Re: Finer grain log timestamps

On Mon, 20 Jun 2022 at 11:01, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

> Alvaro Herrera <alvherre(at)alvh(dot)no-ip(dot)org> writes:
>

> If I were coding it, I would allow only exactly 1 digit (%.Nt) to simplify
> the parsing side of things and bound the required buffer size. Without
> having written it, it's not clear to me whether further restricting the
> set of supported values would save much code. I will point out, though,
> that throwing an error during log_line_prefix processing will lead
> straight to infinite recursion.
>

I would parse the log_line_prefix when it is set. Then if there is a
problem you just log it using whatever format is in effect and don't change
the setting. Then the worst that happens is that logs show up in a format
log processing isn't prepared to accept.

That being said, I think I fall in the “just start putting more digits in
the log” camp, although it is conceivable the counter arguments might be
convincing.