Re: [HACKERS] PG10b2: column-list UPDATE syntax fails with single column - Mailing list pgsql-hackers

Justin Pryzby <pryzby@telsasoft.com> writes:
> ts=# begin; UPDATE eric_enodeb_201707 SET (pmhdelayvarbest50pct,pmlicconnecteduserstimecong)=(0,0) ;
> BEGIN
> UPDATE 3
> ts=# begin; UPDATE eric_enodeb_201707 SET (pmhdelayvarbest50pct)=(0) ;
> BEGIN
> ERROR:  source for a multiple-column UPDATE item must be a sub-SELECT or ROW() expression

Hm.  It's kind of unfortunate that this isn't backwards-compatible, but
it was only accidental that that case was accepted before.  It's really
wrong, because the source for a parenthesized UPDATE target list ought
to be a row value even when there's only one item in the list.  That
is, the correct standards-compliant spelling would be

UPDATE eric_enodeb_201707 SET (pmhdelayvarbest50pct) = ROW(0);

Now, it's true that "(1,2)" is fully equivalent to "ROW(1,2)", but
"(0)" is *not* equivalent to "ROW(0)"; it's just a scalar 0.  So your
existing code is non-spec-compliant and was really being accepted in
error.

We could maybe hack up some weird action-at-a-distance kluge that would
make this case work like before, but I'm afraid it would just introduce
other inconsistencies.

> It may be that our use was wrong (?) or unintuitive (I'm in the middle of
> changing it), but wondered if it was intentional or otherwise if the release
> notes should mention that old syntax is no longer accepted.

Not sure if it's worth getting into in the release notes.  Using the
parenthesis notation for single target columns seems rather weird.
        regards, tom lane