Variable font: odd difference between hb-shape and uharfbuzz about uharfbuzz HOT 12 CLOSED

Note that at TEST=509 the values do match, which makes me think perhaps a negative delta gets rounded differently between the two environments.

from uharfbuzz.

Please try with --font-funcs set to ot vs ft. I feel like that's where the difference is coming from.

from uharfbuzz.

Ha, calling hb-shape with --font-funcs=ot indeed makes a difference. The output then becomes:

482 - 491 491 491.0 True
491 - 495 495 495.5 True
500 - 500 500 500.0 True
509 - 505 504 504.5 False
518 - 509 509 509.0 True

Which is weird, because that also doesn't match the output of uharfbuzz, but differently.

from uharfbuzz.

I have to dig in, but I wonder if Python is changing FPU rounding mode..

from uharfbuzz.

am investigating this also

from uharfbuzz.

Okay, this is fixed with 985eda0 I'll add it as a test also

from uharfbuzz.

What fixed it?

from uharfbuzz.

I'm still seeing the same results in my local build (macOS 10.15)

from uharfbuzz.

What fixed it?

Good question! Maybe somewhere in skbuild / cmake process another roundf implementation / libm was inserted.

I'm still seeing the same results in my local build (macOS 10.15)

Are you passing --font-funcs=ot?

from uharfbuzz.

Of course I should test whether I had the issue before the change, sorry for closing eagerly, at least I can say I don't have the difference on my machine and that's regardless of passing HAVE_ROUNDF to harfbuzz.cc compile or not.

from uharfbuzz.

I updated my hb-shape executable, and the results now indeed match when using --font-funcs=ot, so my original report (based on HB 2.6.2) was out of date. So there's indeed no issue with uharfbuzz.

That said, I'd argue that the values returned with --font-funcs=ot are incorrectly rounded, and should match the results of --font-funcs=ft, but that would be a HB issue.

from uharfbuzz.

Thank you!

That said, I'd argue that the values returned with --font-funcs=ot are incorrectly rounded, and should match the results of --font-funcs=ft, but that would be a HB issue.

Feel free to file that as a bug against harfbuzz.

from uharfbuzz.