Thread: Enable integer datetimes by default
Attached is a patch that enables integer datetimes by default, per recent discussion on -hackers. It makes "--enable-integer-datetimes" the default, and documents the "--disable-integer-datetimes" configure option as a means to get the previous default behavior. Barring any objections, I'll apply this to HEAD tomorrow. -Neil
Attachment
On May 5, 2007, at 22:28 , Neil Conway wrote: > Attached is a patch that enables integer datetimes by default, per > recent discussion on -hackers. It makes "--enable-integer- > datetimes" the > default, and documents the "--disable-integer-datetimes" configure > option as a means to get the previous default behavior. Would it make more sense to have phrase it in the positive sense? i.e., --enable-floating-point-datetimes? I guess that's a bit longer, but it says what you're doing, rather than what you're *not* doing. Michael Glaesemann grzm seespotcode net
On Sat, 2007-05-05 at 22:49 -0500, Michael Glaesemann wrote:
> Would it make more sense to have phrase it in the positive sense?
> i.e., --enable-floating-point-datetimes? I guess that's a bit longer,
> but it says what you're doing, rather than what you're *not* doing.
I think the primary reason people will want to use FP-based datetimes is
because they can't use integer-based datetimes for compatibility reasons
(e.g. no OS support for 64-bit integers, or they need to remain
compatible with old applications). The situation is analogous to
--without-spinlocks: we could call that --enable-slow-locking or
something, but that would sound like we're enabling an additional
feature.
It would also mean there would be an implicit relationship between
"--enable-integer-datetimes" and "--enable-fp-datetimes" (at most one
can be true). IMHO it would be simpler to just keep a single boolean
variable ("integer datetimes enabled or not").
-Neil
Neil Conway wrote: > Attached is a patch that enables integer datetimes by default, per > recent discussion on -hackers. It makes "--enable-integer-datetimes" the > default, and documents the "--disable-integer-datetimes" configure > option as a means to get the previous default behavior. > > Barring any objections, I'll apply this to HEAD tomorrow. > > I object to the short notice. I think we need to give people a chance to adjust their configs, especially on the buildfarm, where those who have currently simply removed the --enable-integer-datetimes setting will need to adjust their configs. Once there is agreement I think we should at the very least give those people a few days to adjust. cheers andrew
On Sun, 2007-06-05 at 00:20 -0400, Andrew Dunstan wrote: > I object to the short notice. I think we need to give people a chance to > adjust their configs Sure, I can wait a few days (although if we're going to do this for 8.3, we should do it promptly). On reflection, it might actually be wiser to delay making this change until the beginning of the 8.4 cycle... -Neil
Peter Eisentraut wrote: > Andrew Dunstan wrote: > >> I think we need to give people a chance to >> adjust their configs, especially on the buildfarm, where those who >> have currently simply removed the --enable-integer-datetimes setting >> will need to adjust their configs. >> > > Why? They have previously been testing the default and now they would > be testing a different default. > > It is not the default for the buildfarm - the default config has --enable-integer-datetimes. I would rather people have a chance to adjust their configs so they keep testing the same config set, regardless of whether or not it's the default. cheers andrew
Andrew Dunstan wrote: > I think we need to give people a chance to > adjust their configs, especially on the buildfarm, where those who > have currently simply removed the --enable-integer-datetimes setting > will need to adjust their configs. Why? They have previously been testing the default and now they would be testing a different default. -- Peter Eisentraut http://developer.postgresql.org/~petere/
Neil Conway wrote: > On reflection, it might actually be wiser to > delay making this change until the beginning of the 8.4 cycle... > > > I think that's probably true, but I can live with it either way as long as there's enough notice. cheers andrew
Neil Conway <neilc@samurai.com> writes:
> On Sun, 2007-06-05 at 00:20 -0400, Andrew Dunstan wrote:
>> I object to the short notice. I think we need to give people a chance to
>> adjust their configs
> Sure, I can wait a few days (although if we're going to do this for 8.3,
> we should do it promptly). On reflection, it might actually be wiser to
> delay making this change until the beginning of the 8.4 cycle...
+1 ... this had not been previously discussed for 8.3 and so it's
arguably too late in the cycle.
regards, tom lane
Added to TODO: * Have configure choose integer datetimes by default http://archives.postgresql.org/pgsql-patches/2007-05/msg00046.php --------------------------------------------------------------------------- Tom Lane wrote: > Neil Conway <neilc@samurai.com> writes: > > On Sun, 2007-06-05 at 00:20 -0400, Andrew Dunstan wrote: > >> I object to the short notice. I think we need to give people a chance to > >> adjust their configs > > > Sure, I can wait a few days (although if we're going to do this for 8.3, > > we should do it promptly). On reflection, it might actually be wiser to > > delay making this change until the beginning of the 8.4 cycle... > > +1 ... this had not been previously discussed for 8.3 and so it's > arguably too late in the cycle. > > regards, tom lane > > ---------------------------(end of broadcast)--------------------------- > TIP 2: Don't 'kill -9' the postmaster -- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB http://www.enterprisedb.com + If your life is a hard drive, Christ can be your backup. +