Skip to content

Conversation

@raytranuk
Copy link
Owner

No description provided.

shrekshao and others added 30 commits May 18, 2020 15:31
Change WebGLTestUtils.awaitOrTimeout to offer a default timeout of
5000 ms, and all tests currently using it to use the default.

Tested both manually and in Chromium's automated WebGL test harness.
Unclear whether this will fix the problems seen on Chromium's bots,
but need to try.

Addresses #3076.
Replace drawQuad with clearAndDrawUnitQuad.
…premultiplyAlpha=false (#3089)

* Add tests to verify toDataURL images are the correct orientation

Adding a test to catch bug: https://bugs.webkit.org/show_bug.cgi?id=156129

* Remove trailing whitespace

* PR feedback
Disable the use of this new API for the image_bitmap_from_video
conformance tests to work around crbug.com/1090960. It will be
re-enabled once the underlying bug is fixed.

Follow-on to #3065.
The earliest version of this code, which was seemingly stable, made the
determination of whether to use a local or remote URL cor cross-origin
images up front. Network timeouts from machines that deliberately
aren't connected to the open Internet can't be relied upon as a fallback
mechanism. Change the code back to examine the test's URL and choose
whether to load local or remote images based on that. Also attempt to
make runningOnLocalhost more robust to IPv6 loopback addresses.

Follow-on to #3052 and #3077. See http://crbug.com/1082525 for more
background.

Should fix #3076 conclusively.

Note that #3097 was proposed as an alternative solution for this problem
and may still be considered in the future if this change destabilizes some
browsers' CI systems.
* Implicitly enable ANGLE_instanced_arrays and WEBGL_multi_draw when needed.
#3086)

* Use ResizeObserver's device-pixel-content-box when available for high-dpi demo.
Restructure the tests to use the Fetch API and async/await (thanks
to suggestion from @jdashg) to make this addition easier to read.

Make tolerance configurable, and use a large value for compatibility
with Firefox Nightly on macOS.

The new JPEG resource was exported from the included Photoshop
document which was constructed from scratch for this test case, based
on the original report from Figma.

Regression test for http://crbug.com/1058160 .
to weed out implementions that incorrectly report precision
kenrussell and others added 21 commits July 17, 2020 12:31
All of the existing tests for this pixel pack parameter with a value
greater than 0 were assumed to fail validation. Adding a positive test
reveals an apparent OpenGL driver bug on macOS.

Test case for http://crbug.com/angleproject/4849 .

This new test passes in Chromium and Firefox on macOS, and in WebKit
on top of ANGLE with the necessary driver bug workaround.
#3119)

According to the spec:
https://www.khronos.org/registry/webgl/extensions/WEBGL_compressed_texture_etc1/

This WebGL extension does not allow using compressedTexSubImage2D:
"Remove ability to use the format with compressedTexSubImage2D,
 as per the GLES extension spec."

The compressed-tex-image test was modified to reflect the spec.
The 2.0.0 snapshot of this test was incorrectly creating a WebGL 1.0
context.
Since the required error was changed from INVALID_ENUM to
INVALID_OPERATION in #3099, the 2.0.0 snapshot must allow both values
in order to admit both old and new implementations.
…3127)

Incorporate fix from #2327, which never made it into the 2.0.0
snapshot. This test was substantially rewritten on top-of-tree in #2806.
* Shader source character set and preprocess validation test

* fix style

* add invalid test
These tests were invalidated in #2386. In order to allow both old and
new implementations to pass these tests, and to minimize complexity in
the 2.0.0 conformance snapshot, simply remove the now-invalid tests.
Backport fixes from #2330 and #2612 to the 2.0.0 snapshot.
Add "new functions" section to WEBGL_multi_draw.

Add needed parentheses to union types in the Web IDL in
WEBGL_multi_draw_instanced_base_vertex_base_instance.
…3134)

When ImageBitmapOptions is not supported by the browser, the implicit
behavior of the premultiplyAlpha option is the default. Per HTML
specification:

https://html.spec.whatwg.org/multipage/imagebitmap-and-animations.html#cropped-to-the-source-rectangle-with-formatting

this is implementation-specific (either premultiplied or not), so when
the input ImageData has alpha != 1.0, there isn't a single correct
answer. Skip these tests.

This was discovered while implementing ImageBitmap from ImageData
support in WebKit in https://bugs.webkit.org/show_bug.cgi?id=183438 ,
though WebKit supports ImageBitmapOptions. Firefox currently doesn't
support ImageBitmapOptions and happens to have been passing these
tests, but WebKit wouldn't have.
INT, UNSIGNED_INT, HALF_FLOAT, INT_2_10_10_10_REV, and
UNSIGNED_INT_2_10_10_10_REV
Revise conformance2/extensions/ext-texture-norm16.html to mandate that
16-bit signed normalized textures are not color renderable. That
functionality is covered under EXT_render_snorm, not currently exposed
to WebGL.

Partially addresses #3114.
This allows the expando-loss.html and expando-loss-2.html tests to be
run correctly in the MiniBrowser with the following two environment
variables set:
  export JSC_useDollarVM=1
  export __XPC_JSC_useDollarVM=1

Thanks to Yusuke Suzuki from Apple for this tip.
@raytranuk raytranuk merged commit 9d9e2f8 into raytranuk:master Aug 28, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.