Re: [ft] Regression in font clarity with Adobe engine

[Dave Arnold]

Hi Mike,

I have reproduced your results with version 0.0.15 of Cantarell-Regular (CFF). 
I do not see any FreeType bugs, but have found some system integration and font 
issues. Here is what I have found.

1) The new CFF rasterizer automatically increases the weight of fonts at small 
sizes, in order to maintain contrast and improve readability. To do this, it 
assumes the system is adjusting gamma for text (as Werner mentioned earlier). 
In most displays, I find a compromise of gamma 1.8 is best, but anything down 
to 1.4 should look ok. Your screen shot shows that that linearized blending is 
not being used on the text. That is, gamma adjustment is not being done, and 
your gamma is 1.0. Using gamma 1.0 has several negative effects on 
anti-aliasing; one is to make black-on-white text look heavier. So, we are 
effectively darkening the text twice. I don't know if this is a Fedora issue or 
a LibreOffice issue. FreeType does have a control to disable darkening, and 
that might help work around the issue if you can't get the system to do gamma 
adjustment.

2) The hints in Cantarell-Regular are wrong for many glyphs. In CFF fonts, the 
hints describe where the important stems of the glyph are located. The 
rasterizer then works to align those hints to the pixel grid. Normally, this 
would align the stems as well. In Cantarell, the hints don't always match the 
stems. For example, the top of the 'c' has a hint at (408,481) while the stem 
is actually at (423,491). On the other hand, for the 'o', which has a very 
similar shape, both hint and stem are (426,490). As a result, the 'c' shows 
overshoot while the 'o' does not. This overshoot contributes to a general 
fuzziness apparent on most lowercase letters.

I have attached a screen shot comparing the rendering you are seeing with our 
expected rendering. This was prepared using FreeType's ftview program, 15 ppem 
Cantarell v0.0.15 and gamma 1.8. It addresses issue 1). Issue 2) is improved 
because the overshoot is lighter.

Please let me know if you have any questions. Thanks.

-Dave


On 1/13/2014 11:52 AM, Michael Cronenworth wrote:
> Dave Arnold wrote:
> > Thanks for sending the screen shots. It gives me a better idea of what you are
> > seeing.
> >
> > Are you finding any problems with Nimbus Sans or Sans, or just with Cantarell?
> Cantarell is the only font that I have seen have negative effects. As shown in
> the screenshots, other fonts appear normal, as do Monospace type fonts in 
> terminals.
>
> > As Werner says, we expect rendering differences in CFF fonts for 2.5.0, so I
> > assume you are using a CFF version of Cantarell. I agree it is surprising how
> > much darker the 2.5.0 rendering of Cantarell is in your screen shot. Also, it is
> > showing overshoot at baseline and x-height. That would contribute to fuzziness.
> >
> > I'd like to understand this better, but I have not been able to locate an
> > OpenType/CFF version of Cantarell; I can find only OpenType/TrueType versions.
> > Could you send me a copy of the font you are using for your test?
> Fedora uses the font directly from upstream with no patches/features applied.
>
> http://ftp.gnome.org/pub/GNOME/sources/cantarell-fonts/0.0/cantarell-fonts-0.0.15.tar.xz
>
> Running "fc-query" on the file reports "fontformat: "CFF"(s)" so I assume the
> latest upstream font generates an OpenType/CFF version, but I am not a font 
> expert.
>
> You can find the Fedora package on the Fedora build system. If you do not have a
> system to extract the otf files I could place them somewhere.
>
> http://koji.fedoraproject.org/koji/buildinfo?buildID=467138
>
> > Thanks.
> Thank you for looking into this. There are many user reports of this issue so
> my, subjective, opinion that the Cantarell font looks "worse" is not unique.
>
> https://bugzilla.redhat.com/show_bug.cgi?id=995643
> https://bugzilla.redhat.com/show_bug.cgi?id=1035486
> https://bugzilla.redhat.com/show_bug.cgi?id=1051689

Compare.png
Description: PNG image