Re: Support tab completion for upper character inputs in psql

"tanghy(dot)fnst(at)fujitsu(dot)com" <tanghy(dot)fnst(at)fujitsu(dot)com> writes:
> I did some tests on it and here are something cases I feel we need to confirm
> whether they are suitable.

> 1) postgres=# create table atest(id int, "iD" int, "ID" int);
> 2) CREATE TABLE
> 3) postgres=# alter table atest rename i[TAB]
> 4) id "iD"
> 5) postgres=# alter table atest rename I[TAB]
> 6) id "iD"

> The tab completion for 5) ignored "ID", is that correct?

Perhaps I misunderstood your original complaint, but what I thought
you were unhappy about was that unquoted ID is a legal spelling of
"id" and so I<TAB> ought to be willing to complete that. These
examples with case variants of the same word are of some interest,
but people aren't really going to create tables with these sorts of
names, so we shouldn't let them drive the design IMO.

Anyway, the existing behavior for these examples is

alter table atest rename i<TAB> --- completes immediately to id
alter table atest rename I<TAB> --- offers nothing

It's certainly arguable that the first case is right as-is and we
shouldn't change it. I think that could be handled by tweaking my
patch so that it wouldn't offer completions that start with a quote
unless the input word does. That would also cause I<TAB> to complete
immediately to id, which is arguably fine.

> I think what we are trying to do is to ease the burden of typing double quote for user.

I'm not thinking about it that way at all. To me, the goal is to make
tab completion do something sensible when presented with legal variant
spellings of a word. The two cases where it currently fails to do
that are (1) unquoted input that needs to be downcased, and (2) input
that is quoted when it doesn't strictly need to be.

To the extent that we can supply a required quote that the user
failed to type, that's fine, but it's not a primary goal of the patch.
Examples like these make me question whether it's even something we
want; it's resulting in extraneous matches that people might find more
annoying than helpful. Now I *think* that these aren't realistic
cases and that in real cases adding quotes will be helpful more often
than not, but it's debatable.

> One the other hand, I'm not so comfortable with the output of "iD" in line 13.
> If user doesn't type double quote, why we add double quote to the output?

That's certainly a valid argument.

> Could we make the output of 13) like below?
> 12) postgres=# alter table atest rename i[TAB]
> ??) id iD

That doesn't seem sensible at all.

regards, tom lane