> On 13 Aug 2025, at 11:05, Erik Nordström <erik@tigerdata.com> wrote: > > I think I found a small typo in the function that extracts a timestamp from a UUIDv7 (uuid_extract_timestamp). Unless I am mistaken, the constant US_PER_MS should be used instead of NS_PER_US when converting milliseconds to microseconds. Fortunately, these constants are the same so the calculation is still correct.
Wow, that's a very good level of proofreading! Yes, you are correct, it's must be US_PER_MS.
> > Anyway, attaching a patch to fix this typo.
LGTM.
Ok, maybe a committer would like to commit this change immediately?
> > On a related note, I am wondering why this function doesn't extract and use the sub-millisecond information in the rand_a bits? These bits are added when generating the UUID, but they don't seem to be extracted. Hopefully somebody could shed some light on this and whether it would be a worthwhile addition.
UUID might be formed by any external system, rand_a bits are not guaranteed to be non-random. Well, in some sense, reading time is reading random number, but anyway, we don't know for sure how those bits are used. And even in Postgres usage of sub-millisecond fractions depends on OS.
I think we can have a function in an extensions, that does pretend that UUID was generated by known algorithm. Or even SQL function to extract microseconds.
