Re: pg_dump / copy bugs with "big lines" ?

Alvaro Herrera wrote:

> I have now pushed this to 9.5, 9.6 and master. It could be backpatched
> to 9.4 with ease (just a small change in heap_form_tuple); anything
> further back would require much more effort.
>
> I used a 32-bit limit using sizeof(int32). I tested and all the
> mentioned cases seem to work sanely; if you can spare some more time to
> test what was committed, I'd appreciate it.

My tests are OK too but I see an issue with the code in
enlargeStringInfo(), regarding integer overflow.
The bit of comment that says:

Note we are assuming here that limit <= INT_MAX/2, else the above
loop could overflow.

is obsolete, it's now INT_MAX instead of INT_MAX/2.

There's a related problem here:
newlen = 2 * str->maxlen;
while (needed > newlen)
newlen = 2 * newlen;
str->maxlen is an int going up to INT_MAX so [2 * str->maxlen] now
*will* overflow when [str->maxlen > INT_MAX/2].
Eventually it somehow works because of this:
if (newlen > limit)
newlen = limit;
but newlen is wonky (when resulting from int overflow)
before being brought back to limit.

PFA a minimal fix.

Best regards,
--
Daniel Vérité
PostgreSQL-powered mailer: http://www.manitou-mail.org
Twitter: @DanielVerite