Most of the tests follow the same pattern.

 * test.pl that uses Test::More, and demonstrates whatever functionality 
   that we're trying to test.  This is the reference code.

 * test.c, which tests the libtap reimplementation of the same functionality.

 * test.t, which compiles the .c program, runs both test scripts, and then 
   diffs their output to make sure it's identical.

   Right now, test.t is identical in every directory.  This sucks somewhat.
   It should either be a symlink to a common script