480984 - Naked uses of alloca()

Status: VERIFIED FIXED

(Tamarin Graveyard :: Garbage Collection (mmGC), defect, P3)

Description

[Tommy Reilly]

After Rishit is done make sure all OS calls go through VMPI layer

There are calls to alloca in GCHeap::DumpHeapRep, but that's not portable and
the function is not protected by any kind of #ifdef DESKTOP_DEBUGGING (for
instance).

Comment 1

[Lars T Hansen]

Hopefully the removal of fprintf will also fix the problem with wrong-size arguments being passed to fprintf (ie, size_t is sometimes 32 and sometimes 64 bits, so it's never correct to use %u or %lu or %llu -- only a cast to a specific size is guaranteed to work).  GCC spews copious warnings for this problem in the new GC code.

Comment 2

[Lars T Hansen]

At least three uses of alloca(), one in GCMemoryProfiler and two in GCHeap.

I've spotted memset() or memcpy() or similar calls, but won't try to find them now.

We can probably machine-search the code for these; if a good script can be found it can become part of the build system.

Comment 3

[Lars T Hansen]

This has too many false positives and doesn't capture everything, but it does filter enough for a quick scan; I can't find any other offenders in the code as it is now (except for alloca):

for n in $(egrep '# *define +VMPI_' platform/mac/mac-platform.h | awk '{ print substr($2,6) }'); do find . -name '*.h' -o -name '*.cpp' | xargs grep '[^_A-Za-z0-9]'$n'[^(a-zA-Z0-9_]' | grep -v '# *define' ; done

Comment 4

[Lars T Hansen]

Attachment: Removes all uses of alloca in core code

Comment 5

[Lars T Hansen]

redux changeset:   3001:14fa373461cd

Comment 6

[Chris Peyer]

Engineering work item.  Marking as verified.