On Tue, Oct 20, 2009 at 3:48 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Dave Page <dpage@pgadmin.org> writes:
>> The patch doesn't change what the code aims to do, only the way it
>> does it. The existing code does this:
>> ...
>> The net result /should/ be the same, but the second method is
>> apparently a little more robust.
>
> Do we have any idea why? =A0I am always distrustful of random changes made
> with no theory as to why they fix a problem. =A0My experience is that such
> changes are almost always wrong, once you find out what the problem
> *really* is.
Honestly? No. I have a vague hand-wavy idea about there being
something preventing us properly modifying the token of an existing
process in some configurations, but nothing even remotely jello-like,
let alone concrete.
On the other hand, I don't see any obvious way for this to cause a
regression - which was born out by my (limited) testing in which the
original problem remained fixed with the new patch. I'd certainly feel
happier if Magnus took a look as well though.
--=20
Dave Page
EnterpriseDB UK: http://www.enterprisedb.com
PGDay.EU 2009 Conference: http://2009.pgday.eu/start