Port last remaining test case in base/stats_table_unittest.cc, and...

Port last remaining test case in base/stats_table_unittest.cc, and
fix the bug it exposes on posix in StatsTable, i.e.
UnregisterThread() breaks when called inside SlotReturnFunction()
on posix because posix clears tls data before calling destructor.


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=4309

[dank, 2008-10-31 17:42:52]

Whaddaya know, I ported the last test case, and it found
a real bug.

Some of the functions needed for the test aren't
available on Mac yet, so it's disabled there.

[dank, 2008-10-31 17:47:39]

Sent to wrong address at first, sorry.

[Evan Martin, 2008-10-31 17:55:59]

Style nit: You've gone over 80 columns in a few places.  You can see them on the
rietveld side-by-side diff because it loses a line number.

http://codereview.chromium.org/8751/diff/8/10
File base/stats_table.h (right):

http://codereview.chromium.org/8751/diff/8/10#newcode88
Line 88: void UnregisterThread();
Maybe we can just eliminate this version?
It appears that stats_table.cc is the only caller of it.

[Evan Martin, 2008-10-31 19:17:13]

LGTM

(Would you mind putting some note about what the bug is in the changelog
description?  It'll help prevent people from "refactoring" the bug back into
existence.)

http://codereview.chromium.org/8751/diff/18/214
File base/stats_table.h (right):

http://codereview.chromium.org/8751/diff/18/214#newcode142
Line 142: void UnregisterThread(StatsTableTLSData *tls_data);
nit: star on the left.

[Dean McNamee, 2008-10-31 19:20:37]

http://codereview.chromium.org/8751/diff/18/213
File base/stats_table.cc (right):

http://codereview.chromium.org/8751/diff/18/213#newcode346
Line 346: 
Is this because you can no longer access the TLS key when you get the callback
on posix?  Can you make a note of that?

http://codereview.chromium.org/8751/diff/18/215
File base/stats_table_unittest.cc (right):

http://codereview.chromium.org/8751/diff/18/215#newcode69
Line 69: class StatsTableThread : public PlatformThread::Delegate {
Can you use SimpleThread, it was designed for unittests so we have a layer
between PlatformThread.