[ft-devel] Pixel Differences after 2ea511eed8 and 7a81b63abc2b3

[Dominik Röttsches]

Hi Werner, Alexei,

in Chrome, I am encountering issues rolling to latest FreeType ToT, see details in https://bugs.chromium.org/p/chromium/issues/detail?id=962932

In particular, these two CLs

(1) https://git.savannah.gnu.org/cgit/freetype/freetype2.git/commit/?id=7a81b63abc2b3da0d7f0950f69377d2b3f54b0fb Optimize Bézier bisections.
(2) https://git.savannah.gnu.org/cgit/freetype/freetype2.git/commit/?id=2ea511eed81770f423544525adebf7f954b8be93 [smooth] Simplify cubic Bézier flattening. 

are each triggering hundreds of layout test failures due to pixel differences.

I appreciate the improved clarity of the code in (1), I agree there are readability improvements. However, it does unfortunately trigger pixel difference in almost all layout tests that render... a font. 

Is there a chance (1) could be modified to not cause different results - or is the intention to keep it as is? Has this change been shown to improve performance? 

Similarly, after (2) (independently) a large number of layout tests is affected.

Do you have additional pending plans for upcoming pixel difference changes? This would help avoiding additional rounds of rebaselines.

The process of regenerating big rebaselines in Chrome causes quite some work. In that sense, I would kindly ask to be careful when landing such changes. Explaining expected rendering result changes in the commits, or adding more context such as a bug reference or link to a mailing list post with more background or benchmark results is always helpful for understanding the benefit of such changes.

Thanks,

Dominik