Re: legitimacy of using PG_TRY , PG_CATCH , PG_END_TRY in C function

On 2017-10-22 23:04:50 -0400, Tom Lane wrote:
> John Lumby <johnlumby(at)hotmail(dot)com> writes:
> > I have a C function (a trigger function) which uses the PG_TRY()
> > construct to handle certain ERROR conditions.
> > One example is where invoked as INSTEAD OF INSERT into a view. It
> > PG_TRYs INSERT into the real base table,
> > but this table may not yet exist (it is a partitioned child of an
> > inheritance parent).
> > If the error is ERRCODE_UNDEFINED_TABLE, then the CATCH issues
> > FlushErrorState() and returns to caller who CREATes the table and
> > re-issues the insert.
> > All works perfectly (on every release of 9.x).
>
> If it works, it's only because you didn't try very hard to break it.
> In general you can't catch and ignore errors without a full-fledged
> subtransaction --- BeginInternalSubTransaction, then either
> ReleaseCurrentSubTransaction or RollbackAndReleaseCurrentSubTransaction,
> not just FlushErrorState. See e.g. plpgpsql's exec_stmt_block.
>
> There may well be specific scenarios where an error gets thrown without
> having done anything that requires transaction cleanup. But when you
> hit a scenario where that's not true, or when a scenario that used to
> not require cleanup now does, nobody is going to consider that a PG bug.

It'd probably be a good idea to expand on this in the sgml docs. This
has confused quite anumber of people...

Greetings,

Andres Freund