Sat Apr 2 12:03:17 PDT 2011

Uli Schlachter <psychon at znc.in> writes:

>>> Both the bugzilla bug report and the report on the mailing list are against
>>> cairo 1.10, so I guess that 1.10 also has this bug. The fix is also quite small
>>> and relatively low-risk (IMHO) since it just skips a fast-path.
>>>
>>> However, I have no clue what the procedure here is. When you feel like something
>>> is worth it, you just push it to the 1.10 branch? Is there anyone who has to be
>>> asked first? Would the change to the test suite be pushed to the 1.10 branch, too?
>> 
>> For a case like this where you have a clear bug fix that impacts users
>> and is low risk, then please do commit to 1.10 branch. And include the
>> test case where convenient. (i.e. unless doing so will cause massive
>> breakage to 1.10).
>
> Ok, thanks for the info.
> I have cherry-picked the bug fix to the 1.10 branch, but I have skipped the
> changes to the "operator-source" test since I still don't know if the new ref
> images are correct. If this change to the test should get in AND the test
> actually works for someone (I have the wrong fonts), then I guess this change
> could get cherry-picked to 1.10, too.

The patch:

+    /* The below code breaks for operator source. This can best be seen with
+     * multiple boxes where black is drawn to dst outside of the boxes. */
+    if (op == CAIRO_OPERATOR_SOURCE)
+       return CAIRO_INT_STATUS_UNSUPPORTED;

looks a little suspicious to me: Should it really check for SOURCE, or
should it check for bounded operators that are unbounded in pixman?
SOURCE is one such operator, but there are others.

Soren