Magnus Hagander <magnus@hagander.net> writes:
> On Mon, Jan 3, 2011 at 19:08, Andrew Dunstan <andrew@dunslane.net> wrote:
>> I'm not going to maintain more than one buildfarm member doing MSVC, and and
>> if we were to adopt your policy I would not be able to use a modern-ish
>> version of the compiler/SDK and also build all the live branches.
> Well, it's perfectly possible to have more tha none version of MSVC on
> the machine.
> And we're not going to be changing the version that's actually used
> for the official binary builds, so all you'll accomplish then is to
> have the buildfarm test something different form what we're shipping.
Are you speaking for EDB on that? Do you even know what they're using
to build the Windows installers?
We've made backpatches before to support building/running older branches
on newer platforms. We do it all the time in fact. (The latest
instance was hacking the Linux wal_sync_method defaults. If you think
this isn't a necessary activity, try building a 7.1 or so branch with a
modern gcc.) It might be reasonable to argue that this particular patch
is too invasive to be safe to back-patch, but I don't agree with the
premise that it isn't a reasonable topic for a back-patch.
I do have some concern about loss of buildfarm coverage for older VS
versions, but if Andrew isn't going to cover those, perhaps someone else
will step up for that.
regards, tom lane
