[cairo] broken pixman bilinear filtering?

[Bill Spitzak]

Keith Packard wrote:
> On Mon, 2006-02-13 at 13:53 -0800, Vladimir Vukicevic wrote:
> 
> 
>>In this particular case, one option is to keep the old code in, and
>>rename the bilinear filter #define to BILINEAR_BROKEN_DEPRECATED or
>>something, and add a new identifier for the "correct" bilinear
>>processing.  But again, I don't know what the situation is with
>>XRender development in the server, and what the potential would be to
>>get this fix out in a release in the short term.
> 
> 
> I'd like to fix the X server reasonably soon (7.1?). I was planning on
> incrementing the minor version number in the Render protocol to mark
> this semantic change so that clients could tell the difference. Of
> course, this presumes that clients would be updated reasonably rapidly
> after the Render changes. I'm not sure how we'd deal with old clients on
> a new server; leaving the broken semantics in place with the 'right'
> name seems like a long-term maintenence problem.

Sounds good to me. It would be best to fix the server entirely, as this 
would fix any clients that did not spend the time figuring out how to 
work around this. I plan to update my code, possibly immediately on the 
assumption that this will be fixed in the next version, and I think this 
would be true of anybody else who went through the trouble of finding 
the workaround.

I'm having some problems figuring out the version. I assumme Xft 
initializes XRender, and have had no trouble calling XRender or 
detecting if it exists and thus running on servers with no XRender. But 
I have been using XRenderFindStandardFormat(xdisplay, 
PictStandardARGB32) returning non-zero to detect if it is working as I 
have had no luck with XRenderQueryExtension or QueryVersion. On the 
machines I have here XRenderQueryVersion returns a non-zero status (1) 
and sets the major/minor to 0.8. Is this correct? And what will the 
fixed version do (it does not seem like this has changed so much it has 
gone from 0.8 to 7.0!)