On Tue, 2023-11-21 at 09:24 -0500, Robert Haas wrote:
> As to the first, could we
> remove the setjmp() and instead have abort-time processing know
> something about this? For example, imagine we just push something
> onto
> a stack like we do for ErrorContextCallback, do whatever, and then
> pop
> it off. But if an error is thrown then the abort path knows to look
> at
> that variable and do whatever.
If I remove the TRY/CATCH entirely, it shows there's room for ~200ms
improvement in my test.
I attached a rough patch, which doesn't quite achieve that much, it's
more like ~100ms improvement and starts to fall within the noise. So
perhaps an improvement, but a bit disappointing. It's not a lot of
code, but it's not trivial either because the nesting level needs to be
tracked (so a subxact abort doesn't reset too much state).
Also, it's not quite as clean as it could be, because I went to some
effort to avoid an alloc/free by keeping the stack within the fcache. I
didn't pay a lot of attention to correctness in this particular patch;
I was mostly trying a few different formulations for performance
measurement.
I'm not inclined to commit this in its current form but if someone
thinks that it's a worthwhile direction, I can clean it up a bit and
reconsider.
Regards,
Jeff Davis