introduce a bit of randomness into the test program. Have it ship with some default 'troublesome' cases, and then generate random other cases. So, for a mid.pl test, the program might ship with input files having one line, two lines, three lines, four lines, lines with \n from three different platforms, maybe files without terminating \n characters (if these inputs are expected to be dealt with), etc. It could then generate a handful of random files (on *each* test run) as input as well.
As the test program is merely advisory, golfers shouldn't expect that a 'pass' necessarily means that they will *always* pass. As with any competition, some days you may do very well, and some days you may trip on your shoelaces. Prepare to the best of your ability to not trip on your shoelaces, and hope it doesn't happen on the day of the event. Perhaps to facilitate this preparation, the test program could be given a parameter indicating how many random input files should be created (for those tests for which random input files are appropriate). That way golfers can abuse their computers and their patience exactly (or at least approximately) as much as is appropriate for their needs. -matt (who is too used to writing code that coworkers can maintain to be a decent golfer, but finds the entries and their discusssion absolutely fascinating)
