on the leak test (fixed in python-pillow#9945) and it's not uploading the artifact before tests work...

[radarhere]

I'm willing to merge the skipping of PyPy leak tests, but less keen on reducing the size for test_empty_image. See what you think of python-pillow#9957

[akx]

but less keen on reducing the size for test_empty_image.

AISI, the idea of the test is to check that these if statements have teeth?

https://github.com/python-pillow/Pillow/blob/3dfc89ea2fd9143e4a3f405da435b9357c4ffcfb/src/libImaging/Fill.c#L26-L29

... and it does so by assuming that image initialization doesn't take over an arbitrary duration. I have another idea, let me open up a PR...

See what you think of python-pillow#9957

Replied there. I don't think it has the intended effect.

[radarhere]

python-pillow#9933 is merged. That should help you evaluate this.

[akx]

Finally managed to take a look at the wheel.

Yeah, looks like the correct symbols are there in this build too.

$ nm -u PIL/.dylibs/libfreetype.6.dylib | grep _hb_ot_
_hb_ot_layout_collect_lookups
_hb_ot_layout_lookup_collect_glyphs
_hb_ot_layout_lookup_would_substitute
_hb_ot_tags_from_script_and_language