1652883 - Shared surfaces memory reporter should give KIND_OTHER a different path

Status: RESOLVED FIXED

(Core :: Graphics: ImageLib, task)

Description

[Andrew McCreight [:mccr8]]

In ImageMemoryReporter::ReportSharedSurface(), the surface is reported as a KIND_NONHEAP if it was not already reported, according to some reasoning I don't understand, or KIND_OTHER if it was already reported.

The problem is that you can end up with paths that vary only in their PID that have different kinds. When you do a diff, it first normalizes the PID, so you end up with multiple entries with the same path but different kinds. This causes an assertion in about:memory. In fact, when I capture a memory report locally and attempt to diff it with itself, it hits this assertion.

I think the easiest fix is to make the last segment something like "/decoded-other" instead of "/decoded-nonheap if it is KIND_OTHER instead of KIND_NONHEAP.

Comment 1

[Andrew McCreight [:mccr8]]

Attachment: Bug 1652883 - Shared surfaces memory reporter should give KIND_OTHER a different path.

If you do a diff, all of the different shared surfaces get merged together,
because the diffing process removes all PIDs from the paths. This means
you end up hitting an assertion when the diffing process tries to combine
entries with the same path but different kinds.

Comment 2

[Andrew McCreight [:mccr8]]

I should probably write a test that we can diff a memory report with itself and it produces an empty report...

Comment 3

[Andrew McCreight [:mccr8]]

I started hacking up a test that diffs a report against itself, but it doesn't produce the exception. It looks like this particular issue requires that WebRender is enabled, and the memory reporter tests are crusty mochitest-chrome tests that presumably don't have it enabled. Maybe I can rewrite it as a browser chrome test or something. It looks like we do run those on Windows with WebRender enabled.

Comment 4

[Andrew McCreight [:mccr8]]

I don't think I really have time to investigate how to write an about:memory test in a new suite so we get WebRender coverage.

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/25360f9d4a9c
Shared surfaces memory reporter should give KIND_OTHER a different path. r=aosmond

Comment 6

[Alexandru Michis [:malexandru]]

https://hg.mozilla.org/mozilla-central/rev/25360f9d4a9c