Hello Tom,
>>>> Here is a POC which defines an internal interface for a PRNG, and use it
>>>> within pgbench, with several possible implementations which default to
>>>> rand48.
>
>>> I seriously dislike this patch. pgbench's random support is quite
>>> overengineered already IMO, and this proposes to add a whole batch of
>>> new code and new APIs to fix a very small bug.
>
>> My intention is rather to discuss postgres' PRNG, in passing. Full success
>> on this point:-)
>
> Our immediate problem is to fix a portability failure, which we need to
> back-patch into at least one released branch, ergo conservatism is
> warranted.
Sure, the patch I sent is definitely not for backpatching, it is for
discussion.
> I had in mind something more like the attached.
Yep.
I'm not too happy that it mixes API levels, and about the int/double/int
path.
Attached an updated version which relies on pg_jrand48 instead. Also, as
the pseudo-random state is fully controlled, seeded test results are
deterministic so the expected value can be fully checked.
I did a few sanity tests which were all ok.
I think that this version is appropriate for backpatching. I also think
that it would be appropriate to consider having a better PRNG to replace
rand48 in a future release.
--
Fabien.
