111651 - trace-malloc blocks low-priority timers

Status: RESOLVED FIXED

(Core :: XPCOM, defect, P1)

Description

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

I was *sure* I tested this before when investigating this problem, but
apparently I didn't do it correctly.

Anyway, the gtk_idle_add that's |#ifdef NS_TRACE_MALLOC| in nsAppShell::Run
prevents low or lowest priority timers from being processed since g_main_pending
in nsTimerGtk's TimerCallbackFunc always returns true.  This prevents a bunch of
things from working when trace-malloc is being used.

Comment 1

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

What's the purpose of flushing the log files, anyway?

Comment 2

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

Attachment: remove the code in nsAppShell (GTK)

Comment 3

[Brendan Eich [:brendan]]

The purpose was to keep real-time tmreaders happy (I didn't want to assume that
Mozilla malloc'd regularly enough to flush important data in a timely fashion!).
 Can't we fix the GTK-based code here so that we flush, yet nothing starves?

Anyway, r/sr=[e-mail] if you need to remove the flushing for now.  We
can add it back if a real-time tmreader needs it.

/be

Comment 4

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

Maybe we could put the flushing on a low-priority timer instead? :-)  Would it
be bad for xpcom to depend on timer |#ifdef NS_TRACE_MALLOC|?

Comment 5

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

Blizzard - any ideas on an easy way to fix this?

Comment 6

[Christopher Blizzard (:blizzard)]

You should put that function elsewhere.  Maybe in the code that wakes up on X
events?  And on exit maybe?  There are lots of options.

Comment 7

[Chris Waterson]

Comment on attachment 59003 [details] [diff] [review]
remove the code in nsAppShell (GTK)

sr=waterson

Comment 8

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

Removal checked in 2001-11-27 20:48 PDT.  We still need to figure out a
replacement, but I wanted our leak / bloat stats to be accurate in the meantime.

Comment 9

[Brendan Eich [:brendan]]

If we turn on trace-malloc for DEBUG builds, don't we need this fix?  Marking a
dependency on bug 112470, which I note is targeted at 0.9.7 still.

/be

Comment 10

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

I checked in the fix so that it doesn't block the timers anymore.  I just
haven't replaced the code that I removed.

Comment 11

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

Hrm, I could probably back out the removal now that pavlov's new timer patch landed.

Comment 12

[Brendan Eich [:brendan]]

rs=[e-mail] on backing out now that we have xp timers.

/be

Comment 13

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

OK, marking FIXED since this was fixed by pav's timer checkin, or something like
that.  I undid the backout 2002-01-15 16:40 PST.