Bug 29618 – Text isn't fully rendered with certain fonts in Win32/Cairo

• Status: RESOLVED
• Resolution: INVALID
• Component: Text

[Marshall Culpepper]

Created attachment 39882 [details]
Screenshot

In some cases, at the end of a line of text, it looks like a few pixels are cut off from the rendered text. This appears in both inline text and buttons as far as I've seen. I've attached a sample HTML file which reproduces this behavior as of r48572, as well as a screenshot of what it looks like in both Windows 7 and Windows XP SP2.

[Marshall Culpepper]

Created attachment 39883 [details]
Testcase HTML

[Marshall Culpepper]

Created attachment 39884 [details]
Screenshot

[Martin Robinson]

Created attachment 45338 [details]
Disable subpixel antialiasing

The rendering glitches are due to limitations in Cairo (or GDI) with rendering ClearType-quality (subpixel) anti-aliased fonts onto a non-DDB surface. The WebKit in Windows currently uses a DIBSection for the backing store.

[WebKit Review Bot]

style-queue ran check-webkit-style on attachment 45338 [details] without any errors.

[Eric Seidel (no email)]

OK.

[Eric Seidel (no email)]

Comment on attachment 45338 [details]
Disable subpixel antialiasing

Does the WinCairo build have pixel tests?  If so, they'll have to be updated as a result of this change.

[Brent Fulgham]

Note that a very similar issue was discussed on the Cairo mailing list over three years ago.  See http://lists.freedesktop.org/archives/cairo/2006-October/008081.html.  While the referenced images seem to be lost in time, the commentary is illuminating:

 "note how the "1" digits that occur as the final character of a string have been partially cut off)"

At the time, the resolution seemed to be to force use of the 24-bit backend for certain operations:

  "If I changed cairo_win32_surface_create() to use RGB24 also when BITSPIXEL is 32, the problem went away, as _cairo_win32_scaled_font_show_glyphs() then uses the "direct" branch. But this seemed like using a awfully large hammer to fix the problem."

[Eric Seidel (no email)]

Comment on attachment 45338 [details]
Disable subpixel antialiasing

I'm confused by bfulgam's recent comment.  Removing cq+ for now.

[Brent Fulgham]

Thanks, Eric.

I would prefer to hold off committing this change (even though I twisted Martin's arm into posting his patch), because what it does is remove the sub-pixel anti-aliasing resulting in very blocky text.  I'd prefer to review this issue with the Cairo team to see if a 'correct' fix could be identified that would allow us to continue seeing ClearType rendered text, but resolve the trailing pixel problem Martin and Marshall identified.

[Brent Fulgham]

BTW, this issue is definitely confirmed.  :-)

[Brent Fulgham]

Created attachment 45417 [details]
Patch to Cairo to correct the win32 rendering problem

Note: This is not a WebKit patch, but is shown here to help other implementors.

[Brent Fulgham]

Created attachment 45418 [details]
Comparison of Cairo rendered glyphs with (right side) and without (left side) the cairo-win32-font patch.

[Brent Fulgham]

I believe removing the sub-pixel antialiasing is ill-advised.  The true problem is in the underlying Cairo library.  Making the recommended change (which is already in use on FireFox) will resolve the problem properly.