On Wed, Jan 04, 2023 at 04:18:59PM -0500, Andrew Dunstan wrote: > On 2023-01-02 Mo 10:44, Tom Lane wrote: >> I don't think that just concatenating those strings would make for a >> pleasant API. More sensible, perhaps, to have a separate function >> that returns a record. Or we could redefine the existing function >> that way, but I suspect that "just the primary error" will be a >> principal use-case. >> >> Being able to get the SQLSTATE is likely to be interesting too. > > OK, here's a patch along those lines.
My vote would be to redefine the existing pg_input_error_message() function to return a record, but I recognize that this would inflate the patch quite a bit due to all the existing uses in the tests. If this is the only argument against this approach, I'm happy to help with the patch. -- Nathan Bossart Amazon Web Services: https://aws.amazon.com