Re: pg_recvlogical fixes

On 2015-10-21 11:52:31 -0300, Euler Taveira wrote:
> While testing wal2json, I faced some problems with pg_recvlogical. Attached
> is a serie of patches that can improve pg_recvlogical. Patches #2 and #3 are
> bugfixes (and should be applied to 9.5 too). Patch #1 is not mandatory to
> 9.5.

> #1: add a bunch of checks to complain when using an option that is not
> available in the specified action;

I'm not a fan of doing that. Doing that for every option tends to be
more annoying than helpful. E.g. pgbench's checks requires you to
pointlessly remove a lot of harmless options just to be able to pass -i.

> #2: there is a wrong check because startpos option can be specified with
> --create-slot;

> #3: doesn't ignore startpos in --create-slot because that action could be
> specified together with --start action (that uses that option);

It doesn't make sense to specify it with --create-slot though - you
can't even know what a the start position would mean before the slot is
created. You can't get older records than the ones at slot creation
time, and you can't know what a feature xlog position would mean.

Greetings,

Andres Freund