Thread
Commits
-
Reject "23:59:60.nnn" in datetime input.
- a9632830bb05 13.0 landed
- b2c64f571db4 10.14 landed
- a958b07bc453 12.4 landed
- 6490376e56b0 11.9 landed
-
time values past 24:00:00 (or rather 23:59:60)
Christoph Berg <myon@debian.org> — 2020-05-20T12:58:07Z
This is ok: # select '23:59:60'::time; time ────────── 24:00:00 # select '23:59:61'::time; ERROR: 22008: date/time field value out of range: "23:59:61" This is not ok: # select '23:59:60.999'::time; time ────────────── 24:00:00.999 -- value isn't read back: # select '24:00:00.999'::time; ERROR: 22008: date/time field value out of range: "24:00:00.999" -- we can even go to :01 # select '23:59:60.9999999'::time; time ────────── 24:00:01 PG 12.3. Christoph -
Re: time values past 24:00:00 (or rather 23:59:60)
Tom Lane <tgl@sss.pgh.pa.us> — 2020-05-21T05:53:01Z
Christoph Berg <myon@debian.org> writes: > This is not ok: > # select '23:59:60.999'::time; > time > -------------- > 24:00:00.999 Yeah. Here's a draft patch for that (sans any regression tests as yet). I found several places with largely the same logic, so I tried to centralize the tests; although it turns out we need a minimum of two versions, because some places deal with fractional seconds differently. The places using float seconds had their own bugs: one forgot to worry about NaN, and neither was being careful about imprecise input. It's slightly annoying to have the check functions duplicate the calculation that some level of caller is going to do, but avoiding that would take more refactoring than I think is warranted. In practice a few extra multiplications aren't likely to be noticeable here. regards, tom lane
-
Re: time values past 24:00:00 (or rather 23:59:60)
Tom Lane <tgl@sss.pgh.pa.us> — 2020-06-04T20:54:34Z
I wrote: > Christoph Berg <myon@debian.org> writes: >> This is not ok: >> # select '23:59:60.999'::time; >> time >> -------------- >> 24:00:00.999 > Yeah. Here's a draft patch for that (sans any regression tests as yet). I pushed this fix, but only as far back as v10. To apply it in the 9.x branches, we'd need to make the code support float timestamps, and I didn't think it was worth the trouble. regards, tom lane
-
Re: time values past 24:00:00 (or rather 23:59:60)
Christoph Berg <myon@debian.org> — 2020-06-05T10:27:43Z
Re: Tom Lane > I pushed this fix, but only as far back as v10. To apply it in the > 9.x branches, we'd need to make the code support float timestamps, > and I didn't think it was worth the trouble. Aye. Thanks! Christoph