Closed
Bug 1053078
Opened 10 years ago
Closed 10 years ago
ASAP mode breaks refresh driver throttling
Categories
(Core :: Layout, defect)
Tracking
()
RESOLVED
FIXED
mozilla34
People
(Reporter: mattwoodrow, Assigned: mattwoodrow)
References
Details
(Whiteboard: [talos_regression])
Attachments
(1 file)
7.28 KB,
patch
|
roc
:
review+
|
Details | Diff | Splinter Review |
The refresh driver throttling code has a safety mechanism where we unblock painting if we get 10 refresh driver ticks while frozen. This exists to stop us getting deadlocked if a response from the compositor gets lost somehow. ASAP mode (where we trigger refresh driver ticks as fast as possible) breaks this since we accumulate the 10 ticks almost immediately, even when the compositor is responding normally. We use this mode for most of our talos test, so the results (particularly on OSX) will likely appear to regress when we fix this since we'll correctly be keeping content in sync with the compositor instead of overproducing.
Assignee | ||
Comment 1•10 years ago
|
||
Adding avih and jmaher to the cc list since this is going to show up as a talos regression.
Assignee | ||
Comment 2•10 years ago
|
||
Assignee: nobody → matt.woodrow
Attachment #8472173 -
Flags: review?(roc)
Attachment #8472173 -
Flags: review?(roc) → review+
Assignee | ||
Comment 3•10 years ago
|
||
https://hg.mozilla.org/integration/mozilla-inbound/rev/76d4e6230498
Comment 4•10 years ago
|
||
(In reply to Matt Woodrow (:mattwoodrow) from comment #1) > Adding avih and jmaher to the cc list since this is going to show up as a > talos regression. Do you know this for a fact? Last time I tried I couldn't notice overproduction IIRC with tscrollx (can't find the bug where I said it though, but I do recall roc was part of the conversation as well).
Comment 5•10 years ago
|
||
https://hg.mozilla.org/mozilla-central/rev/76d4e6230498
Status: NEW → RESOLVED
Closed: 10 years ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla34
Comment 6•10 years ago
|
||
Matt, thanks for including us on this bug. I see 3 regressions from this change: 10.8 tscrollx 339% 10.8 tart 9.2% 10.6 tscrollx 202% as this change landed with 2 other patches (http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=f9c51223ffc9&tochange=76d4e6230498), it is possible that another patch caused the tart regression. here is a graph for 10.6: http://graphs.mozilla.org/graph.html#tests=[[287,63,21]]&sel=1407987418000,1408160218000&displayrange=7&datatype=running Is this the expected behavior from this patch?
Flags: needinfo?(matt.woodrow)
Assignee | ||
Comment 7•10 years ago
|
||
Yeah that's expected, any test that uses ASAP mode would potentially be affected.
Flags: needinfo?(matt.woodrow)
Comment 8•10 years ago
|
||
Thanks Matt, appreciate the heads up and reply!
Blocks: 1042202
Whiteboard: [talos_regression]
You need to log in
before you can comment on or make changes to this bug.
Description
•