Hello, I am trying to test a module, but to that end, I must fork two processes, and the parent and the two kids each perform a series of tests.
The output is such: 1..7 ok 1 - The proxy requests what we expect ok 1 - Got what we wanted ok 2 - The proxy requests what we expect ok 2 - Got what we wanted ok 1 - Served the correct number of requests not ok 3 - Got what we wanted # Failed test (t/20simple.t at line 84) ok 3 - The proxy requests what we expect # Looks like you planned 7 tests but only ran 3. (Before you ask: yeah, I know one of the tests fails, I am also investigating this) Each instance of the forked Test::Builder thinks it's all alone in the world, and starts at 1. Which is perfectly understandable. So I tried to add the following lines at the beginning of my test script: my $test = Test::Builder->new; $test->use_numbers( 0 ); But Test::Builder is still confused and only sees the three tests run by the parent process. 1..7 ok - The proxy requests what we expect ok - Got what we wanted ok - The proxy requests what we expect ok - Got what we wanted ok - Served the correct number of requests not ok - Got what we wanted # Failed test (t/20simple.t at line 87) ok - The proxy requests what we expect # Looks like you planned 7 tests but only ran 3. Ooh, I think I get it: the parent process' Test::Builder warns about the fact that it has not seen all tests (since most of them were run by the children). On the other hand, Test::Harness sees everything right, since it's looking at the test script output only. Hence with make test, I simple have garbled output due to the diagnostics. I guess my question is something like is there's a way for the children's test output to be taken into account by the main Test::Builder object? Must I really use some IPC to report back to the parent process' Test::Builder object? Or is it considered ok just to have the kids spit "ok" and "not ok" and to let Test::Harness organise the resulting mess? Be aware that the test results do not look that good (when no test fails): Failed 1/5 test scripts, 80.00% okay. 0/29 subtests failed, 100.00% okay. Not that great. So I guess I'll have to go the IPC way, after all. Any advice or success story? -- Philippe "BooK" Bruhat Even the worst guesser is right once in a while. (Moral from Groo The Wanderer #72 (Epic))