Check for failed memory allocations in libpq

Dave Allen <daveallen@acm.org> writes:

Attached is a patch (against 7.3.4) to check the return values of some
calls (malloc, realloc, etc.) for failed memory allocations in libpq.

You sure you aren't just trading one misbehavior for another? The
change in PQmakeEmptyPGresult, for example, just moves the core dump
somewhere else, unless you find reasonable fallbacks for *all* its
callers (including applications you don't have the source code for,
but in any case including every one of the calls in libpq).

regards, tom lane

Apologies if I missed anything, but I thought I fixed any callers of
PQmakeEmptyPQresult that weren't already checking (parseInput and
getRowDescriptions were the only ones).

I of course can't fix any applications I don't have the source for, but
if I missed something in libpq, I'd be more than happy to go back and
try to fix it.

On Thu, 2003-07-31 at 07:05, Tom Lane wrote:

Dave Allen <daveallen@acm.org> writes:

Attached is a patch (against 7.3.4) to check the return values of some
calls (malloc, realloc, etc.) for failed memory allocations in libpq.

You sure you aren't just trading one misbehavior for another? The
change in PQmakeEmptyPGresult, for example, just moves the core dump
somewhere else, unless you find reasonable fallbacks for *all* its
callers (including applications you don't have the source code for,
but in any case including every one of the calls in libpq).

regards, tom lane

---------------------------(end of broadcast)---------------------------
TIP 5: Have you checked our extensive FAQ?

http://www.postgresql.org/docs/faqs/FAQ.html

--
Dave Allen
daveallen@acm.org

Dave Allen <daveallen@acm.org> writes:

You sure you aren't just trading one misbehavior for another?

Apologies if I missed anything, but I thought I fixed any callers of
PQmakeEmptyPQresult that weren't already checking (parseInput and
getRowDescriptions were the only ones).

Well, I'm concerned about the global implications. For instance, I
think making parseInput simply return on malloc failure *without doing
any more* will convert out-of-memory from a core dump into an infinite
loop; which is hardly an improvement. (See PQgetResult for one example
of a place that will loop till it gets something.)

I'd be the first to admit that libpq's error handling isn't very good,
but I think fixing it will take some wholesale rework of the internal
calling conventions, not a few lines of localized patches.

Having said that, though, CVS tip does have some progress on this
matter; it will usually behave reasonably when it runs out of memory for
a query result. If you'd like to pursue the matter, please pull down
CVS tip and see what you can do with it. The patch as given would not
be much use to us anyway because of the changes since 7.3...

regards, tom lane