Fix realloc to not call the new_handler (which is for failures)...

Fix realloc to not call the new_handler (which is for failures)
when the request was for realloc(ptr, 0).  Calling realloc(ptr, 0)
is valid, and we should delete ptr and return NULL.

BUG=none
TEST=none

[jar (doing other things), 2009-09-08 16:03:52]

http://codereview.chromium.org/194040/diff/2001/2002
File third_party/tcmalloc/allocator_shim.cc (right):

http://codereview.chromium.org/194040/diff/2001/2002#newcode163
Line 163: // realloc should free the ptr and return zero if the requested new
size
nit: rather than "... and return zero if..." I'd prefer:

....and return NULL if ...

Also, the entire block is covering the subtle case, and hence perhaps even the
larger comment:

Subtle warning: NULL return doesn't always indicate out-of-memory.  If the
requested new size is zero, realloc should free the ptr and return NULL.

[jar (doing other things), 2009-09-08 16:04:31]

With the nit/comment adjusted as per previous review.... LGTM.