On 07/31/2017 01:02 PM, Tom Lane wrote:
> Record full paths of programs sought by "configure".
>
> Previously we had a mix of uses of AC_CHECK_PROG[S] and AC_PATH_PROG[S].
> The only difference between those macros is that the latter emits the
> full path to the program it finds, eg "/usr/bin/prove", whereas the
> former emits just "prove". Let's standardize on always emitting the
> full path; this is better for documentation of the build, and it might
> prevent some types of failures if later build steps are done with
> a different PATH setting.
>
> I did not touch the AC_CHECK_PROG[S] calls in ax_pthread.m4 and
> ax_prog_perl_modules.m4. There seems no need to make those diverge from
> upstream, since we do not record the programs sought by the former, while
> the latter's call to AC_CHECK_PROG(PERL,...) will never be reached.
>
> Discussion: https://postgr.es/m/25937.1501433410@sss.pgh.pa.us
The problem with this commit, as jacana is demonstrating, is that on Msys it finds the wrong prove. configure needs to
runagainst the perl we build against, i.e. a native Windows perl, but prove needs to run with the perl from the MSys
DTKthat understands MSys virtualized paths. I have a hack that will allow the buildfarm to overcome the difficulty,
(essentiallyit passes 'PROVE=prove' to make) but that's fairly ugly and certainly non-intuitive for someone running an
MSysbuild and TAP tests without the buildfarm client.
cheers
andrew
--
Andrew Dunstan https://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services