On Tue, 18 Jun 2019 at 10:39, Tom Lane <tgl@xxxxxxxxxxxxx> wrote:
Tim Cross <theophilusx@xxxxxxxxx> writes:
> On Tue, 18 Jun 2019 at 09:34, Ken Tanzer <ken.tanzer@xxxxxxxxx> wrote:
>> Thanks Adrian, though I wasn't really seeking tips for column names. I
>> was instead trying to understand whether this particular tab expansion was
>> intentional and considered useful, and if so what that usefulness was,
>> because it's rather escaping me!
> Have to say, I fid that behaviour unusual as well.
I don't think it's intentional. A look into tab-complete.c shows that it
makes no attempt to offer completions beyond the "=" part of the syntax;
so there's room for improvement there. But then what is producing the
"DEFAULT" completion? After looking around a bit, I think it's
accidentally matching the pattern for a GUC "set" command:
else if (TailMatches("SET", MatchAny, "TO|="))
{
/* special cased code for individual GUCs */
...
else
COMPLETE_WITH("DEFAULT");
}
So perhaps that needs to look more like this other place where somebody
already noticed the conflict against UPDATE:
else if (TailMatches("SET|RESET") && !TailMatches("UPDATE", MatchAny, "SET"))
COMPLETE_WITH_QUERY(Query_for_list_of_set_vars);
More generally, though, I'm inclined to think that offering DEFAULT
and nothing else, which is what this code does if it doesn't recognize
the "GUC name", is just ridiculous. If the word after SET is not a known
GUC name then we probably have misconstrued the context, as indeed is
happening in your example; and in any case DEFAULT is about the least
likely thing for somebody to be trying to enter here. (They'd probably
have selected RESET not SET if they were trying to do that.)
regards, tom lane
Given that without adding a full blown sql parser in order to identify legitimate candidates following a '=' in an update statement, my suggestion would be to refine the rules so that no completion is attempted after the =. Would rather have tab do nothing over tab replacing what I've already typed with 'default'.
regards,
Tim
--
Tim Cross