Or you could have asked how to do it, instead of trawling (fishing) for "concrete reasons" ... :)

See another example http://search.cpan.org/~evo/Class-MakeMethods-1.01/test.pl, the magic part is naming it "tests/*t" instead of "t/*t", works like a charm, there is no TAP swallowing or anything :)

I had been already to break out the Champagne when I cleaned out everything and rebuilt from scratch. Once again, even with all the tests in tests rather than t, I saw the same behavior. test.pl runs fine when it runs directly on the command line perl ./test.pl. TAP gets swallowed when test.pl is executed by make test.

Clearly whatever is causing this on-again off-again TAP swallowing is not merely about the name of the test directory. In fact it seems to have nothing to do with it. You can get make test to work with the files in t or in test. The only factor that seems to make a difference is the "test" key:

# this creates a broken Makefile
WriteMakefile
(
 'INSTALLDIRS' => 'site',
 , 'NAME' => MyModule::Foo
 , 'VERSION' => 0.999_003
 , 'PREREQ_PM' => { 'Fiddle::Faddle' => 0 }
 , 'test' => { TESTS => 'test.pl' }
);

# this does not (test key removed)
WriteMakefile
(
 'INSTALLDIRS' => 'site',
 , 'NAME' => MyModule::Foo
 , 'VERSION' => 0.999_003
 , 'PREREQ_PM' => { 'Fiddle::Faddle' => 0 }
);
[download]

I'm not going to be entirely comfortable until I understand why this makes a difference. Just as yours and my empirical observations about t vs. tests turned out to be unlikely, so too could my empirical observations about the 'test' key.

It does however look like we are well on our way, without having to reinvent the wheel.

As for "you could have asked" :-).... don't underestimate your role in causing me to revisit an option I'd rejected. By believing it could be done and even offering to look at a reduced code sample, you convinced me that I might have given up too quickly. You all but invited me to ask for help - and you are correct that I did not explicitly do so - it isn't easy for me - I tend to believe I have to take care of myself at all times. However, had you not issued that quasi-invitation, I would not have told you the story of why I gave up and perhaps you would not have offered evidence that my original intent to call runtests could work if things outside my test.pl code were set up properly.

Thanks again for your help. You do honor to the name "Anonymous Monk".

I asked for a patch before; here is how I work it

module-starter --eumm  --module=TestPL     --author=name  --email=emai
+l
cd TestPL
wget http://cpansearch.perl.org/src/EVO/Class-MakeMethods-1.01/test.pl
perl Makefile.PL && nmake test
[download]

This does the unwanted

C:\perl\bin\perl.exe "-MExtUtils::Command::MM" "-e" "test_harness(0, '
+blib\lib', 'blib\arch')" t/*.t
C:\perl\bin\perl.exe "-Iblib\lib" "-Iblib\arch" test.pl
[download]

which is why you rename t to tests, so that only test.pl gets run.

If you add test  => { TESTS => 'test.pl' }, you get No plan found in TAP output, which is not surprising, since test.pl runs its own harness, so there is nothing for the other harness to consume.

You don't want to run test.pl through a harness, test.pl is the harness.

Do you see what I see?