Introduce the ANNOTATE_LEAKING_OBJECT_PTR annotation that can be used to mark

Introduce the ANNOTATE_LEAKING_OBJECT_PTR annotation that can be used to mark
heap allocated objects as intentionally leaked ones.
Annotate the histograms produced by {Histogram,BooleanHistogram,LinearHistogram,CustomHistogram}::FactoryGet(), as leaked.

Rename StatsHistogram::StatsHistogramFactoryGet to StatsHistogram::FactoryGet, annotate the result as leaky, update the suppressions.

BUG=79322
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=87278

[Alexander Potapenko, 2011-05-27 11:24:08]

Please take a look.

Two differences between this and http://codereview.chromium.org/6869009:
 -- each histogram object is ignored only once, as demanded by IgnoreObject
documentation
 -- if heapchecker is disabled, the macros expands to (void)0 in order to avoid
the compiler errors about missing if-bodies.

[commit-bot: I haz the power, 2011-05-27 12:32:57]

No LGTM from valid reviewers yet.

[jar (doing other things), 2011-05-27 18:13:01]

See comments below.

http://codereview.chromium.org/7071036/diff/18/base/metrics/histogram.cc
File base/metrics/histogram.cc (right):

http://codereview.chromium.org/7071036/diff/18/base/metrics/histogram.cc#newc...
base/metrics/histogram.cc:105: ANNOTATE_LEAKING_OBJECT_PTR(histogram);  // see
crbug.com/79322
Perchance this line should appear only once, inside of
RegisterOrDeleteDuplicate().

This will also mean that you won't have to make so many changes in the patern
outside this file.

http://codereview.chromium.org/7071036/diff/18/net/disk_cache/stats_histogram.h
File net/disk_cache/stats_histogram.h (right):

http://codereview.chromium.org/7071036/diff/18/net/disk_cache/stats_histogram...
net/disk_cache/stats_histogram.h:38: static StatsHistogram* FactoryGet(const
std::string& name);
I like this change, and consider it cleanup (to more consistently use an
existing pattern)... so  I don't mind if you do this even if you don't change
the function.

[Alexander Potapenko, 2011-05-30 08:00:38]

http://codereview.chromium.org/7071036/diff/18/base/metrics/histogram.cc
File base/metrics/histogram.cc (right):

http://codereview.chromium.org/7071036/diff/18/base/metrics/histogram.cc#newc...
base/metrics/histogram.cc:105: ANNOTATE_LEAKING_OBJECT_PTR(histogram);  // see
crbug.com/79322
Indeed. I've forgotten to upload the actual diff for this file that compared
|histogram| to |tentative_histogram| in order to tell whether the object needs
to be annotated.
Anyway, adding the annotation to RegisterOrDeleteDuplicate() is more simple and
clear. Please see the updated patch.

[Alexander Potapenko, 2011-05-30 08:25:44]

(See Patch Set 4)
Some tests call RegisterOrDeleteDuplicate with (lock_ == NULL) ||
(!histograms_), so we need to annotate the corresponding return statements, then
cross the fingers and hope nobody is trying to register the same histogram
twice.

I've proposed a fix to google-perftools that allows annotating an object twice,
so we hopefully won't need to check this so strictly soon.

[jar (doing other things), 2011-05-30 17:07:51]

LGTM