On 11/23/21 12:03, Tom Lane wrote: > [ moving thread to -hackers for a bit more visibility ] > > Attached are a couple of patches I propose in the wake of commit > 405f32fc4 (Require version 0.98 of Test::More for TAP tests). > > 0001 responds to the failure we saw on buildfarm member wrasse [1] > where, despite configure having carefully checked for Test::More > being >= 0.98, actual tests failed with > Test::More version 0.98 required--this is only version 0.92 at > /export/home/nm/farm/studio64v12_6/HEAD/pgsql.build/../pgsql/src/test/perl/PostgreSQL/Test/Utils.pm > line 63. > The reason is that wrasse was choosing "prove" from a different > Perl installation than "perl", as a result of its configuration > having set PERL to a nondefault place but doing nothing about PROVE. > > We already installed a couple of mitigations for that: > (a) as of c4fe3199a, configure checks "prove" not "perl" for > appropriate module versions; > (b) Noah has modified wrasse's configuration to set PROVE. > But I'm of the opinion that (b) should not be necessary. > If you set PERL then it's highly likely that you want to use > "prove" from the same installation. So 0001 modifies configure > to first check for an executable "prove" in the same directory > as $PERL. If that's not what you want then you should override > it by setting PROVE explicitly. > > Since this is mainly meant to prevent an easy-to-make error in > setting up buildfarm configurations, we should back-patch it.
Do we really have much of an issue left to solve given c4fe3199a? It feels a bit like a solution in search of a problem. > > 0002 is written to apply to v14 and earlier, and what it wants > to do is back-patch the effects of 405f32fc4, so that the > minimum Test::More version is 0.98 in all branches. The thought > here is that (1) somebody is likely to want to back-patch a > test involving Test::More::subtest before too long; (2) we have > zero coverage for older Test::More versions anyway, now that > all buildfarm members have been updated to work with HEAD. > This one seems like a good idea. cheers andrew -- Andrew Dunstan EDB: https://www.enterprisedb.com
