Notable WebGPU CTS changes:
- Demotions from tier 2 to tier 3:
- `webgpu:shader,execution,expression,unary,{i32,u32}_conversion:f32:*` started regressing on Windows.
- Promotions to tier 2 from tier 3:
- Across all platforms:
- `webgpu:shader,execution,expression,call,builtin,cross:f32:*`
- `webgpu:shader,execution,expression,unary,i32_conversion:abstract_float:*`
- `webgpu:shader,execution,expression,unary,u32_conversion:abstract_float:*`
- `webgpu:shader,execution,limits:nesting_depth_braces:*`
- `webgpu:shader,validation,expression,call,builtin,value_constructor:vector_copy:*`
- `webgpu:shader,validation,expression,call,builtin,value_constructor:vector_elementwise:*`
- `webgpu:shader,validation,expression,call,builtin,value_constructor:vector_mixed:*`
- `webgpu:shader,validation,expression,call,builtin,value_constructor:scalar_zero_value:*` was promoted on Linux and Windows, but stays in macOS on account of failing `f16` functionality. My guess is that these failures are also present on the other platforms, but we don't see the because `shader-f16` is not aviailable on our other CI platforms yet.
- From macOS, completing the set of all platforms passing:
- `f16` functionality, which, again, was likely not visible on our other CI platforms before:
- `webgpu:shader,execution,expression,call,builtin,cross:f16:*`
- `webgpu:shader,validation,decl,override:type:*`
- `webgpu:api,validation,capability_checks,limits,maxUniformBufferBindingSize:createBindGroup,at_over:*`
was promoted on Windows.
Differential Revision: https://phabricator.services.mozilla.com/D243593
This implements the crash helper service used to move child process crash
report generation out of the main process and into its own process. This is
implemented as a separate executable that is launched on startup by the main
process on the desktop platforms and as a service hosted by a separate process
on Android.
The crash helper process is started when the first exception handler is set
on desktop platforms and before loading libxul on Android. In both cases
starting this process happens asynchronously so that neither the main process
nor child processes have to wait for it to come up. If a crash happens before
the crash helper has started, the crashed process will wait for it to fully
come up and then proceed with regular crash generation.
IPC with the crash helper is implemented using Unix sockets on Linux and macOS
with the former using sequential packets and the latter using stream sockets.
On Windows we use named pipes. In all cases the choice of IPC was dictated both
by the requirement to eventually talk directly to child processes from within
the sandbox, and to external processes in case of Windows as the Windows
Error Reporting exception handler must be able to reach out to the helper from
within a restricted context. These particular requirements are not used yet but
will be as we move more logic out of the main process logic.
Differential Revision: https://phabricator.services.mozilla.com/D231083
Split out, because the upgrades are split into multiple commits, but we
require a single vendoring to avoid duplicated packages.
Differential Revision: https://phabricator.services.mozilla.com/D241959
Split out, because the upgrades are split into multiple commits, but we
require a single vendoring to avoid duplicated packages.
Differential Revision: https://phabricator.services.mozilla.com/D241959
WRT WebGPU CTS, we have some interesting changes:
- Promotions from tier 3 to tier 2:
- `webgpu:shader,execution,expression,call,builtin,textureStore:out_of_bounds:*`
- All of `webgpu:shader,execution,expression,binary,ai_arithmetic:*` on Windows and Linux.
- Some of `webgpu:shader,execution,expression,binary,bitwise:*` on Windows and Linux:
- `…:bitwise_and:*`
- `…:bitwise_exclusive_or:*`
- `…:bitwise_or:*`
- `webgpu:shader,execution,expression,call,builtin,textureStore:out_of_bounds:*`
- Some of `webgpu:shader,execution,expression,unary,*` on Windows and Linux:
- `…,ai_assignment:abstract:*`
- `…,ai_assignment:abstract:*`
- `…,ai_complement:complement:*`
- `webgpu:shader,execution,limits:switch_case_selectors:*` on Windows and macOS.
- `webgpu:shader,execution,limits:workgroup_array_byte_size_override:*` on all but Windows debug.
- `webgpu:shader,execution,zero_init:compute,zero_init:*` on macOS.
- `webgpu:shader,validation,expression,access,vector:abstract:*`
- `webgpu:shader,validation,expression,call,builtin,textureGather:offset_argument,non_const:*` on all but Windows debug.
- `webgpu:shader,validation,expression,call,builtin,textureSample:offset_argument,non_const:*`
- `webgpu:shader,validation,expression,call,builtin,textureSampleGrad:offset_argument,non_const:*`
- Most of `webgpu:shader,validation,statement,switch:*`:
- `…:case_types_match:*`
- `…:condition_type_match_case_type:*`
- `…:parse:*`
- Demotions from tier 2 to tier 3:
- `webgpu:api,validation,capability_checks,limits,maxColorAttachmentBytesPerSample:beginRenderPass,at_over:*`
- Otherwise notable potential regressions:
- `webgpu:shader,validation,expression,matrix,mul:overflow_scalar_abstract:*`
Differential Revision: https://phabricator.services.mozilla.com/D242218
This vendors the `desktop-138` branch [1] of application-services. The `main`
branch currently does not have some PRs that desktop needs because they break
mobile, and we need this on desktop ASAP.
This patch is larger than usual because the vendor includes some major changes
to the application-services `suggest` component, including new and changed API.
As a consequence this patch includes the following important changes:
## New `suggest` API & uniffi
`SuggestStoreBuilder::remote_settings_service` and `RemoteSettingsService::new`
are exposed to JS as synchronous functions. There's no need for them to be
off-main-thread.
## Telemetry
The labels of `suggest.ingest_time`, `ingest_download_time` and `query_time` had
to be updated due to changes in the Rust component. These are minor updates that
don't need a data review.
## Urlbar
I had to make the following changes to urlbar. I tried to keep them to a minimum
for now. There are opportunities for improvements in follow-ups.
* Appropriate minimal integration changes to `SuggestBackendRust` for creating
the `SuggestStore` and setting up the RS config
* The Rust component uses new RS collections, which breaks tests. I tried to fix
them without touching too many lines. There are definitely opportunities to
improve these tests and test helpers that I'd like to come back to.
* A fix to `RemoteSettingsServer` that's required due to the new RS client used
by the Rust component
## Late writes due to the new RS client & `AsyncShutdown`
This is a urlbar change but it's worth calling out separately. I pushed all
these changes to tryserver, and there was a failure at the end of the browser
Suggest tests due to `LateWriteObserver` [2].
The late write happens when the app exits: `SuggestStore` is dropped, which
causes the new app-services RS client to drop its Sqlite connection, which
causes Sqlite to sync the RS client's DB to disk. This hasn't been a problem
before because `suggest` currently uses the old RS client, which doesn't keep a
DB. (`suggest` does have its own separate Sqlite DB, and I didn't investigate
why this isn't a problem for it, mainly because it makes sense that the new RS
client would sync its DB when it's dropped and that might be considered a "late
write" when it happens on app shutdown.)
According to the stack in the log, `SuggestStore` is dropped by
`nsCycleCollector`. I can't see how `SuggestBackendRust.#store` is involved in a
cycle, and I don't know if something in this patch is causing a cycle where
there wasn't one before. Maybe there always was. And I don't know if the cycle
is what's causing the all this to happen too late on shutdown. Maybe it's
unrelated. (I'll paste the stack in a Phabricator comment.)
The `SuggestStore` is definitely kept alive until
`AsyncShutdown.profileBeforeChange` since we have a barrier for that phase that
calls `interrupt()` on it. Maybe that's simply the problem and we're using a
phase that's too late in shutdown. But again I don't know why it wouldn't also
be a problem for Suggest's own DB.
The only fix I found is to replace `AsyncShutdown.profileBeforeChange` with
either `quitApplicationGranted` or `profileChangeTeardown`, and then null out
`#store` in the callback (after we call `interrupt()` on it). I assume that
fixes it because `profileBeforeChange` runs later than those other two phases.
So I replaced `profileBeforeChange` with `profileChangeTeardown`. I don't know
which of `quitApplicationGranted` or `profileChangeTeardown` is better. I think
it probably doesn't matter. I chose `profileChangeTeardown` because it's closer
to `profileBeforeChange`. (The order is: `quitApplicationGranted`,
`profileChangeTeardown`, `profileBeforeChange`.)
[1] https://github.com/mozilla/application-services/tree/desktop-138
[2] https://treeherder.mozilla.org/jobs?repo=try&revision=1639f87aa46f1afaf50901d80c8282861700019b
Differential Revision: https://phabricator.services.mozilla.com/D240919
Special thanks to Vladimir Vetrov <pdastore1@gmail.com>, who provided a patch
that would eventually inform parts of this patch (in particular, how to use
some of the certificate-related Android APIs).
Differential Revision: https://phabricator.services.mozilla.com/D238821
Nested declarations were getting the location of their parent rule, which would
cause issue in the DevTools Inspector (e.g. for InspectorUtils.getRelativeRuleLine).
We now track the first declaration of the nested declaration to set it as the location
of the rule.
This required some changes in upstream cssparser that are vendored in this patch:
- 5e477ab819
- 958a3f098a
Differential Revision: https://phabricator.services.mozilla.com/D238911
