This commit prepares the decks for turning specific Raptor tasks into
Raptor + browsertime tasks. The `--browsertime` flag to `mach try
...` flips the switch; eventually, the Raptor harness will recognize
the `--browsertime` flag and use browsertime to perform the pageload
measurements.
To run browsertime, we need:
1) Node.js
2) the browsertime `node_modules` (provided by the
`toolchain-browsertime` task)
3) ffmpeg (for producing videos from captured frames)
4) chromedriver (in the future, when targeting Chrome/Chromium)
5) geckodriver (provided by the `toolchain-*-geckodriver` tasks)
6) `PATH` configured
This commit arranges those things.
Since the configuration varies by test platform, and eventually we
expect the changes implemented by the flag to be moved into YAML task
definitions, we elect to use `by-test-platform` conditionals as much
as possible. The end expression is pleasant, thanks to
`evaluate_keyed_by`.
Handling PATH, however, is a rabbit hole. At this time, it's not
possible to use `fetch` task repackaging, because `releng-hardware`
doesn't support `zstandard` (Bug 1576244) and there's no appetite to
avoid `zstandard` entirely (Bug 1576698). Generally PATH is
configured using `mozharness` configuration files, which can execute
arbitrary Python and configure the PATH only for browsertime jobs.
However, the Raptor mozharness script itself runs the Raptor harness
in a stripped down environment, throwing away modifications to PATH.
It's not clear what impacts changing that has, so we leave it alone,
and add a `--browsertime-ffmpeg` flag and custom handling in the
Raptor harness. This can transition smoothly into a browsertime flag
(so that the PATH doesn't need to be set at all) and into a unified
interface for Raptor and `mach browsertime` to configure the
browsertime execution environment.
Differential Revision: https://phabricator.services.mozilla.com/D38781
--HG--
extra : moz-landing-system : lando
Remove test manifest annotations that specifically target fennec,
or likely target the android 4.3 emulator.
Differential Revision: https://phabricator.services.mozilla.com/D45018
--HG--
extra : moz-landing-system : lando
In the next part, the patch generated by a script removes "-moz" prefix
to all multi-column properties.
However, some of the tests already have standard multi-column properties
specified. To avoid duplicating CSS multi-column rules in these
files (after applying the next part), I manually remove the prefixed
rules beforehand as many as possible.
Differential Revision: https://phabricator.services.mozilla.com/D44869
--HG--
extra : moz-landing-system : lando
We were assuming that in the case that the status matched expected we'd have
no expected entry in the log data. But actually it was being set to None, so
handle both those cases.
Differential Revision: https://phabricator.services.mozilla.com/D44874
--HG--
extra : moz-landing-system : lando
Flex containers resolve a main-axis size for each of their flex items, and they
impose this size on the flex items by stomping on the items'
ReflowInput::ComputedISize() or ComputedBSize() data. For cases when we're
stomping on the flex item's block-axis, then this can make us improperly treat
this size as definite (i.e. use it for percent resolution).
The flexbox spec does call out some cases where the item's size *is* supposed
to be considered definite[1], but if we're not in one of those cases, we need
to be careful to treat the size as indefinite even though it's been resolved.
This patch achieves this by:
- adding a flag to FlexItem, which gets set in cases where we know we need to
treat the size as indefinite.
- adding a flag to ReflowInput, which gets set whenever a ComputedBSize is
imposed (if the FlexItem had its flag set).
- adding some code to the ReflowInput percent-resolution codepath to test for
this flag and skip past percent-resolution if it's set.
This patch makes us pass all of the existing testcases in the web-platform-test
"percentage-heights-003.html". This patch also adds a few more subtests
there to exercise cases where the flex container and item have orthogonal
writing-modes.
(Note the XXXdholbert comments in the patch about items being "fully
inflexible" in some cases in order to be treated as having a definite size.
If I were to address that comment here, then we would start failing the
web-platform-test "percentage-heights-005.html", which all browsers currently
pass[2]. Therefore, I'm not adding that restriction at this point.)
[1] https://drafts.csswg.org/css-flexbox/#definite-sizes
[2] https://wpt.fyi/results/css/css-flexbox/percentage-heights-005.html
Differential Revision: https://phabricator.services.mozilla.com/D43705
--HG--
extra : moz-landing-system : lando
https://drafts.css-houdini.org/worklets/#fetch-and-invoke-a-worklet-script
requires that the addModule() promise is rejected with an "AbortError"
DOMException. https://github.com/w3c/css-houdini-drafts/issues/509 proposes a
more specific error, but that has not been implemented in another browser.
StealExceptionFromJSContext() set ErrorCode to
NS_ERROR_INTERNAL_ERRORRESULT_JS_EXCEPTION if there was a pending exception
(or NS_ERROR_OUT_OF_MEMORY on failure to record the JS exception) or
NS_ERROR_UNCATCHABLE_EXCEPTION if there was no pending exception.
StealNSResult() converted NS_ERROR_INTERNAL_ERRORRESULT_JS_EXCEPTION to
NS_ERROR_DOM_INVALID_STATE_ERR.
Information about the error is lost on conversion of a SyntaxError to the
promise rejection, so allow the SyntaxError to be reported the console.
Depends on D44601
Differential Revision: https://phabricator.services.mozilla.com/D44602
--HG--
extra : moz-landing-system : lando
Previously, picture caching code would use the viewport of the
scroll root to find a clipping rect for picture cache tiles. This
viewport rect was also used to eliminate fixed position clip rects
on primitives that would otherwise cause unwanted invalidations
due to them moving relative to the scroll root when scrolls occur.
Now, the picture caching code uses a similar technique to Gecko
to find shared clips on primitives in a picture cache. These clips
are filtered out from being applied on a per-primitive basis, and
instead applied once during compositing the tiles into the parent
picture.
This is a potential performance improvement, since less per-item
work is required when building clip chains. More importantly, it
means the picture caching code correctly handles cases where the
scroll root contains fixed position elements (or other scroll
roots). This is a requirement before we can enable picture caching
on multiple slices.
Differential Revision: https://phabricator.services.mozilla.com/D44618
--HG--
extra : moz-landing-system : lando
Cosmin Sabou
Gurzau Raul
Jonathan Kingston
Narcis Beleuzu
John Dai
Nick Alexander
Sebastian Hengst
Geoff Brown
David Heiberg
Daisuke Akatsuka
Ting-Yu Lin
James Willcox
Thomas Nguyen
James Graham
Daniel Holbert
Coroiu Cristina
Rob Wood
Dorel Luca
Hiroyuki Ikezoe
Karl Tomlinson
Glenn Watson
Gijs Kruitbosch
Benjamin