1028456 - More leaks with _PR_Getfd

Status: NEW

(Core :: XPCOM, defect)

Comment 1

[Andrew McCreight [:mccr8]]

Attachment: BC1 and BC3 fd-related leaks

Two more fd-related leaks showed up on the merge from m-c.  The bc1 wasn't happening 100% of the time.  I landed a new fairly broad suppression of _PR_Getfd for them, as there are a few variants of this that have shown up, and I have no idea what is going on.

Comment 2

[Andrew McCreight [:mccr8]]

We're actually suppressing the _PR_Getfd leaks pretty broadly now, FWIW, because they kept coming up.  Though pretty small each time.

Comment 3

[Nathan Froyd [:froydnj]]

Just out of curiosity, are LSAN runs --enable-debug or --disable-debug?

Comment 4

[Andrew McCreight [:mccr8]]

Disable.  This does cause problems in some areas of the code that don't bother running cleanup code in non-debug builds.

Comment 5

[Nathan Froyd [:froydnj]]

(In reply to Andrew McCreight [:mccr8] from comment #4)
> Disable.  This does cause problems in some areas of the code that don't
> bother running cleanup code in non-debug builds.

Hm.  NSPR has some code to recycle its internal file descriptors and comments claim that it is "disabled in opt builds", but reading the code, it looks like that comment is inaccurate.  The caching logic looks...confusing.

The cache ought to be cleared on shutdown via PR_Cleanup, but I don't think we ever call that...?

Comment 8

[Ryan VanderMeulen [:RyanVM]]

This is still pretty widespread across the mochitest suites.