<html> <head> <meta http-equiv="content-type" content="text/html; charset=ISO-8859-1"> </head> <body bgcolor="#FFFFFF" text="#000000"> (I am sending this to <a class="moz-txt-link-abbreviated" href="mailto:cairo@cairographics.org">cairo@cairographics.org</a> rather than <a class="moz-txt-link-abbreviated" href="mailto:cairo-bugs@cairographics.org">cairo-bugs@cairographics.org</a> because that's what the contact page told me, I hope this is correct) I recently encountered a bug in GIMP where the edge pixels of text would become greenish rather than the intended color: <img alt="incorrect antialiasing in GIMP" src="cid:part1.09000803.04080101@hotmail.com" width="152" height="156"> By coincidence, this bug doesn't affect Ubuntu and some other distributions, because it only appears when the fontconfig file explicitly defines the pixel layout of the display. Ubuntu doesn't do this, but it does enable 'lcdfilter', and apparently this causes cairo to use RGB by default. But if you explicitly define the layout as RGB or anything else (which seems to be required for Qt 5 applications to get any subpixel antialiasing at all), this bug appears. I thought this was a bug in GIMP (it doesn't disable subpixel rendering, so this problem was to be expected), but when I tried to fix it, I discovered that cairo simply ignores what the application says and forces subpixel rendering even when you explicitly disable it. This makes no sense. The documentation states: <table border="0"> <tbody> <tr> <td> <code class="literal">CAIRO_ANTIALIAS_DEFAULT</code> </td> <td>Use the default antialiasing for the subsystem and target device, since 1.0 </td> </tr> <tr> <td> <code class="literal">CAIRO_ANTIALIAS_NONE</code> </td> <td>Use a bilevel alpha mask, since 1.0 </td> </tr> <tr> <td> <code class="literal">CAIRO_ANTIALIAS_GRAY</code> </td> <td>Perform single-color antialiasing (using shades of gray for black text on a white background, for example), since 1.0 </td> </tr> <tr> <td> <code class="literal">CAIRO_ANTIALIAS_SUBPIXEL</code> </td> <td>Perform antialiasing by taking advantage of the order of subpixel elements on devices such as LCD panels, since 1.0 </td> </tr> <tr> <td> <code class="literal">CAIRO_ANTIALIAS_FAST</code> </td> <td>Hint that the backend should perform some antialiasing but prefer speed over quality, since 1.12 </td> </tr> <tr> <td> <code class="literal">CAIRO_ANTIALIAS_GOOD</code> </td> <td>The backend should balance quality against performance, since 1.12 </td> </tr> <tr> <td> <code class="literal">CAIRO_ANTIALIAS_BEST</code> </td> <td>Hint that the backend should render at the highest quality, sacrificing speed if necessary, since 1.12 </td> </tr> </tbody> </table> In my tests, CAIRO_ANTIALIAS_GRAY actually results in subpixel rendering. And ironically, <code class="literal">CAIRO_ANTIALIAS_GOOD results in no subpixel rendering. I could fix the bug in GIMP by using CAIRO_ANTIALIAS_GOOD rather than CAIRO_ANTIALIAS_GRAY, but that seems like a bad idea as it could easily break in the future, or on other platforms. Besides, the documentation also states:</code> <blockquote><code class="literal">In the case of glyph rendering, <code>CAIRO_ANTIALIAS_FAST</code> and <code>CAIRO_ANTIALIAS_GOOD</code> will be mapped to <code>CAIRO_ANTIALIAS_GRAY</code>, with <code>CAIRO_ANTALIAS_BEST</code> being equivalent to <code>CAIRO_ANTIALIAS_SUBPIXEL</code>.</code> <code class="literal"></code></blockquote> <code class="literal">This is simply wrong, my test clearly shows different results for CAIRO_ANTIALIAS_GOOD and CAIRO_ANTIALIAS_GRAY. So I would like to propose this patch: </code><code class="literal"></code> <pre>diff --git a/src/cairo-ft-font.c b/src/cairo-ft-font.c index 1ab56be..6ceec03 100644 --- a/src/cairo-ft-font.c +++ b/src/cairo-ft-font.c @@ -1768,8 +1768,7 @@ _cairo_ft_options_merge (cairo_ft_options_t *options, } if (other->base.antialias == CAIRO_ANTIALIAS_SUBPIXEL && - (options->base.antialias == CAIRO_ANTIALIAS_DEFAULT || - options->base.antialias == CAIRO_ANTIALIAS_GRAY)) { + options->base.antialias == CAIRO_ANTIALIAS_DEFAULT) { options->base.antialias = CAIRO_ANTIALIAS_SUBPIXEL; options->base.subpixel_order = other->base.subpixel_order; } </pre> <code class="literal">The effect is minimal: CAIRO_ANTIALIAS_GRAY will be respected when the fontconfig file specifies subpixel antialiasing. I've tested this patch for a few weeks on my computer, and I don't see any issues whatsoever.</code> Most applications just use CAIRO_ANTIALIAS_DEFAULT - the only situation where an application would ever specify <code class="literal">CAIRO_ANTIALIAS_GRAY is when it's actually required, so the system settings should never override this. I know that similar bugs have already been reported years ago, and apparently not everyone agrees on how _cairo_ft_options_merge() should behave. But this bug is obviously a problem, since it makes it impossible to write an application like GIMP which requires subpixel rendering to be disabled reliably. I hope this can get fixed soon, so I can fix this bug in GIMP. Maarten Baert </code> </body> </html>