702984 - CSS Rulehash memory reporter may undercount memory usage

Status: RESOLVED INVALID

(Core :: CSS Parsing and Computation, defect)

Description

[Justin Lebar (not reading bugmail)]

In bug 702301 comment 9, we saw the following DMD output:

==24692== Unreported: 3,145,728 (cumulative: 58,999,300) bytes in 16 heap block(s) in record 5 of 34899:
==24692==  Requested bytes unreported: 3,145,728 / 3,145,728
==24692==  Slop      bytes unreported: 0 / 0
==24692==    at 0x4C276AA: malloc (vg_replace_malloc.c:263)
==24692==    by 0x87A51C6: ChangeTable(PLDHashTable*, int) (pldhash.cpp:565)
==24692==    by 0x87A5657: PL_DHashTableOperate (pldhash.cpp:650)
==24692==    by 0x800EB77: RuleHash::AppendRuleToTable(PLDHashTable*, void const*, RuleSelectorPair const&) (nsCSSRuleProcessor.cpp:563)
==24692==    by 0x800EBE1: RuleHash::AppendRule(RuleSelectorPair const&) (nsCSSRuleProcessor.cpp:595)
==24692==    by 0x8011212: nsCSSRuleProcessor::RefreshRuleCascade(nsPresContext*) (nsCSSRuleProcessor.cpp:2784)
==24692==    by 0x8011386: nsCSSRuleProcessor::GetRuleCascade(nsPresContext*) (nsCSSRuleProcessor.cpp:3038)
==24692==    by 0x8011834: nsCSSRuleProcessor::RulesMatching(AnonBoxRuleProcessorData*) (nsCSSRuleProcessor.cpp:2307)
==24692==    by 0x80414C7: bool EnumRulesMatching<AnonBoxRuleProcessorData>(nsIStyleRuleProcessor*, void*) (nsStyleSet.cpp:477)
==24692==    by 0x80436C9: nsStyleSet::FileRules(bool (*)(nsIStyleRuleProcessor*, void*), void*, nsIContent*, nsRuleWalker*) (nsStyleSet.cpp:785)
==24692==    by 0x8043A66: nsStyleSet::ResolveAnonymousBoxStyle(nsIAtom*, nsStyleContext*) (nsStyleSet.cpp:1207)
==24692==    by 0x7F49514: nsCSSFrameConstructor::ConstructRootFrame(nsIFrame**) (nsCSSFrameConstructor.cpp:2541)

This is a bit strange since a memory reporter for this was added in bug 676048.

Nick, is it possible that the DMD patch just isn't aware of this memory reporter?

Comment 1

[Justin Lebar (not reading bugmail)]

Is nsCSSRuleProcessor::SizeOf ever called?  DXR says no, but that could be wrong.

http://dxr.mozilla.org/mozilla/mozilla-central/layout/style/nsCSSRuleProcessor.cpp.html#l2561

Comment 2

[Justin Lebar (not reading bugmail)]

Okay, yes, it's called.  I wonder if this is just a DMD false negative.

Comment 3

[Boris Zbarsky [:bzbarsky]]

Actually ccing Nick.

Comment 4

[Nicholas Nethercote [inactive]]

Yeah, the patch in bug 676724 telling DMD about memory reporters doesn't have annotations for style stuff.  I'm planning to convert all the existing reporters to use the new style from bug 698968, which will mean they don't need any annotations for DMD to know about them.

I'll assume this bug report is invalid for now, and if anything is missing after I convert I'll file new bugs.