392118 - port trace-malloc to Mac OS X

Status: RESOLVED FIXED

(Core :: General, defect)

Description

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

I have patches to get trace-malloc working on Mac OS X, modulo the Mac OS X stack walking code in XPCOM, which I don't have working yet.

It was pretty easy; I checked that the thread-local storage stuff using pthreads seem to think we have a new thread at about the right places, and the nsGetTypeName code even works.

The malloc-hooking required a quick look at Apple's APSL-ed Libc code to find out what the hooks were, since they're not documented in the headers.

Comment 1

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Attachment: patch 1: trace-malloc port to Mac OS X

Comment 2

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Attachment: patch 2: build system patch, part 1: fix bad dependency

So I hit a bad dependency rule that I think nobody's hit yet.  I'm not even sure the rule is necessary; the correct solution may be to remove it.  But instead I made it work, I think, although I'm not sure how portably.

The problem was that EXTRA_DSO_LIBS is not a sensible thing to append to DSO_LDOPS_DEPS; there's no file called "-ltracemalloc".  To work around that, I made an EXTRA_DSO_LIB_FILES variable.  But then, the makefile where I was having the problem happened to override DLL_PREFIX and DLL_SUFFIX, which my definition of EXTRA_DSO_LIB_FILES depended on, so I gave that makefile a different way to do what it was doing (IS_FRAMEWORK).

I'm not sure if EXTRA_DSO_LIBS are always shared libraries, though.  In fact, I'm pretty sure they're not, which means this patch is wrong.  But I'm not sure what we *do* want to do here.  Advice?

Comment 3

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Patch 3 isn't written yet, but I need to reorganize tools/trace-malloc/Makefile.in somehow so that it links leaksoup with the C++ compiler rather than the C compiler.  That may involve moving spacetrace into its own subdirectory.

Comment 4

[Brendan Eich [:brendan]]

Comment on attachment 276564 [details] [diff] [review]
patch 1: trace-malloc port to Mac OS X

The declarator name in the function typedef doesn't need parens around it, but I guess that's how the hidden Apple source wrote it.

r/a=me.

/be

Comment 5

[Benjamin Smedberg]

GNU make can calculate dependencies of the form -lfoo, kinda. It uses the VPATH to determine where to look, which is IMO not ideal.

I'll look at this more tomorrow when I can concentrate.

Comment 6

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Patch 1 landed on trunk 2007-08-14 09:35 -0700.

Comment 7

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Attachment: patch 2: build system patch, part 1: fix bad dependency

Right, it turns out that the half of the patch that I wrote second is sufficient on its own.  Whatever makes -ltracemalloc work requires DLL_PREFIX and DLL_SUFFIX (or at least one of them) to be set correctly.

Comment 8

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Attachment: patch 3: build system patch, part 2: set CPP_PROG_LINK

This is the other half of the required build system changes: this makefile is pretty weird already, and this minimal fix makes us use g++ as the linker, which is needed for leaksoup.  rules.mk sets CPP_PROG_LINK if CPPSRCS or CMMSRCS are set; we're setting SIMPLE_CPPSRCS instead.

Comment 9

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 276984 [details] [diff] [review]
patch 2: build system patch, part 1: fix bad dependency

a=bzbarsky

Comment 10

[Boris Zbarsky [:bzbarsky]]

Comment on attachment 276986 [details] [diff] [review]
patch 3: build system patch, part 2: set CPP_PROG_LINK

a=bzbarsky

Comment 11

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Both build system patches checked in, 2007-08-19 17:00 -0700.