On Friday, March 24, 2017, Heikki Linnakangas <hlinn...@iki.fi> wrote:
> On 03/22/2017 07:44 PM, Robert Haas wrote: > >> On Wed, Mar 22, 2017 at 10:13 AM, Alvaro Herrera >> <alvhe...@2ndquadrant.com> wrote: >> >>> Heikki Linnakangas wrote: >>> >>>> I did some archeology, and found CheckTokenMembership() in MinGW's >>>> w32api >>>> packages version 3.14 >>>> (https://sourceforge.net/projects/mingw/files/MinGW/Base/ >>>> w32api/w32api-3.14/, >>>> in include/winbase.h). According to the timestamps on that download >>>> page, >>>> that was released in 2009. That was the oldest version I could find, so >>>> it >>>> might go even further back. >>>> >>>> Dave, do you know exactly what version of MinGW narwhal is running? And >>>> how >>>> difficult is it to upgrade to something slightly more modern? Ease of >>>> upgrade is another good data point on how far we need to support old >>>> versions. >>>> >>> >>> Given that this was backpatched and that it broke narwhal in all >>> branches, I think the solution needs to make narwhal work again without >>> requiring it to upgrade; so we should acquire CheckTokenMembership via >>> dynloading just like we do the other functions. If we want to require a >>> newer mingw version in pg10, that's acceptable, but it should be a >>> separate patch. >>> >> >> +1 for not moving the minimum system requirements in the back-branches. >> > > Ok. I reverted this patch in the back-branches, and applied the much less > invasive "V2" patch [1] instead. HEAD is unchanged, so narwhal still fails > there. > > Dave: the consensus is that we no longer support the old version of MinGW > that narwhal is using, for PostgreSQL v 10. Can you modify the > configuration of narwhal to not try building 'master' anymore, or upgrade > the toolchain, please? > I've disabled it. Thanks. -- Dave Page PostgreSQL Core Team http://www.postgresql.org/
