Thread: Avoid is possible a expensive function call (src/backend/utils/adt/varlena.c)

On Mon, 4 Mar 2024 at 18:39, Ranier Vilela <ranier.vf@gmail.com> wrote:
>
> Hi,
>
> The function var_strcmp is a critical function.
> Inside the function, there is a shortcut condition,
> which allows for a quick exit.
>
> Unfortunately, the current code calls a very expensive function beforehand, which if the test was true, all the call
timeis wasted.
 
> So, IMO, it's better to postpone the function call until when it is actually necessary.

Thank you for your contribution.

I agree it would be better, but your current patch is incorrect,
because we need to check if the user has access to the locale (and
throw an error if not) before we return that the two strings are
equal.

Kind regards,

Matthias van de Meent
Neon (https://neon.tech)

On Mon, Mar 04, 2024 at 03:08:03PM -0300, Ranier Vilela wrote:
> I can't see any user validation at the function pg_newlocale_from_collation.

Matthias is right, look closer.  I can see more than one check,
especially note the one related to the collation version mismatch that
should not be silently ignored.

> meson test pass all checks.

Collations are harder to test because they depend on the environment
where the test happens, especially with ICU.
--
Michael

Michael Paquier <michael@paquier.xyz> writes:
> On Mon, Mar 04, 2024 at 03:08:03PM -0300, Ranier Vilela wrote:
>> I can't see any user validation at the function pg_newlocale_from_collation.

> Matthias is right, look closer.  I can see more than one check,
> especially note the one related to the collation version mismatch that
> should not be silently ignored.

The fast path through that code doesn't include any checks, true,
but the point is that finding the entry proves that we made those
checks previously.  I can't agree with making those semantics
squishy in order to save a few cycles in the exact-equality case.

            regards, tom lane