On Sat, Apr 18, 2026 at 09:06:02PM +1200, David Rowley wrote:
> On Fri, 10 Oct 2025 at 11:30, Michael Paquier <michael@paquier.xyz> wrote:
> > Thanks for double-checking. Applied after running an indent.
>
> I was working on test_bitmapset.c to add some tests for a new
> bitmapset function. I noticed a few weird things.
>
> 1. test_random_operations() is coded to use GetCurrentTimestamp() as a
> seed when the given seed is <= 0. Of course, it'll be a while before
> the return value of that wraps beyond 2^63 (292250 years), but I still
> can't help but think that NULL is a better value to use to have the
> seed auto-generate.
Fine by me.
> 2. Doing #1 means the function can't be STRICT. I do think it's wrong
> that the function is marked as strict. That's normally reserved for
> functions that we needn't call because NULL input(s) yield a NULL
> output. That's not the case for this function.
Using the existing HEAD approach where STRICT avoids these extra NULL
checks, or adding explicit NULL checks without STRICT does not strike
me as a big difference in this context.
> 3. There's no CHECK_FOR_INTERRUPTS() in test_random_operations(). If
> someone uses a large num_ops, there's no way to cancel the query.
> 4. If there happened to be some rare bug in bitmapset.c that
> test_random_operations() we might struggle to find it again, as we
> don't report which seed we used in the ERROR message.
These make sense.
> I felt it was worth fixing these now as the function I plan to add
> there does #1, #2, #3 and #4. If I add the new function for v20, the
> discrepancy seems questionable.
It is a test module, it would be a big issue if new pieces are
backpatched in this area. In short I'm fine with these. Thanks for
asking.
--
Michael
