After bug 2010890, addProfilerMarker's data argument is
(UTF8String or object): a non-string msg takes the object branch and
is JSON-stringified. info() always wants a text marker, so stringify
here to take the UTF8String branch unconditionally.
Differential Revision: https://phabricator.services.mozilla.com/D302741
Same reasoning as the mochitest cleanup in the previous commit. Neither
xpcshell task in taskcluster/kinds/test/xpcshell.yml sets
test-manifest-loader: null, so taskgraph always pre-resolves their
manifests and passes them via MOZHARNESS_TEST_PATHS. desktop_unittest.py
never appends --total-chunks/--this-chunk to the xpcshell command,
self.totalChunks stays at its default of 1, and the chunk_by_slice call
in runxpcshelltests.py never fires.
Differential Revision: https://phabricator.services.mozilla.com/D302813
This reverts commit bde4bcc89a.
Revert "Bug 2019113 - Part 3: Provide the window to createLinuxDesktopEntry when pinning. r=nrishel"
This reverts commit f38b471713.
Revert "Bug 2019113 - Part 2: Add tests for DynamicLauncher portal usage from the browser. r=emilio,extension-reviewers,robwu,gerard-majax,Thinker,ahal"
This reverts commit 89a81eb63b.
Revert "Bug 2019113 - Part 1: Use the DynamicLauncher portal from createLinuxDesktopEntry and deleteLinuxDesktopEntry. r=emilio,nrishel"
This reverts commit b6b202794e.
selftest.py: pass timeoutFactor to runTests() so the per-sub-test timeout
is not silently reset to 30s. Previously, setUp() set
self.x.harness_timeout=60 for ccov, but runTests() unconditionally
overwrites harness_timeout from HARNESS_TIMEOUT*timeoutFactor, so that
assignment had no effect.
Coverage flushing (flushCounters + IOUtils.move) before and after each
sub-test adds significant overhead on ccov builds - on Windows this
consistently pushes the sub-test past the 30s default timeout, causing
TEST-UNEXPECTED-TIMEOUT. A 1.5x factor is applied universally to
provide headroom on slow workers; ccov builds get 2.0x to account for
the extra overhead.
Differential Revision: https://phabricator.services.mozilla.com/D299583
Pass the exception's actual stack via the harness's stack field instead of
synthesizing one rooted in simpletestOnerror or assertNoUncaughtRejections.
For uncaught rejections, also pass the rejection date as the assertion's
timestamp rather than baking it into the failure message, so otherwise
identical failures fingerprint to the same message. The rejection failure
message is simplified to "uncaught rejection: <message>".
Plumbing changes:
- Assert.report / Assert.AssertionError accept a string stack and an
optional time.
- StructuredLog.testStatus accepts an optional time that overrides the
default Date.now().
- The mochitest reporter and browser-harness.xhtml forward the new time
field through testResult and addResult.
Differential Revision: https://phabricator.services.mozilla.com/D296797
The profiler-related code paths in head.js run after the test file has
been loaded into the harness' global scope, so a test that declares
`var Cc = ...` or `var Services = ...` at top level shadows the bindings
that head.js relies on. Switch those paths to the existing `_Services`
pattern and add an `_Cc` alias captured before any test file is loaded.
test_utf8_illegals.js was hitting this: `var Cc = Components.Constructor`
made head.js throw a TypeError when setting up the profiler timeout
timer, which skipped `_PromiseTestUtils.uninit()` and
`Cu.exitIfInAutomation()` and led to a shutdown crash. Rename the
test's local to `CC` to match the convention used elsewhere.
Differential Revision: https://phabricator.services.mozilla.com/D295280
`NS_ERROR_NET_INTERRUPT` is not in the restart condition list, so using
it silently drops the request without fallback to H1/H2. Use
`NS_ERROR_NET_RESET` instead.
Differential Revision: https://phabricator.services.mozilla.com/D287813
`NS_ERROR_NET_INTERRUPT` is not in the restart condition list, so using
it silently drops the request without fallback to H1/H2. Use
`NS_ERROR_NET_RESET` instead.
Differential Revision: https://phabricator.services.mozilla.com/D287813
`NS_ERROR_NET_INTERRUPT` is not in the restart condition list, so using
it silently drops the request without fallback to H1/H2. Use
`NS_ERROR_NET_RESET` instead.
Differential Revision: https://phabricator.services.mozilla.com/D287813
The maintenance cost of supporting non-profiler builds now outweighs the benefits.
While previously maintained for Tier 3 platforms, the profiler now supports marker recording across these targets.
Removing this toggle simplifies the build system and allows for faster development.
Differential Revision: https://phabricator.services.mozilla.com/D284849
The maintenance cost of supporting non-profiler builds now outweighs the benefits.
While previously maintained for Tier 3 platforms, the profiler now supports marker recording across these targets.
Removing this toggle simplifies the build system and allows for faster development.
Differential Revision: https://phabricator.services.mozilla.com/D284849
This logic was only tracking privileged parent process javascript,
whereas JS_CODE_COVERAGE_OUTPUT_DIR is the typical equivalent and works
on all processes.
Differential Revision: https://phabricator.services.mozilla.com/D285084
The error summary formatter sums raw test_start-to-test_end times for
xpcshell manifests, but this overestimates wall-clock time when tests run
in parallel across multiple threads. Pass the thread count from the
xpcshell harness via group_start extra data, and divide individual test
times by that count during the parallel phase.
Differential Revision: https://phabricator.services.mozilla.com/D284511
This reverts commit 0382300c2d.
Revert "Bug 2015841 - ruff: fix 'String is split more times than necessary' r=linter-reviewers,ahal,sergesanspaille"
This reverts commit 1e0b4c3c7b.
Revert "Bug 2015841 - Reformat the code with ruff-format 0.15.0 r=linter-reviewers,perftest-reviewers,mozperftest-reviewers,releng-reviewers,sessionstore-reviewers,dom-storage-reviewers,dom-worker-reviewers,whimboo,chutten,ahal,sparky,sclements"
This reverts commit 15a20a737e.
Revert "Bug 2015841 - Update to ruff 0.15.0 r=linter-reviewers,ahal"
This reverts commit 4b0785d39e.
This reverts commit 5eca473e96.
Revert "Bug 2015841 - ruff: fix 'String is split more times than necessary' r=linter-reviewers,ahal,sergesanspaille"
This reverts commit 654a09320c.
Revert "Bug 2015841 - Reformat the code with ruff-format 0.15.0 r=linter-reviewers,perftest-reviewers,mozperftest-reviewers,releng-reviewers,sessionstore-reviewers,dom-storage-reviewers,dom-worker-reviewers,whimboo,chutten,ahal,sparky,sclements"
This reverts commit 93b38f0d79.
Revert "Bug 2015841 - Update to ruff 0.15.0 r=linter-reviewers,ahal"
This reverts commit cc04e8271b.
The remote XPCShell test configuration for Android builds up `XPCShellRemote.env` with environment
variables for the remote device. We also want some of these on the host for httpserver/NodeServer.sys.mjs,
but some of these may break host apps. In particular the 'HOME' variable is used on host for
'adb' to run so we want to avoid clobbering it with the value used for the Android device.
Differential Revision: https://phabricator.services.mozilla.com/D278226
Randell Jesup