Open Bug 962911 Opened 6 years ago Updated 6 years ago

Bring back more-than-one-native-event-per-gecko-event processing

Categories

(Core :: Widget, defect)

x86_64
Linux
defect
Not set

Tracking

()

People

(Reporter: smaug, Assigned: smaug)

References

Details

Attachments

(1 file)

In bug 930793 I took the rather extreme approach to process only one native event
per gecko event. But as bug 961869 for example shows, that was a bit too much.
Instead of bringing back max 20ms processing, the patch gives max 10ms.

(pushed from a bit old tree)
https://tbpl.mozilla.org/?tree=Try&rev=25da36e626c7
https://tbpl.mozilla.org/?tree=Try&rev=a1fdcbb35dd2

(pushed from up-to-data)
https://tbpl.mozilla.org/?tree=Try&rev=72e57f6b02f2

Based on try the patch seems to work rather well.
tp doesn't seem to regress.

The old code was
http://mxr.mozilla.org/mozilla-aurora/source/widget/xpwidgets/nsBaseAppShell.cpp?mark=254-254,275-278#253

(apparently roc is on vacation, so perhaps you avih can review this partial backout.)
Attachment #8364098 - Flags: review?(avihpit)
Comment on attachment 8364098 [details] [diff] [review]
more_native_events.diff

Review of attachment 8364098 [details] [diff] [review]:
-----------------------------------------------------------------

Good move. This is closer to how native events were processed before bug 930793 landed, but still without the starvation. /me likes.
Attachment #8364098 - Flags: review?(avihpit) → review+
Backed out for making B2G reftests much more failure prone. R1 and R5 were near perma-fail after this landed. It's worth noting that the R1 failures are almost all on file, so it's probably exacerbating a preexisting timing issue.
https://hg.mozilla.org/integration/mozilla-inbound/rev/93440685551d

https://tbpl.mozilla.org/php/getParsedLog.php?id=33461914&tree=Mozilla-Inbound
https://tbpl.mozilla.org/php/getParsedLog.php?id=33461884&tree=Mozilla-Inbound
Bah. 

Tryserver push with limit 20, which is the old value.

https://tbpl.mozilla.org/?tree=Try&rev=ddf67ad3fae6
But I need to now test whether 20ms causes other regressions.
I thought it did...
looks like the fractal demo is still ok with 20ms.
Confirmed that B2G is a lot more green post-backout.
Nice, 20ms fails too. And I think wrapper.html is just broken.
...fixing another test.
You need to log in before you can comment on or make changes to this bug.