Andrew Dunstan <andrew@dunslane.net> writes:
> On 11/20/21 20:10, Noah Misch wrote:
>> The skip would be unnecessary if configure just tested whether $PROVE can run
>> a test requiring the module. We're testing $PERL, but we're actually
>> indifferent to $PERL's Test::More.
> Yeah, we could do something along these lines:
> [ script ]
Seems reasonable to me, although it's not entirely clear how to hook
the output into configure's practices --- in particular, people using
"./configure -q" might not be pleased by unwanted diagnostic output.
But that could probably be dealt with.
> prove is pretty much always a script - if we want to know which perl is
> invoked we could look at its shebang line.
Do we care though?
I think the $64 question is whether it's a great idea to run the
TAP tests with a different Perl than is used for (a) build tooling
and (b) building plperl against. I guess that it should work in
principle, and the msys animals apparently need it, but I'm
concerned that somebody will waste a lot of time being confused by
unexpected behavioral differences. ("This works when I run a perl
script by hand, why doesn't it work in my TAP test?") So I still
think that the best default behavior is to pick a prove associated
with the selected perl, and if you really want something other than
that then you have to set PROVE explicitly.
IOW, I think we should make (and back-patch) both of the changes
discussed in this thread.
regards, tom lane
