On 04.01.23 01:37, Nathan Bossart wrote:
> In general, +1.
>
> On Tue, Dec 20, 2022 at 06:30:40AM +0100, Peter Eisentraut wrote:
>> (Still need to think about Robert's comment about lack of error context.)
>
> Would adding the name of the GUC be sufficient?
>
> ereport(ERROR,
> (errmsg("could not build %s", guc_name),
> errdetail("string ends unexpectedly after escape character \"%%\"")));
done
The errors now basically look like an invalid GUC value.
>> + * A value may be NULL. If the corresponding placeholder is found in the
>> + * input string, the whole function returns NULL.
>
> This appears to be carried over from BuildRestoreCommand(), and AFAICT it
> is only necessary because pg_rewind doesn't support %r in restore_command.
> IMHO this behavior is counterintuitive and possibly error-prone and should
> result in an ERROR instead. Since pg_rewind is the only special case, it
> could independently check for %r before building the command.
Yeah, this annoyed me, too. I have now changed it so that a NULL
"value" is the same as an unsupported placeholder. This preserves the
existing behavior.
(Having pg_rewind check for %r itself would probably require replicating
most of the string processing logic (consider something like "%%r"), so
it didn't seem appealing.)