573268 - Intentional XPT leak interferes with leak debugging

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Description

[Jesse Ruderman]

See bug 570488 comment 10 forward.  It should probably just be made not to leak, since that's what we do with everything else, but I'd settle for not leaking if compiled with { --enable-logrefcnt, --enable-valgrind, --enable-trace-malloc }.

Comment 1

[Benjamin Smedberg]

Attachment: Destroy the arena in trace-malloc builds, rev. 1

Comment 2

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

Comment on attachment 455465 [details] [diff] [review]
Destroy the arena in trace-malloc builds, rev. 1

>+#if defined(NS_TRACE_MALLOC)

defined(NS_TRACE_MALLOC) || defined(NS_BUILD_REFCNT_LOGGING)

>+#define NS_FREE_PERMANENT_DATA
>+#endif
>-    // Don't destroy the arena; we're shutting down, why touch the
>-    // memory when we don't have to?
>+#ifdef NS_FREE_PERMANENT_DATA
>+    XPT_DestroyArena(gXPTIStructData);
>+#endif
> }        

Maybe good to leave the comment, slightly reformulated?  (Or maybe, even better, add a little more explanation of the same above NS_FREE_PERMANENT_DATA to explain why it exists?)

r=dbaron with that

Comment 3

[Jesse Ruderman]

Also || defined(MOZ_VALGRIND)?

Comment 4

[Benjamin Smedberg]

http://hg.mozilla.org/mozilla-central/rev/fd1592c0dae1

Comment 5

[Daniel Holbert [:dholbert]]

Attachment: trivial followup: use c-style (not c++-style) comment in nscore.h

This patch added a //-style comment to nscore.h, which makes us spam build warnings for .c files that include nscore.h (e.g. pldhash.c)

Attached followup switches the comment to be /**/-style.

Comment 6

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

Comment on attachment 459926 [details] [diff] [review]
trivial followup: use c-style (not c++-style) comment in nscore.h

r=dbaron