[cairo] Write ups on Microsoft Avalon
Bill Spitzak
spitzak at d2.com
Mon May 17 10:02:14 PDT 2004
On Sunday 16 May 2004 08:19 pm, Peter Dennis Bartok wrote:
> There are a few situations where the ability to 'turn off anti-aliasing'
> is required. For example, when drawing fonts smaller than a certain size,
> in order to keep it actually readable and not 'smudged' it is important.
> I'm working on Mono (SWF) and one of the sample apps I use brings this
> difference out quite strong, when I compare the output of MS with the
> output of Mono through Cairo, the Mono version is much harder to read with
> a 6pt font.
>
> Also, for situations like drawing a Hotkey marker or an underline for
> text, anti-aliasing is really defeating the purpose, it simply will look
> off if you have a blurry line underlining the hotkey character or the text
> line.
Neither of these are good examples of "need to turn off antialiasing". Both
of them may indicate that the rasterization is not producing the most
beautiful pictures possible, but that means the rasterization should be
fixed, not that we need an option to change the algorithim. Such options lock
the algorithim, for instance on Windows it prevents use of the lcd rgb
shifting even though that can be done in a "non-antialiased" way, due to the
fact that programs assumme other effects of turning off antialiasing.
The font renderer certainly can change how it renders the glyphs to make them
as nice as possible. This does not mean "turn off antialiasing", what it
means is to shift and adjust lines in the letters so they are 1 pixel wide
and aligned to 1 pixel vertically, and 1 or 1/3 pixel horizontally. Curves in
the glyphs can and should still be antialiased.
Character widths are a much bigger problem. I am pretty convinced that we
cannot have character widths depend on the current transform and the font
renderer. If the renderer produces glyphs renderered at several different
fractional positions the output looks quite good despite this, the only
problem is when a character or pattern is repeated, for instance "iiiiiiii"
will look irregular even if the antialiasing was perfect. The only solution I
can see for this is for Cairo's low-level renderer to accept strings of
glyphs and for it to be allowed to adjust the positions of the internal ones
so the string is about the same width but more regular looking (somebody else
here posted some documentation on these algorithims). This may not sit very
well with the Pango people, however.
Line widths could be solved by special-casing '0' line width to mean the
nearest multiple of pixels to 1/96 inch, and that the line centers are
adjusted to be in the middle of pixels. This would be the default line width.
The reason for this special case is to discourage the many programmers who
try to "fix" it by adjusting the CTM so that the integers are at the centers
of the pixels rather than the corners, which makes predictable results for
integer rectangles impossible (this is a personal peeve of mine, the NeWS
guys completely broke the renderer in an attempt to "fix" integer rectangles,
because they thought the default had to center integers in the pixels so that
1-pixel lines would draw right).
The only real reason for no-anti-aliasing is so that you can predicatably
"erase" an image by drawing it again in the background color. This can either
be ignored, or handled by a new compositing mode that replaces the alpha a
with (a>0)?1:0
--
,~,~,~,~ ~ ~ ~ ~
/\_ _|_========___ Bill Spitzak
~~~/\/\\~~~~~~\____________/~~~~~~~~ spitzak at d2.com
More information about the cairo
mailing list