my recent testcase

[Benjamin Otte]

Hey,

I just added a testcase in 84bbf179c375622d2c7b4e21b1b8ce189b5a18f2
that exposes some issues. I fixed the important one, but a few things
remain that I don't know how to solve:

1) disabling the test on vector backends
As the test is targetted at xlib in particular and causes aliasing
artifacts, I want to disable it on backends where it's not necessary
to run it. I used CAIRO_ANTIALIAS_NONE to get rid of the artefacts,
but the vector backends can't honor that flag. Is there a way to
disable the test there?

2) bug on intel
I noted a bug on intel in the commit message and was about to post
about it here. But then I decided to install the newest X and driver
from git and the bug is gone now. So you can ignore that in the commit
message.

3) CAIRO_FIXED_16_16_MIN/MAX
cairo-xlib-surface.c defines these two values to -32768 / 32767 and
then compares them to cairo_fixed_t without conversion in
_line_exceeds_16_16. I'm 90% sure this is wrong, but I'd like Chris
(or someone else) to say something like "yes, you need to multiply it
with CAIRO_FIXED_ONE there" before going to fix that.
So far it just seems to cause way a lot of unnecessary projections
occuring (one of which triggered the original bug I fixed).

Cheers,
Benjamin
_______________________________________________
cairo mailing list
cairo@...
http://lists.cairographics.org/mailman/listinfo/cairo

[Chris Wilson-11]

Excerpts from Benjamin Otte's message of Wed Nov 04 11:21:26 +0000 2009:
> 1) disabling the test on vector backends
> As the test is targetted at xlib in particular and causes aliasing
> artifacts, I want to disable it on backends where it's not necessary
> to run it. I used CAIRO_ANTIALIAS_NONE to get rid of the artefacts,
> but the vector backends can't honor that flag. Is there a way to
> disable the test there?

Add "raster" to the requirements line in CAIRO_TEST()

> 3) CAIRO_FIXED_16_16_MIN/MAX
> cairo-xlib-surface.c defines these two values to -32768 / 32767 and
> then compares them to cairo_fixed_t without conversion in
> _line_exceeds_16_16. I'm 90% sure this is wrong, but I'd like Chris
> (or someone else) to say something like "yes, you need to multiply it
> with CAIRO_FIXED_ONE there" before going to fix that.
> So far it just seems to cause way a lot of unnecessary projections
> occuring (one of which triggered the original bug I fixed).

Indeed I do have a patch that fixes it as
_cairo_fixed_from_int (-32768), along with tens of thousands of other
lines. Gah.
-ickle
--
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
cairo mailing list
cairo@...
http://lists.cairographics.org/mailman/listinfo/cairo

[Benjamin Otte]

On Wed, Nov 4, 2009 at 1:43 PM, Chris Wilson <chris@...> wrote:
> Add "raster" to the requirements line in CAIRO_TEST()
>
Just for the record, it was "target=raster".

> Indeed I do have a patch that fixes it as
> _cairo_fixed_from_int (-32768), along with tens of thousands of other
> lines. Gah.
>
So I just pushed this fix. Have fun merging. :)

Excerpts from Benjamin Otte's message of Wed Nov 04 11:21:26 +0000 2009:
> 2) bug on intel
> I noted a bug on intel in the commit message and was about to post
> about it here. But then I decided to install the newest X and driver
> from git and the bug is gone now. So you can ignore that in the commit
> message.
>
The bug seems to be back. I have no idea why it wasn't happening
directly after I restarted X, but now the problem is back.
test/cairo-test-suite xcomposite-projection fails for me with seams
looking like off-by-one errors.
It'd be nice if someone could confirm it and have a look at it.

Benjamin
_______________________________________________
cairo mailing list
cairo@...
http://lists.cairographics.org/mailman/listinfo/cairo