Debug Assertion on page already registered in the profiler browsing to youtube.com
Categories
(Core :: Gecko Profiler, defect, P2)
Tracking
()
Tracking | Status | |
---|---|---|
firefox-esr60 | --- | unaffected |
firefox67 | --- | wontfix |
firefox68 | --- | fixed |
People
(Reporter: jesup, Assigned: canova)
References
(Regression)
Details
(Keywords: assertion, regression)
Attachments
(3 files)
STR: with a fairly new profile with a Debug non-opt build, browse to a site (I used mozilla.github.io), then start the profiler (already installed) and browse to youtube.com.
Hit an assertion at tools/profiler/core/platform.cpp:323 due to a page already being in the page cache. Dumps of the 3 pages in the cache and the new page are attached -- the collision was on HistoryId 13, which one has as about:blank, and the other has as "https://accounts.google.com/ServiceLogin?service=youtube&continue=https%3A%2F%2Fwww.youtube.com%2Fsignin%3Ffeature%3Dpassive%26hl%3Den%26action_handle_signin%3Dtrue%26next%3D%252Fsignin_passive%26app"... (truncated).
They compare as equal in PageInformation::Equals() (same docshell, id, and subframe).
Unsure if this is repeatable. Note: browsing with a non-opt debug build is Very Slow....
I presume this was introduced by bug 1417976
Reporter | ||
Comment 1•6 years ago
|
||
Reporter | ||
Comment 2•6 years ago
|
||
If you remove the ifdef, and make it MOZ_RELEASE_ASSERT(), it failed on the the first pageload of youtube for me (after loading https://mozilla.github.io/webrtc-landing). Trying again and 1 or two other URLs didn't hit, but clearly there's a hole here.
Reporter | ||
Comment 3•6 years ago
|
||
Assertion with opt build (modified to include the assert) on a CNN page.
Assignee | ||
Comment 4•6 years ago
|
||
Thanks for filing this Randell. I'm aware of that issue and actually I have a fix for that here: https://phabricator.services.mozilla.com/D18331.
But I wasn't working on that for a while after publishing the patch because of other works that I need to finish. I came back to it and finishing up/writing some tests now. It will be fixed after that.
Reporter | ||
Comment 5•6 years ago
|
||
Nazim, we need to disable this, or fix it - it makes debug builds die with the profiler way too often. Currently I build with patches to make that code an #if 0 to avoid crashing
Assignee | ||
Comment 6•6 years ago
|
||
Ok, how about disabling it until we fix it? Will submit a patch that disables that for now. Sorry about that.
Assignee | ||
Comment 7•6 years ago
|
||
Comment 9•6 years ago
|
||
bugherder |
Updated•6 years ago
|
Updated•5 years ago
|
Updated•3 years ago
|
Updated•3 years ago
|
Description
•