Closed MarimeGui closed 4 years ago
A test font I used showed no diff with the new atan2, but my test font is kinda bad. I'll swap to libm's correct atan2 since this doesn't impact performance, all these slower math bits are part of the precomputation on font creation and not the raster pipeline. Thanks for catching this.
Can you give it a try with Sun's atan function (what libm uses)? I assume for fonts with not enough curves, the inaccuracy actually looks bad. The fonts I test with all are wall over done with geometry (Like roboto). Here's the commit: https://github.com/mooman219/fontdue/commit/f0054b67a709d953ab19107dbdb20616e62a908c
Yup, that seems to have done the trick ! The rendered characters look exactly like before ! Thanks for the swift modification, I'll close this now.
Awesome, I'll publish 0.1.1 today eventually
Hello again !
After upgrading my console from version 0.0.4 to 0.1.0 of fontdue, visual quality of the rendered characters significantly degraded.
Here is what it looked like before:
This is what it looks like now:
(I am using font size 15.0 btw)
Certain characters look different, just slightly "denser" or "thinner" in certain areas. This is especially visible on 'o', 'a', 'g' and 'c' characters (mostly the ones that curve apparently).
It traced it back to this commit, fixing the std problem where std ops were used by implementing a custom atan2 fn.
I tried replacing it by copy-pasting the one found in libm, and it fixed this issue, restoring back the look before this commit.
I tried looking for what's wrong in the new code, but I'm not smart enough sadly.
Do you think a fix is possible ?
Thanks again, Marime Gui