Re: Summary: changes needed in function defaults behavior - Mailing list pgsql-hackers

From Tom Lane
Subject Re: Summary: changes needed in function defaults behavior
Date
Msg-id 12263.1229554295@sss.pgh.pa.us
Whole thread Raw
In response to Summary: changes needed in function defaults behavior  (Tom Lane <tgl@sss.pgh.pa.us>)
Responses Re: Summary: changes needed in function defaults behavior  ("Pavel Stehule" <pavel.stehule@gmail.com>)
Re: Summary: changes needed in function defaults behavior  ("Asko Oja" <ascoja@gmail.com>)
List pgsql-hackers
I wrote:
> * Two functions that could match a given call after adding defaults
> are considered ambiguous only if they would add the same number of
> defaults; otherwise we prefer the one with fewer parameters.  This
> generalizes the rule that an exact match (no defaults) is preferred
> over one that requires adding defaults.

Experimenting with the revised code, I found a curious case that might
be worth worrying about.  Consider the example that started all this:

create function foo(f1 int, f2 int = 42, f3 int = 43) ...
create view v1 as select foo(11);

The patch I've got correctly reverse-lists v1 as "select foo(11)".
Now suppose we add

create function foo(f1 int, f2 int = 42) ...

or even

create function foo(f1 int) ...

The view is still gonna reverse-list as "select foo(11)" --- in fact,
we really haven't got much choice about that.  However, if dumped and
reloaded along with one of these shorter-argument-list functions, the
view will be reconstituted as a reference to the shorter function instead
of the original 3-argument function.

I'm not sure how critical this is, since you'd have to be pretty dumb to
put together a set of functions like this that didn't work compatibly.
Still, this is the first instance I know of in which dump/reload isn't
going to be guaranteed to match the same function as was being called
in the dumped database.

If we think this is critical enough to be worth sacrificing something
for, what I'd suggest is that we abandon the concept that shorter
argument lists are allowed to win over longer ones.  This would mean
that
foo(f1)foo(f1 int, f2 int = 42)foo(f1 int, f2 int = 42, f3 int = 43)

would all be considered equally good matches for a call foo(11)
and so you'd get an "ambiguous function" failure.  While that doesn't
prevent you getting into this sort of trouble, what it would do is
ensure that the dump reload gives an error instead of silently picking
the wrong function.  Also, you'd most likely have gotten a few failures
and thus been shown the error of your ways before you dumped the old
DB at all.

Thoughts?
        regards, tom lane


pgsql-hackers by date:

Previous
From: Teodor Sigaev
Date:
Subject: Re: Review: B-Tree emulation for GIN
Next
From: Simon Riggs
Date:
Subject: Re: Preventing index scans for non-recoverable index AMs