17390 - leaking documents and webshells while running editor

Status: VERIFIED FIXED

(Core :: XUL, defect, P1)

Description

[Warren Harris]

Akkana Peck did this:

  setenv XPCOM_MEM_BLOAT_LOG 1
  script
  apprunner -edit
  (typed a few characters, did Debug->Output HTML, then quit)
  ^D (to end the script)

and said:

   We're leaking 1023 nsCSSSelectors and 12 CSSStyleSheetImpls.

Pierre, can you investigate.

[PTD team: we don't know the extent of the memory bloat due to this yet. We
just know it's unnecessary.]

Comment 1

[Pierre Saslawsky]

Attachment: LEAK AND BLOAT STATISTICS: Win32 console output

Comment 2

[Pierre Saslawsky]

The style leaks are the result of leaks at a higher level. As you can see in the
attached Win32 console output, we are leaking the following objects:

     object               total leaked
     ------               ------------
     nsDocument              5
     DocumentViewerImpl     15
     nsWebShellWindow        3
     nsWebShell              6
     PresShell               7
     nsPresContext           7

We don't have a clear owner for nsDocument or nsWebShell so I'm humbly submitting
this to beard who's got the best tools in the Valley to do that kind of work.

Changed the summary line to "leaking documents and webshells" from "leaking
nsCSSSelectors and CSSStyleSheetImpls".

Comment 3

[Pierre Saslawsky]

*** Bug 17389 has been marked as a duplicate of this bug. ***

Comment 4

[Patrick C. Beard]

I don't see these same leaks, but I see others. Leak detection is a
moving target.

Comment 5

[Patrick C. Beard]

Style person should follow up on these leaks. They are evidently shutdown leaks, 
as they don't appear as transient leaks with the Boehm leak detector.

Comment 6

[gordon]

On Mac, just starting up and shutting down leaks these, which prevent the 
nsIOService (and related services) from shutting down properly.

                       Total      Rem
CSSLoaderImpl             18        5
CSSParserImpl             90        7
CSSStyleSheetImpl         92       91
CSSStyleSheetInner        54       53

...but I don't see the same problem on WindowsNT.

Comment 7

[Pierre Saslawsky]

CCd attinasi. Gordon says that it only happens in the Mac but it may be related 
to the other XP leaks that Marc has on his list.

Comment 8

[Pierre Saslawsky]

Really CCd attinasi...

Comment 9

[Pierre Saslawsky]

How come this is still on my list?
Trying for the 3rd time to reassign to attinasi...

Comment 10

[Marc Attinasi]

This bug has been marked "future" because the original netscape engineer working 
on this is over-burdened. If you feel this is an error, that you or another 
known resource will be working on this bug,or if it blocks your work in some way 
-- please attach your concern to the bug for reconsideration.

Comment 11

[Warren Harris]

=> TK at AOL

Comment 12

[Warren Harris]

reassigning to dbaron

Comment 13

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

I believe the fix I attached yesterday to bug 24645 should fix this bug.  The 
root of the leak was the XUL keybinding document table, which was intentionally 
leaked to fix bug 27739.  However, just releasing the table was not sufficient 
because nobody was breaking the circular references between the keybinding 
documents (nsXULDocument objects) and their script objects, which would explain 
the state of this bug before.  (Note that there were 3 XULDocumentImpl leaked in 
pierre's original bloat log attached above.  This is what I saw before breaking 
the circular references.)

Comment 14

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

Proposed patch is:
http://bugzilla.mozilla.org/showattachment.cgi?attach_id=11213

Comment 15

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

Actually, this doesn't fix all the leaks when starting with mozilla -editor.  
I'm working on this new leak.  In fact, I'm writing a tutorial by writing down 
exactly what I do to find it.  It's going to be an interesting tutorial...

Comment 16

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

Part or all of the remaining leak has something to do with DocumentType objects
not having their script objects un-rooted when the
nsDocument::SetScriptGlobalObject(null) is called.  This function needs to call
SetDocument on all the elements in mProlog and mEpilog instead of just on
mRootContent.  Probably it needs to do the same elsewhere.

I'm not sure if that will fix the whole thing.  There may be a problem with
SetOpenerWindow too...

Comment 17

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

Attachment: patch to fix the "./mozilla -editor" startup leak

Comment 18

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

Attachment: revised patch

Comment 19

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

Attachment: revised patch, cleaning up a bunch of other issues in the same file

Comment 20

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

Attachment: also needed

Comment 21

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

Attachment: a diff -uw (ignoring whitespace) of that patch "also needed"

Comment 22

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

I checked in the keybindings table / keybindings document part of the fix.

Comment 23

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

Was unable to check in due to cvs lock in layout/html/style/src.  Will try again
another time.

Comment 24

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

Fix checked in, 2000-08-06 20:03 PDT.

Comment 25

[Stuart Parmenter]

*** Bug 42970 has been marked as a duplicate of this bug. ***

Comment 26

[shrirang khanzode]

verified that the patches are in. Marking as such.

Comment 27

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

*** Bug 39855 has been marked as a duplicate of this bug. ***