Re: snprintf.c hammering memset()

Andres Freund <andres(at)anarazel(dot)de> writes:
> On 2018-10-01 20:19:16 -0400, Tom Lane wrote:
>> That patch takes the memset out of the main line, but it'd still be
>> a performance problem for formats using argument reordering; and the
>> stack-space concern would remain the same.

> What I mean is that it shouldn't be that hard to only zero out the
> portions of the array that are actually used, and thus could refrain
> from introducing the limit.

Well, we use the zeroing exactly to detect which entries have been used.
Probably there's another way, but I doubt it'd be faster.

In any case, the stack-space-consumption problem remains, and IMO that
is a *far* greater concern than the cycles. Keep in mind that we'll be
calling this code when we have already hit our stack space consumption
limit. I'm a bit surprised that we've not seen any buildfarm members
fall over in the error recursion test. We have not designed the system
on the assumption that ereport() could eat half a meg of stack.

regards, tom lane