[cairo] win32 build

[Vladimir Vukicevic]

On 6/25/05, Owen Taylor <otaylor at redhat.com> wrote:
> On Sat, 2005-06-25 at 09:42 -0700, Keith Packard wrote:
> > On Sat, 2005-06-25 at 12:12 -0400, Owen Taylor wrote:
> >
> > > I'm not really sure we want backend-specific reference images for
> > > the full set of tests we have now. That's a lot of images.
> >
> > Yes, I think this is exactly what we want.  A new backend should come
> > with a set of reference images for it.  Given that we have intentionally
> > departed from pixel-exact rendering, I think this is the best we can
> > manage.
> 
> This makes the process of adding new regression tests pretty
> annoying for everyone, which isn't really what we want.

I don't think it has to -- I think what Keith is suggesting is that
the standard reference images generated by the image surface/pixman
backend should be the default to compare against, but any backend
should be able to override those results to say "for this backend, the
correct result will look like this instead".  cairo_test might want to
read an input file describing which tests to run on which backends,
and what output file should be used if not the default.

> Based on all experience with GLib, people really like 'make check'
> to pass and file bugs liberally when it doesn't... having make check
> fail until people do some sort of manual verification and replacement
> doesn't sound like a good idea to me.

I'd think a new regression test that generates output on some surfaces
that's different than what the pixman backend generates *needs* to be
manually inspected and either OK'd (by adding a new reference image
for that backend) or have the regression-causing patch be flagged as
potentially buggy.

> I'm already concerned about the Xlib failures on most servers.

Not sure what the specific issue is, but if the xlib surface doesn't
have an explicit workaround for a bug/failure in Xlib or a particular
X server, then I think that test should fail, no?

     - Vlad