Re: [PATCH] mtrace: properly handle realloc (p, 0)

[Roland McGrath <roland at hack dot frob dot com>]

You all have been thoroughly in the realm of standards-lawyering for a
while now.  Please do take that someplace else, where it might have
some useful effect.  Frankly, I've lost track of exactly what change
to libc's actual behavior is under consideration.  Just on principle
about standardizing existing practice, as standards actually did once
upon a time, we should push back at least moderately on any proposed
change to, or reinterpretation of, the ISO C or POSIX.1 standards that
would retroactively declare libc's existing practice to be
nonconformant.  But some change of behavior in a corner case that was
not previously well-specified may be acceptable to us for the benefit
of greater harmonization.  Paul is an excellent person to follow up on
that issue.

The original subject of this thread is quite straightforward and none
of this fritter about the standards is relevant to it at all.  The
change was to make mtrace's analysis of libc calls consistent with
what they actually do.  That's clearly a correct change and that's all
there is to it.  If the behavior of realloc changes in the future, for
whatever reason, then it will be appropriate and necessary to update
mtrace so it continues to accurately reflect the actual behavior.


Thanks,
Roland