PHILIP/Test-Spec-0.47.tar.gz fails with 1.301001071

[andk]

Seems to fails since 1.301001_051 or thereabouts. Sample fail report:

http://www.cpantesters.org/cpan/report/47392685 # 5.21.5
http://www.cpantesters.org/cpan/report/47392754 # 5.20.1

HTH&&Thanks

[exodist]

Looks like a conflict. Test::More now exports a context function, looks like this module also uses something with that name.

Not sure best solution here. Will look deeper when I can get to a computer.

[andk]

Maybe releated: ADAMK/Aspect-1.04.tar.gz

The diag there is:

#   Failed test 'no warnings'
#   at /home/sand/src/perl/repoperls/installed-perls/perl/v5.21.5-114-g03d05f6/a2da/lib/site_perl/5.21.6/Test/NoWarnings.pm line 45.
# There were 1 warning(s)
#       Previous test 0 ''
#       Subroutine context redefined at t/00_context.t line 57.
#  at t/00_context.t line 57.

[exodist]

Aspect is fixed simply by moving 'context()' to not be exported by default, which is fine since it is new. This change also removes a warning from Test-Spec, but it does not fix it. I have dug deeper and found the problem.

define.t:

116 {   
117   no warnings 'once';
118   my $stub = Stub->new;
119   local *A::builder          = sub { $stub };
120   local *Test::More::builder = sub { $stub };
121   A->runtests;
122 } 

The test that fails tries to hide 2 tests by replacing the Test::Builder singleton with a stub class that simply returns on all sub calls (autoload).

This is a problem because one of the key goals of the alphas is to kill the singleton and the problems having a monolithic singleton caused. Instead there is a single minimal singleton that has 1 job, funnel results to the right place. It also has a built-in api that lets you do the kinds of thing this test wants to do, ie hide or intercept results.

Basically the fact that authors had to do this kind of hackery at all is the very limitation of Test::Builder that we are hoping to overcome with the alphas.

I can think of one possible fix on our end, but I am not sure if it might cause other issues. The fix would be to have Test::Stream 'notice' that the singleton has been replaced with an object that is not a subclass or instance of Test::Builder, if it detects this it can try to go through it... But I am pretty sure this would do the following:

• Make the logic very complicated and hard to maintain
• Break a lot of things
• Slow Test::* down enough that blead might reject it again.

I think the better option would be to ask the author to make a small change to their code:

define.t:

116 {
117   if ($INC{"Test/Stream.pm"}) {
118     Test::Stream->intercept(sub { A->runtests });
119   }
120   else {
121     no warnings 'once';
122     my $stub = Stub->new;
123     local *A::builder          = sub { $stub };
124     local *Test::More::builder = sub { $stub };
125     A->runtests;
126   }
127 }

I have confirmed this solves the problem. I will submit it to the author.

[exodist]

Closing this ticket in favor of the module specific pull request: kingpong/perl-Test-Spec#18

[exodist]

https://rt.cpan.org/Ticket/Display.html?id=100034

Added an RT ticket since it is their prefered tracker method. I forgot pull requests were glorified issues