Re: [ft-devel] merged CFF and Type1 hinting code now in master

[Nikolaus Waxweiler]

> I dunno. Is LIGHT autohinter that bad?

No, it is perfectly fine. It's just that the CFF renderer does not
have to guess, which gives a slight hinting quality boost in some
situations. Plus, I still have to get around to implement actually
good stem darkening for the autohinter. I have no opinion on MONO.

> Does it correspond to fontconfig with light, medium, and strong?

No. As I was told by a GNOME developer, the fontconfig setting
originally attempted to let the user chose the hinting strength, going
from no hinting to v35-like strength (monochrome vs. grayscale vs.
subpixel rendering is a different setting). The problem with this
scheme is that it doesn't really work (and never has) with two
opposing hinting schools of thought (bytecode vs. Type1 hints) and how
the user has no say in what hinting strategy a TrueType font uses.
Sure, you can implement x-axis-hinting in the CFF renderer, but that
doesn't solve the rendering consistency problem with TTFs. v35 only
really works if all you look at is Microsoft's core web fonts.

Microsoft doesn't even try to solve this and just sets a default for
their various GUI toolkits that gets applied to every font passing
through it (witness the ugliness of Arial next to some random
ttfautohinted web font). Apple ignores the issue entirely.
Fontconfig-using toolkits may ignore it, too. GTK-using applications
ignore what fontconfig says and just give the user the option of
slight/medium/full across all font formats.

I care about making text displayed on the Linux desktop look
harmonious and consistent, no matter the source format. For the above
reasons, I think the entire light/normal and slight/medium/full
differentiation will just never work well without manual intervention.
I practiced manual intervention for years before trying to change FT
(e.g. with the Infinality patch set) and it just... didn't deliver.

My proposal is to let it go and chose light as the default and have
toolkit makers implement linear alpha blending and gamma correction
instead. It looks good (also on HiDPI-screens!) and just works for the
vast majority of fonts out there.