1041219 - I sometimes see ghost-windows from openstreetmap.org

Status: RESOLVED DUPLICATE of bug 1238427

(Core :: DOM: Core & HTML, defect)

Description

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

I sometimes see ghost windows from openstreetmap.org (where I both browse and edit maps).

The first time I tried to debug the problem I crashed due to bug 1041218 after taking CC/GC dumps and having the right breakpoints set on destructors.

The second time I think I hit it (although I'm not sure, since I'd only taken a non-verbose about:memory which doesn't list the URLs for the ghost windows), I crashed due to bug 1036692 while doing a verbose about:memory.

I'm not inclined to try debugging further in the near future, at least until I have reason to think going to about:memory is less likely to crash.

Comment 1

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: find_roots.py output for first window

When I did get a chance to debug partially, there were three ghost windows, 2 of which showed up in the CC log.  Here's what heapgraph/find_roots.py had to say about one of them.

Comment 2

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: find_roots.py output for second window

Comment 3

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Actually, it turns out that about:memory's [Measure] button without "verbose" checked is pretty reliable, so I'm keeping an eye on that to see if I see the problem again (although without "verbose" I can't be sure if it's openstreetmap or not).  It's the verbose measurement that's crashing more often than not.

I haven't managed to see it yet.

Comment 4

[Nicholas Nethercote [inactive]]

> It's the verbose measurement that's crashing more often than not.

That's surprising. Verbose mode does not change how the data is collected; it only changes how the data is displayed. In non-verbose mode all the data is there, it's just that you have to click on less significant sub-trees to see expand them.

Comment 5

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Yes, but the current about:memory crash is bug 1036692, which is related to about:memory causing a slow script dialog (while already holding a cache lock, I think).

Comment 6

[Andrew McCreight [:mccr8]]

Are you still seeing ghost windows on that page, dbaron?

Comment 7

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

It's been a long time since I looked at about:memory regularly.  I'll try to remember to start again and get back to you.

Comment 8

[Andrew McCreight [:mccr8]]

I'm going to optimistically assume that this is a dupe of another geolocation ghost window leak.