Running the following command with Nushell (`nu`) on
[`gpuweb/cts`:0ba03b43]:
[`gpuweb/cts`:0ba03b43]: 0ba03b439c/src/webgpu/listing_meta.json
```
open src\webgpu\listing_meta.json
| reject _end _comment
| transpose
| rename test data
| update data {
$in.subcaseMS
| into float
| into int
| into duration --unit ms
}
| rename --column { data: subcaseMS }
| sort-by subcaseMS
| insert violates_the_law { $in.subcaseMS > 30sec }
| filter { $in.violates_the_law }
| reject violates_the_law
| last
```
…which results in this output:
```
╭───────────┬───────────────────────────────────────────────────────────────╮
│ test │ webgpu:shader,validation,expression,call,builtin,max:values:* │
│ subcaseMS │ 4min 20sec 241ms │
╰───────────┴───────────────────────────────────────────────────────────────╯
```
…we can determine that Chromium's test runs of WebGPU CTS expect their
longest-running test to take approximately 4.3 minutes. Rounding up,
this approximately 5 times the base long timeout of a WPT test for
Firefox, by default. So, apply a 5x test timeout to Firefox's WebGPU CTS
tests, in hopes that we can avoid the massive number of `TIMEOUT`s we've
historically run into.
Differential Revision: https://phabricator.services.mozilla.com/D216439
This patch adds a small delay to the URL navigation in talos tests to resolve some permafailure timeouts on windows 11. At the same time some test timeouts, and max runtimes are adjusted to handle the new delay.
Differential Revision: https://phabricator.services.mozilla.com/D216388
Reruns of mochitest-plain tasks on Windows for older commits also fail. The
issue is located in the generated conditioned profile.
Differential Revision: https://phabricator.services.mozilla.com/D215575
Currently cstm-car-m tests for sp3 are tier 3, so failures are not
triaged properly. This patch bumps up the tier for this to 2.
Additionally, motionmark 1.3 is currently tier 2 for all platforms. For
the time being it may make more sense to only have it as tier 2 for Fx
applications, and tier 3 otherwise.
Differential Revision: https://phabricator.services.mozilla.com/D214035
Change the android-ui-tests docker image to be based on Ubuntu (as it was on github) rather than Debian, to restore the preferred log formatting of summary results.
Differential Revision: https://phabricator.services.mozilla.com/D214910
Change the android-ui-tests docker image to be based on Ubuntu (as it was on github) rather than Debian, to restore the preferred log formatting of summary results.
Differential Revision: https://phabricator.services.mozilla.com/D214910
Previously we set use_v8_context_snapshot to false to speed up build
times and be consistent with all platforms (since for some platforms
this was perma/highly intermittent e.g. win/linux, and others taking a
very long time e.g. arm mac cross build)
cstm-car-m tests have been perma failing recently (sometime between June
10th & 11th). Enabling this flag seems to help, but it is presently
unclear why that is, and why other platforms were not affected.
Differential Revision: https://phabricator.services.mozilla.com/D214257
We wanted to have a temporary workaround to explicitly allow/block
domains to use MFCDM due to the bug 1901334. As this is only a
workaround, we'd like to do it in a simplest way in order to save our
resource.
Our strategy is to keep Nightly unfiltered so that we could discover
more websites affected by the CSS problem. On other channels, we would
currently only allow Netflix to use MFCDM as we have been testing
with them and know that issue won't affect them.
We will first try to use the allow list first to ensure the MFCDM
playback quality, and gradually switch to the block list to enable the
MFCDM ability to more websites.
Differential Revision: https://phabricator.services.mozilla.com/D213820