[cairo] [PATCH 0/7] Bunch of distcheck fixes

[Bryce W. Harrington]

On Fri, Mar 14, 2014 at 04:00:52PM +0100, Uli Schlachter wrote:
> Hi,
> 
> On 14.03.2014 02:17, Bryce W. Harrington wrote:
> > Looks like the attachment was too large for the mailing list.
> > 
> > I've posted it here instead:
> > 
> >   # Test case changes since 1.12.8:
> >   http://www.bryceharrington.org/files/test_changes.txt
> 
> I took a look at this. This resulted in the two patches that I already send. My

493 + 8 fixed cases!  :-)

> other observations are for someone else to figure out (and I won't touch ref
> images):
>
> test-fallback: clip-fill clip-stroke-unbounded clip-twice hatchings random-clip
> rotated-clip
> bisects to 7b80613d6d483cdfd5d0c6311de0f8586092b408, ref images need to be
> updated, volunteers?
> 
> 
> PASS -> FAIL # world-map-fill.xlib-render-0_0.rgb24
> Can't reproduce, video driver bug?

Entirely possible; could also be an intermittent failure, I only ran the
suite one time.

> image16: world-map-fill paint-source-alpha paint-with-alpha record-paint-alpha
> record-paint-alpha-clip
> Bisects to 98fef3cef2d0f7f463a2e4f9f1b35b09f7b6ea77, ref images need to be
> updated, volunteers?
> 
> 
> FAIL -> XFAIL # bug-seams.image.argb32
> This change is caused by the following commit (which doesn't explain why a
> previously-good ref image is suddenly bad, Chris?)
> 
> commit 9eb1237e006bb19098144cc045c0a03d167a82b4
> Author: Chris Wilson <chris at chris-wilson.co.uk>
> Date:   Thu Jul 4 09:40:37 2013 +0100
> 
>     test: Add a few reference images found lurking on my machine
> 
> 
> 
> PASS -> FAIL # record-fill-alpha.image16.rgb24
> Bisects to 8cfbdf2f02ba01d5638a91c9f3f7fc228b402caa, dunno, I'd propose a new
> ref image
> 
> 
> PASS -> FAIL # record-paint-alpha-clip-mask.image16.rgb24
> Bisects to c986a7310bb06582b7d8a566d5f007ba4e5e75bf, dunno, I'd propose a new
> ref image
> 
> 
> PASS -> FAIL # device-offset-scale.image.argb32
> PASS -> FAIL # large-source-roi.image.argb32
> PASS -> FAIL # surface-pattern-big-scale-down.image.argb32
> PASS -> FAIL # surface-pattern-scale-down-extend-none.image.argb32
> PASS -> FAIL # surface-pattern-scale-down-extend-pad.image.argb32
> PASS -> FAIL # surface-pattern-scale-down-extend-reflect.image.argb32
> PASS -> FAIL # surface-pattern-scale-down-extend-repeat.image.argb32
> PASS -> FAIL # surface-pattern-scale-down.image.argb32
> Bisects to fb57ea13e04d82866cbc8e86c83261148bb3e231, definitely a regression for
> large-source-roi, the others might need new ref images

I plan to re-run the suite with KK's new patch applied this evening.

> > On Fri, Mar 14, 2014 at 01:04:13AM +0000, Bryce W. Harrington wrote:
> >> I've gone ahead and pushed patches 1, 2, 3, 6, and 7 from this series to
> >> master.
> >>
> >> Patch 4 I modified as suggested by Uri and pushed, to indicate skia has
> >> been in the codebase since 1.10 rather than 1.12.
> >>
> >> Patch 5 I dropped and used Uri's alternate fix instead, after checking
> >> that it resolves the gtk-doc issues.
> 
> No idea who this Uri is, but thanks!

Uli!  Sorry.  You have permission to refer to me as Bruce from now on.

> >> As mentioned below, there is still some minor cruft in distcheck, but
> >> nothing IMHO that is worth holding up the release for.  However, tests
> >> are in a poor shape right now.
> >>
> >> I ran 'make check' comparisons between the 1.12.8 release (as tagged in
> >> git), and master HEAD from earlier today:
> >>
> >>                    1.12.8      HEAD
> >>  ----------------------------------------------------------------------
> >>  Tests run:        22751       24838
> >>  Passed:           15809       15881
> >>  Failed:           4198        5385 
> >>  Expected Failed:  405         330      
> >>  Error:            1           1        
> >>  Crashed:          227         1018 
> >>  Untested:         2111        2223 
> >>  Total:            22751       24838
> >>
> >> I've attached a listing of the test cases that changed their results.
> >> Failures seem to be rather across the board, although there does seem to
> >> be a lot relating to clipping/masking and transform/scaling.
> [...]
> 
> How did you generate the (not-)attached list? I'm curious...

I did up a couple simple scripts.  One greps the results out of the
*.log files and creates a results.txt, the other parses two results.txt
files into dicts and identifies test cases that were present in both
runs and had differing results.  If you're interested I can stick them
in test/.

Bryce