Bug 55745 – [chromium] MAC: update expectations for CSS3 linear gradient tests

• Component: CSS
• Status: RESOLVED FIXED

[noel gordon]

After the CG changes on bug 28152, the chromium MAC port should pass these tests. However, the do not pass currently, not sure why.

fast/gradients/css3-color-stop-units.html
fast/gradients/css3-color-stops.html
fast/gradients/css3-linear-angle-gradients.html
fast/gradients/css3-radial-gradients.html
fast/gradients/css3-radial-gradients2.html
fast/gradients/css3-radial-gradients3.html
fast/gradients/css3-repeating-radial-gradients.html

[James Robinson]

fast/gradients/css3-color-stops.html looks like 32bit vs 64bit CoreGraphics differences (Chrome always compiles in 32 bit currently whereas Safari is a 64 bit app on Snow Leopard).

[noel gordon]

Maybe fast/gradients/css3-color-stop-units.html fast/gradients/css3-linear-angle-gradients.html could go in the same bucket (32bit vs 64bit).

The radial gradient tests are broken though.  Chrome-mac shows "circles", other platforms show "ellipsoids".  For example ...

http://test-results.appspot.com/dashboards/flakiness_dashboard.html#tests=fast%2Fgradients%2Fcss3-radial-gradients3.html&showExpectations=true&group=%40ToT%20-%20chromium.org

[noel gordon]

Got a Snow Leopard box setup, and compiled chromium mac, and see the 32bit vs 64bit compile issue.

Gradient.h includes <config.h>
   which includes -> <wtf/Platform.h>
     which includes -> /Developer/SDKs/MacOSX10.5.sdk/usr/include/AvailabilityMacros.h
 
Snow Leopard is not defined in the 10.5 SDK AvailabilityMacros.h, so the (32-bit) chromium-mac compile
thinks it is BUILDING_ON_LEOPARD via <wtf/Platform.h> :o

The makes Gradient.h define:
  #define USE_CG_SHADING defined(BUILDING_ON_TIGER) || defined(BUILDING_ON_LEOPARD)

and makes the gradient paint code take the USE_CG_SHADING (wrong) path in
  http://trac.webkit.org/changeset/74915/trunk/WebCore/platform/graphics/cg/GradientCG.cpp

Circular gradients result, rather than ellipsoids. Evil bug.

[noel gordon]

Created attachment 85779 [details]
Proposed patch

A workaround.  Rebaselines will come later if we go with this patch.

[James Robinson]

But we do build on Leopard - does this patch work there?

[noel gordon]

Fair call, chrome try jobs sent ...
  http://build.chromium.org/p/tryserver.chromium/builders/mac_layout_rel/builds/119
  http://build.chromium.org/p/tryserver.chromium/builders/mac_layout/builds/463

[Mihai Parparita]

(In reply to comment #6)
> Fair call, chrome try jobs sent ...
>   http://build.chromium.org/p/tryserver.chromium/builders/mac_layout_rel/builds/119
>   http://build.chromium.org/p/tryserver.chromium/builders/mac_layout/builds/463

I'm pretty sure that Mac try servers run on Snow Leopard. From the output:
    setenv DISTCC_COMPILER "gcc version 4.2.1 (Apple Inc. build 5659)"
    setenv DISTCC_HOSTS localhost
    setenv DISTCC_SYSTEM "10.6.6 (10J567, i386)"

[noel gordon]

Thanks Mihai, I'm searching round for leopard machine now ...

[noel gordon]

Found a 10.5.8 machine, built chromium dumpRenderTree, and the repeating linear gradient tests are well broken ...

[noel gordon]

Created attachment 85919 [details]
repeating linear gradient test failure

[noel gordon]

Best to fix this once there's a chrome 10.6 build/release.

[noel gordon]

To salvage something from this bug, noted the following tests are passing on
dumpRenderTree --chromium on both OSX 10.5, OSX 10.6 chrome builds.
 
fast/gradients/css3-color-stop-units.html
fast/gradients/css3-color-stops.html
fast/gradients/css3-linear-angle-gradients.html

Change the bug title to match.

[noel gordon]

Filed bug 56736 about fixing the radial gradient cases.

[noel gordon]

Created attachment 86306 [details]
Patch

[Mihai Parparita]

Comment on attachment 86306 [details]
Patch

I'm not seeing any new or updated images in the patch (and it's only 1.96K). Also note that we're currently passing these tests on Snow Leopard, so we want to keep using the mac/ baseline there. These baselines should go into chromium-mac-leopard (or better yet, mac-leopard if you can verify that the mac port would pass with them on Leopard).

[Mihai Parparita]

(In reply to comment #12)
> To salvage something from this bug, noted the following tests are passing on
> dumpRenderTree --chromium on both OSX 10.5, OSX 10.6 chrome builds.

Oh, I see now, you're saying that these tests pass with the current baselines. However, I'm not seeing that on the dashboard:

http://test-results.appspot.com/dashboards/flakiness_dashboard.html#tests=fast%2Fgradients%2Fcss3-color-stop-units.html%2Cfast%2Fgradients%2Fcss3-color-stops.html%2Cfast%2Fgradients%2Fcss3-linear-angle-gradients.html&group=%40ToT%20-%20chromium.org

Where did you see them pass on 10.5?

[noel gordon]

Tests look fine to me on the on the dashboard, closing works for me.

http://test-results.appspot.com/dashboards/flakiness_dashboard.html#showExpectations=true&tests=fast%2Fgradients%2Fcss3-color-stop-units.html%2Cfast%2Fgradients%2Fcss3-color-stops.html%2Cfast%2Fgradients%2Fcss3-linear-angle-gradients.html

[noel gordon]

Oh, now we have skia on the mac, some rebaselines are in order ...

[noel gordon]

Created attachment 120283 [details]
Patch

[noel gordon]

Created attachment 121050 [details]
Patch for landing

[Stephen White]

Comment on attachment 121050 [details]
Patch for landing

OK.  r=me

[WebKit Review Bot]

Comment on attachment 121050 [details]
Patch for landing

Clearing flags on attachment: 121050

Committed r104128: <http://trac.webkit.org/changeset/104128>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.