Re: [Freetype] FT 2.0.9 rendering for some glyphs [San-Serif fonts]

[Vadim Plessky]

Hello David, others!

On Wednesday 17 April 2002 16:33, David Turner wrote:
|  Hi Vadim,
|
|  Vadim Plessky a écrit :
|  > 1) testcases (using HTML) for Arial and TNR attached (5k all together)
|  > 2) screenshots uploaded to this location:
|  >
|  > http://htmltests.newmail.ru/test/ft209-arial.png
|  > http://htmltests.newmail.ru/test/ft209-TNR.png
|  >
|  > So, as you can see, TNR (as was expected) has much worse rendering, and
|  > text at 10px-12px sizes is almost unreadbale.
|  > Plus, number of artefacts increased (comparing to Arial)
|  >
|  > BTW: I also noticied that (in Arial) "Ð" and "Ç" are more narrow than
|  > "Î" - to my best knoledge, they should have equal widths (according to
|  > native Cyrillic reader, and general font definitions for those glyphs)
|
|  Thanks for the images. They demonstrate two kinds of weaknesses in the
| auto-hinter:
|
|    - apparently, some stems are not properly recognized and hinted,
|      which creates strange artefacts like un-aligned or disappearing
|      bars. I'll need to investigate this in more details

I tested with other fonts, and finally uploaded new image
http://htmltests.newmail.ru/test/san-serif-cyr-FT209.png
testcase:
http://htmltests.newmail.ru/test/sansserif-cyr.html

Basically, you don't need any special software/setup to reproduce problem - 
just any browser and Windows fonts (I use set from Win98SE plus MS Typography 
webfonts)

Note that problem I highlighted with Arial exists also in Verdana, Tahoma, 
Century Gothic and Arial Narrow.
Besides, Century Gothic has additional artefacts for diagonal stems - see 
letters - м М л Л д Д

After all, I tested Greek Delta character (Cyrillic Д has origin in Greek 
Delta) - and Greek alphabet was affected, too.
 
|
|    - generally speaking, serifed fonts like TNR are not processed too well.
|      the problem being that the auto-hinter detects too many potential
|      "hint" zones (due to the serifs), and try to align them all, which
|      isn't easy (especially when many of the stems overlap).
|
|  there are several ways to fix all of these, but they require quite a
|  bit of experimentation. I'd like to note that I won't start working on
|  this before 2.2 is out (which should be pretty soon I hope).

ok, I see.
Anyway, rendering of Serif fonts in Windows is so terrible that I tend not to 
use it at all. Most people I know like more Arial (or Pragmatica) than any 
version of Serif fonts (Times, TNR, Garamond, Antiqua, New Century 
Schoolbook)
Serif fonts are nice - when you get them printed :-)
but on screen they look terrible...

BTW: pls note that Linotype Palatino has the worst rendering with FT 2.0.9 
from all Serif fonts.
That's somewhat strange for me, as it had best rendering with FT 
2.0.3/Bytecode Interpreter On.

As about ways to fix: I think the only right solution for it is "to learn" 
(detect)  serif segments in outlines and bput them into (PS) subroutines.
Than you can apply one of subroutines for every character, it will save space 
(and execution time) plus imprrove rendering quality and consistency.

I don't have solution for fonts which require Flex hint, though...

|
|  > Testing environment:
|  > * Konqueror 3.0/KDE 3.0
|  > * QT 3.0.1
|  > * XFree86 4.2, anti-aliasing for 8pt-12pt font sizes in Turned On (using
|  > XftConfig)
|  > * FreeType 2.0.9 from Mandrake Cooker (I guess Bytecode Interpreter is
|  > Turned Off - at least that was the case with FT 2.0.6)
|
|  Yes, it seems so !
|
|  Thanks
|
|  - David Turner
|  - The FreeType Project
|
|  _______________________________________________
|  Freetype mailing list
|  address@hidden
|  http://www.freetype.org/mailman/listinfo/freetype

-- 

Vadim Plessky