On my perl-5.10 I had Test-Harness-2.64. I updated to version 3.10, and Math-Pari still builds and tests flawlessly for me.

I investigated further, here's my report:

With perl5.8.8 make test runs fine, with 5.10.0 (and blead) it doesn't. Both perls use the same version of ExtUtils::Command::MM (6.44) and Test::Harness (3.10). Furthermore I isolated the test file which seem to cause this strange error, it's t/55_objets.t (sic!). This test file uses test_eng/Testout.pm, which loads pari-2.1.7/src/test/32/objets (at least on my system). I reduced this file to

   realprecision = 38 significant digits
   echo = 1 (on)
? truncate(1.7,&e)
1
? \p57
   realprecision = 57 significant digits
? length(divisors(1000))
16
? \q
[download]

Now we run the t/5_objets.t test file:

$ PERL_DL_NONLAZY=1 /opt/perl-33445/bin/perl5.11.0 "-MExtUtils::Comman
+d::MM" "-e" "test_harness(0, 'blib/lib', 'blib/arch')" t/55_objets.t
t/55_objets...... All 3 subtests passed

Test Summary Report
-------------------
Files=1, Tests=3,  0 wallclock secs ( 0.00 usr  0.00 sys +  0.04 cusr 
+ 0.02 csys =  0.06 CPU)
Result: FAIL
Failed 1/1 test programs. 0/3 subtests failed.
[download]

-- Frank

Blead throws up a number of issues that I'll be taking a look at as time permits. (I don't have a lot of time right at the moment ... work is interfering again.)

Firstly, when running the 5.11.0 test suite, op/alarm.t hangs, and killing the hang aborts the test procedure entirely - so I only got to run a small number of the test files.

After installing 5.11.0, I couldn't even get the 'Makefile.PL' step of the Math-Pari build process to complete. It hangs at "...I will use assembler build of type 'ix86'". Killing the hang kills the entire process, so no makefile even gets written.

While I'm at work tonight I'll try it out on my XP laptop. (I'm using Vista here at home.) Interesting ....

Cheers,
Rob
Update: On my Windows XP box I get essentially the same error as you reported with Math::Pari.

With 5.10.0 (but not 5.11.0) t/55_objets.t doesn't produce any problems when run as perl -Mblib t/55_objets.t. But when run as part of the 'make test' process (with both 5.10.0 and 5.11.0) I get a segfault during cleanup, when the script is about to exit after all of the tests have been successfully run. Strange that it doesn't affect Vista the same way.