On Tue, Jul 6, 2010 at 10:42 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> log_temp_files was introduced in 8.3, so we'll need to backpatch this
>> to 8.3, not just 8.4. Greg Smith tells me Simon has been busy with
>> other things, so I'm going to pick this up. Barring objections, I'm
>> going to revert Simon's patch and change the behavior instead, per the
>> attached patch. I will also back-patch the fix to 8.4 and 8.3.
>
> I agree with doing this for 9.0, but I'm much less sold that it's a good
> idea to change the meaning of the parameter in minor releases of 8.4 and
> 8.3. Aren't you risking breaking installations that work now?
I think my least favorite option is changing the behavior only in
HEAD. I think the reasonable options are:
1. Change the behavior in HEAD, 8.4, and 8.3, per previous discussion.If we do this, we should do what I proposed in my
previousemail.
2. Change the comments and documentation in 8.4 and 8.3 along the
lines that Simon already did in HEAD. If we do this, we also need to
change the GUC units to something other than GUC_UNIT_KB, as noted
upthread. I'm not sure what would be appropriate.
The reason I think it's OK to change the behavior in the back-branches
is that (a) the only thing it affects is logging, so it shouldn't
really "break" anything, and (b) apparently nobody has noticed that
the interpretation of the GUC is off by three orders of magnitude, so
either nobody's using it or they're not looking at what's actually
happening too carefully. But I'm OK with going the other way and
changing the code and docs in the back-branches, too. I just think we
should be consistent.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise Postgres Company