Re: Some regular-expression performance hacking

Hi,

On 2021-02-23 12:52:28 -0500, Tom Lane wrote:
> I wrote:
> > Hmph. There's an "assert(depth >= 0)" immediately in front of that,
> > so I'm not looking too kindly on the compiler thinking it's smarter
> > than I am. Do you have a suggestion on how to shut it up?

gcc can't see the assert though, in an non-cassert optimized build... If
I force assertions to be used, the warning vanishes.

> On reflection, maybe the thing to do is convert the assert into
> an always-on check, "if (depth < 0) return false". The assertion
> is essentially saying that there's no arc leading directly from
> the pre state to the post state. Which there had better not be,
> or a lot of other stuff is going to go wrong; but I suppose there's
> no way to explain that to gcc. It is annoying to have to expend
> an always-on check for a can't-happen case, though.

Wouldn't quite work like that because of the restrictions of what pg
infrastructure we want to expose the regex engine to, but a
if (depth < 0)
pg_unreachable();
would avoid the runtime overhead and does fix the warning.

I have been wondering about making Asserts do something along those
lines - but it'd need to be opt-in, since we clearly have a lot of
assertions that would cost too much.

Greetings,

Andres Freund