Re: IF (NOT) EXISTS in psql-completion

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> Yeah, maybe, but we'd need a committer to take more of an interest in
> this patch series. Personally, I'm wondering why we need a series of
> 19 patches to add tab completion support for IF NOT EXISTS. The
> feature which is the subject of this thread arrives in patch 0017, and
> a lot of the patches which come before that seem to change a lot of
> stuff without actually improving much that would really benefit users.

FWIW, one reason this committer hasn't jumped in is that we already
rewrote tab-complete.c pretty completely in 9.6. If we accept a patch
that completely rewrites it again, we're going to be faced with
maintaining three fundamentally different implementations for the next
three-plus years (until 9.5 dies). Admittedly, we don't back-patch
fixes in tab-complete.c every week, but a look at the git history says
we do need to do that several times a year.

Also, the nature of the primary refactoring (changing the big else-chain
into standalone ifs, if I read it correctly) is particularly bad from a
back-patching standpoint because all you have to do is insert an "else",
or fail to insert one, to silently and almost completely break either
one or the other branch. And I don't really understand why that's a good
idea anyway: surely we can return at most one set of completions, so how
is turning the function into a lot of independent actions a win?

So I'd be a whole lot happier if it didn't do that. Can we really not
add the desired features in a more localized fashion?

regards, tom lane