julian67 wrote:OK, but I'm still struggling to see the problem these patches supposedly solve. As I mentioned earlier, can anyone point out or describe the legibility issue in the screenshots I posted, which are 1:1 jpgs from my screen? The Xfce config screenshot is of my actual config by the way, with custom DPI so that 1" on screen does measure 1".
They solve the problem of using subpixel rendering with medium and slight hinting. Without these packages there will be severe colour fringing. The characters, C, D and M in the screenshot you posted would all look much nicer with subpixel rendering.
julian67 wrote:I can't help noticing that the legibility in the Gnome preferences panel used in the example is unquestionably bad, having poor sharpness and contrast. Maybe it's a low res screenshot, though you'd imagine someone writing such patches and promoting them would take the trouble to make decent screenshots and actually
show the benefits. The final
screenshot he posted looks so poor that my natural reaction is to rub my eyes and clean my glasses..... maybe it's a joke? But on a real system (mine) using only the fonts and Xfce config tool from Debian main there is no such problem.
Agreed. In fact, I am not convinced that it is even subpixel rendered at all. Seems like greyscale AA to me.
julian67 wrote:I do like seeing solutions to problems but I'm wary of solutions being posed when the problem doesn't seem apparent. Maybe it's an issue with Gnome of 2006 and not relevant now?
It is still very much apparent, however only shows itself when using the slight and medium hinting options in FreeType.
julian67 wrote:In Debian the bytcode interpreter has been enabled by default since 2006.
The byte-code interpreter is evil. Now, I am not particularly bothered about software patents, however, I do care about high quality type. The byte-code interpreter encourages the use of highly aggressive font hinting. This is bad for several reasons.
Firstly, it destroys any hope of gaining resolution independence. Aggressive font hinting results in a 'drift' of around 1em for every 10 characters rendered. This means that the actual width of a rendered string will be ±1em from what the font designer intended or the result of printing it. A consequence of this? You can no longer zoom text on your screen as the locations of line breaks becomes dependant on the zoom factor.
This is why PDF documents render differently to other pieces of text. PDF files are by design resolution independent, so the use of any horizontal hinting is out of the question. It also has consequences for WYSIWYG editors, which need to either disable hinting or insert small amounts of space to counteract the drift caused by hinting.
Secondly, only TrueType files have bytecode. The other major font format, Adobe Type 1 (PFB), has virtually no hinting controls. Further, since most OpenType fonts store their outlines in Type 1 format hinting bytecode is not universally available. This means you still need to fall back to the auto-hinter for a lot of fonts.
Finally it is interesting to note that Apple — who own the (questionable) patents on TrueType bytecode — have all but abandoned it in OS X — which as far as I can tell does not make use of font hinting information but instead uses anti-aliasing, subpixel positioning and subpixel rendering.
Since many have asked what the big deal is I've attached a screenshot of my website --
http://freddie.witherden.org -- as rendered by Firefox 3.5 on a Gentoo system with the aforementioned Ubuntu patches. Serif fonts are somewhat more interesting as the results are usually more pronounced than with sans serif fonts. The DPI is 99x98, as reported by xdpyinfo.

- subpixel.png (75.79 KiB) Viewed 6802 times
Regards, Freddie.