Confirmed users
995
edits
Electrolysis/Meetings/2016-02-25: Difference between revisions
Electrolysis/Meetings/2016-02-25 (view source)
Revision as of 17:25, 25 February 2016
, 25 February 2016→RyanVM
(jimm) |
(→RyanVM) |
||
| (16 intermediate revisions by 10 users not shown) | |||
| Line 17: | Line 17: | ||
** FINALLY landed this. smaug, mccr8 and ting helped huge, and we found some leaks along the way. | ** FINALLY landed this. smaug, mccr8 and ting helped huge, and we found some leaks along the way. | ||
** (The short story is that my patch fixed half of a leak, which caused many more things to be added to the cycle-collector graph, but those things were kept alive too long so the graph got larger and larger, and then on shutdown when analyzing the cc logs, we'd take too long and miss a native event, which would cause us to leak nsBaseAppShell.) | ** (The short story is that my patch fixed half of a leak, which caused many more things to be added to the cycle-collector graph, but those things were kept alive too long so the graph got larger and larger, and then on shutdown when analyzing the cc logs, we'd take too long and miss a native event, which would cause us to leak nsBaseAppShell.) | ||
* '''Perf P2''' {{bug|1174770}} - tpaint regressions (3%-33%) in e10s mode compared to non-e10s mode | |||
** Found that we're sending two sync IPC messages, one after the other, accounting for > 50% of the regression. Filed {{bug|1251032}} | |||
** We're also sending two sync IPC messages for IME. Filed {{bug|1251063}} | |||
* {{Bug|666365}} - Figure out what TabChild::SetVisibility needs to do | |||
** I have a needinfo on this, and it's taking me a while to sort this out. I _think_ we're doing the right thing by just having the TabChild be "visible" by default, but I haven't fully verified this yet. | |||
== jimm == | == jimm == | ||
* | * {{bug|1232181}} (M8, plugin window captures) - under review | ||
== gabor == | |||
* {{bug|1179735}} (p1 perf, - tp5o_scroll regression) - APZ causing the regression, and it causes other regressions as well. {{bug|1144120}} has some fresh data about other regressions. | |||
* {{bug|1186585}} (p1 perf - 28.2% - 48.5% tps talos regression with e10s enabled) - did some profiling, it seems like the compositor spends more time in RecvUpdateNoSwap, I wonder why does that take more time in e10s. The compositor should do the same work for tab switch in both case no? It is not (just) APZ I tested it. | |||
* {{bug|1248806}} - (p1 crash - Recursive initialization of nsLayoutStatics when capability.policy.policynames is set) - this flag is used for various corporations for intranet and what not. Legacy stuff but it's important (Bobby tried to remove it once but had to put it back in the end for the high demand) I see a few solutions for the problem want to talk to Bobby or Boris about which one should I pick. | |||
== mrbkap == | |||
* {{bug|1183915}} - Landed drag and drop of images from Firefox to certain other applications. | |||
* {{bug|1095484}} - WIP - Making links to extension-added custom protocol handlers work from content processes. Debugging some bugs in the patch. | |||
* More test fixes and work related to tests. | |||
== poiru == | |||
* [https://gist.github.com/poiru/1ceec405172bcafa7305 Latest analysis] | |||
* {{bug|1222890}} Crash rate seems to vary significantly build by build, trying to figure out why | |||
== chutten == | |||
Chatted with :avih and :jmaher about talos tests and e10s (specifically concerned with tp5o_responsiveness). We now have a handle on why the number is so much better in e10s (it's measuring in the parent only) and :avih's tracking down someone who wants to be responsible for the test longer-term. | |||
== bobowen == | |||
* {{bug|1245309}} - D3D11CreateDevice fails with E_FAIL in sandbox - mchang has reproduced this, but I can't on any of the three machines I have with an nvidia card. It's blocking {{bug|1246505}} to let the low integrity sandbox ride the trains. | |||
* {{bug|1247959}} - window.dump() does not work on Windows when sandboxed - {{bug|1173698}} has been duped to this one. The same work around of redirecting stdout/err to a file seems to work. | |||
== bsmedberg == | |||
* Regression in slow-script dialog count has been identified and cpeterson will arrange a meeting to determine next steps. | |||
* https://wiki.mozilla.org/Electrolysis/Release_Criteria | |||
== RyanVM == | |||
* Avi has been working to identify owners for the Talos test suites and their current status of e10s vs. non-e10s. Will be going through the results to work to identify owners of tests that need investigation. | |||
* Ongoing discussions about manual test coverage for e10s | |||