[Devel] Re: my experience with 2.1.4

[bulia byak]

>   Though I'm still way too overworked, Werner asked me to
>   answer to this specific thread :-)

Thanks for the informative answer. My points below.

> We have already integrated some of David's improvements into the font =
> engine,
> and we're ready to add more of them, as long as we have the time to test =
> them
> completely.

David's patched versions are in use for more than a year, and 
I've been using them most of that time. I can attest that I've 
consistently seen good rendering quality from that software for 
any fonts I used it with, including many Cyrillic fonts. 

> This generally involves checking that the new code doesn't to stupid =
> things
> with regards to performance or internal consistency, as well as inspect =
> the
> *global* result of the improvement. That's because it's fairly easy to =
> come
> up with an improvement that will improve the rendering of some fonts, =
> while
> ruining that of others. 

Truth is, good free fonts for Linux are few and can be tested 
quickly. We don't want to spend our time trying to make crap 
fonts look good. Though I must say that almost any font looks 
reasonably legible and evenly kerned with David's rendering.

> Moreover, we need to check that all character
> sizes are consistently improved (an improvement at size 16 against a
> degradation at size 12 is difficult to balance).

As I see it, the problem is mostly with very small sizes. My 
impression (I may be wrong) is that you (freetype developers) 
are trying to improve the crispness of very small font sizes 
at the expense of disregarding major flaws with larger sizes 
(such as bad kerning). But screen resolutions and pixel font 
sizes grow all the time, so you should "develop for the future 
(for it will be here sooner than you think)" and pay more 
attention to larger sizes. I for one would much prefer evenly 
kerned and smooth average-sized font at the expense of a bit 
of a blur in small sizes. After all, those preferring small 
font sizes can turn off AA altogether.

> In short, it's not that we're not willing to accept improvements, it's =
> that
> it takes a lot more time to validate them than what you might expect. A
> screenshot doesn't tell you the whole story.

What about a series of screenshots showing common fonts in 
common sizes?

> > Your comparison clearly shows that in the *_LIGHT mode, kerning is=20
> > not only bad but *intolerably* bad.=20
> >=20
> Boy, you're using strong words here :-) 

Sorry. I'm not going personal, I'm speaking about rendering 
examples that I see. Please do not take it personally.

> Personally, while I do agree that the kerning of LIGHT is worse than =
> SMOOTH,
> I also find the stem width imbalance of SMOOTH a lot more annoying on =
> the
> PC I'm using here at work. I also know from experience that my feeling =
> will
> be slightly different if I switch to a different screen (e.g. better =
> quality,
> or LCD instead of CRT, etc...)

Do you agree that the kerning problem with LIGHT is real, 
while the stem width problem is subjective and/or depending 
on display? Your paragraph above seems to suggest that you do. 
But what directly follows from this is that we must fix the 
kerning _first_ and worry about stem widths _later_.

OK, I've got an idea. With so many people claiming that it's 
"subjective," let's use the proper way of resolving subjective 
issues, which is an opinion poll. Let's prepare a 
representative series of screenshots for LIGHT and SMOOTH 
showing common fonts in common sizes and let people vote for 
them on the freetype site. A link from a high traffic linux 
site (slashdot) would be helpful too. If one of the renderings 
gets a large enough margin, we'll consider this to be the 
final decision on what to make the default mode in the next 
release of freetype. What do you think?

Another idea (I'm not sure if it's doable) is to make the 
kerning strategy selectable at runtime, with some sort of a 
config file. With that option, at least I won't have to 
squirm when I have another Linux newbie talking about "crap
fonts" - I'll be able to give them simple instructions on how
to fix that. Right now, I cannot suggest that they "patch and
recompile" because that's too scary for newbies.

>   - disable hinting completely. No more funky kerning and stem widths =
> :-)
>     that's what Apple does in OS X and nobody seems to complain

An interesting idea that I'd like to try. Can anyone please 
tell me how to do that? I tried to find a magic switch to turn 
off autohinter in 2.1.4 sources but didn't succeed (I'm not a 
programmer).
 
>   - install a patched version of FreeType on your system.

Doing that now. Unfortunately it's not an option for most users 
and distro providers, which results in Linux having such a bad 
reputation for font rendering.

>   - try to find other fonts that FreeType 2.1.4 hints better.

Not quite practical. Good fonts for desktop are rare, even if
you look for non-free ones.

> Sorry, but we're not going to rush for this right now. I lack time to =
> review
> David's patch in details. Moreover, there are other interesting inputs =
> from
> Japan that I'm trying to integrate as well, and I doubt the two of them =
> will
> match 100%. Which means finding some ways in between while performing =
> always
> more tests..

I'm not trying to rush you into accepting untested patches. I'm 
just trying to suggest that David's SMOOTH patch is pretty well 
tested already, and that the kerning problem with LIGHT is 
("subjectively perceived by some as") way more urgent than you 
seem to imply.

-- bb



-- 
__________________________________________________________
Sign-up for your own FREE Personalized E-mail at Mail.com
http://www.mail.com/?sr=signup