This adds the following configurations on central and autoland:
> test-linux1804-64-asan/opt-web-platform-tests-fis-e10s
Depends on D110553
Differential Revision: https://phabricator.services.mozilla.com/D110554
This adds the following configurations on central and autoland:
> test-windows10-64-qr/debug-web-platform-tests-fis-e10s
Differential Revision: https://phabricator.services.mozilla.com/D110553
This makes the snapback less aggressive, and feels more like Safari to me.
Reducing the stiffness makes the animation take longer to complete.
To counteract the longer time, I've also reduced the damping a little bit;
reducing the damping makes the animation complete a little more quickly.
Depends on D110845
Differential Revision: https://phabricator.services.mozilla.com/D110846
CLOSED TREE
Backed out changeset 813c7ce3607b (bug 1395819)
Backed out changeset 46083fddeb0d (bug 1395819)
Backed out changeset 990441f04a86 (bug 1395819)
Previously the test only visited a single site.
We now capture metrics over a user journey of visiting 5 sites cold and then the same sites warm.
This gives a more representative view of performance.
Differential Revision: https://phabricator.services.mozilla.com/D110833
This patch makes MediaManager notify of recording-device-events as soon as any
gUM-track has ended, as opposed to when all tracks in a gUM-request have ended.
This effectively makes temporary permissions per-track rather than per-request.
Differential Revision: https://phabricator.services.mozilla.com/D109943
The system changes always affect all windows, and it makes sense to have change
listening in the same file where the state is read.
This also removes the need to unregister our listener, because we only have one
global leaked listener.
Differential Revision: https://phabricator.services.mozilla.com/D110769
This fixes some transparent colors, such as the odd tree row color in dark mode (see bug 1698763).
It also seems like a more sensible default.
If we find problems with unexpectedly-transparent colors that we want to be opaque, we can fix those on a per-color basis.
Differential Revision: https://phabricator.services.mozilla.com/D110767
-[NSOpenGLContext pixelFormat] is 10.10+
-[NSScreen backingScaleFactor] is 10.7+
+[NSEvent isSwipeTrackingFromScrollEventsEnabled] is 10.7+
Differential Revision: https://phabricator.services.mozilla.com/D110772
This change constitutes a way we can redirect users to an actual page that
explains what captive portals are and why we are making these requests.
Normally users should not see this page, as we only compare the contents
of a small html file. The meta redirect only happens when loaded in a
page.
The SUMO URL https://support.mozilla.org/kb/captive-portal will automatically
redirect to the appropriate locale.
Differential Revision: https://phabricator.services.mozilla.com/D99773
All NativeGetFont implementations do, so child processes see something
different from parent processes which is pretty unfortunate, and isn't
the goal of RemoteLookAndFeel.
I found this by code inspection. Seems this is mostly used for debugging
etc?
Differential Revision: https://phabricator.services.mozilla.com/D110914
Commit message, comments, and test cases for s-s patch that landed
separately.
Several interlocking parts to the baseline part of the patch:
First, the pre-reservation of the value stack in the decode loop does
not work for opcodes with more than a fixed number of results: a
function or a block can return a larger number of values than any
reasonable pre-reserved size. The pushResults method must reserve
space for the values it will push, and must propagate OOM up the call
tree. This requires a fair amount of plumbing but is straightforward.
Second, the way the reservation should be made both in pushResults and
in the decode loop is to take into account the max number of pushes
per opcode, which can be defined as a constant (here very
conservatively as 10), but which is not "2" as asserted by a comment
in the existing code, and even if it were "2" it would not be that for
the reason given in the comment.
Third, the reservation code in the decode loop is cleaned up by
removing the guard that runs reservation code only once every 50
decode iterations, since this guard was always of questionable utility
and seems only to complicate the reasoning around the reservation
logic.
Finally, the guarded reservation code also contained a check for
available space in the temp allocator the baseline compiler uses for
allocating delayed code generation objects. This check was necessary
because that allocator is infallible. However, the compiler uses the
allocator rarely, and all existing uses of the allocator check for
allocation failure already and propagate the OOM, so the reservation
check can be removed if the allocator is simply marked as fallible.
On the ion side of things:
Every time we process an open-ended list of values for which we may
allocate nodes, we must call ensureBallast on the allocator, since it
is infallible. This was done in one place in WasmIonCompile, but not
both places where it was needed.
As far as the test case is concerned:
This test was enough to find the errors in both baseline and ion. It
can probably be pared down from the 100..300 loop it has now to
something smaller, but there you have it.
Differential Revision: https://phabricator.services.mozilla.com/D98513
We only need to react to a completed captive portal check when the captive
portal was previously locked. When not in a captive portal, this notification
only means the check has completed - no network properties have changed.
Differential Revision: https://phabricator.services.mozilla.com/D109723