Testing Questions

[Andy Armstrong]

On 14 May 2007, at 23:23, Nicholas Clark wrote:
> I'm always surprised that so few people seem to do this. Often  
> people I work
> with write tests, and each is/like/... runs once. I like putting  
> them in
> loops or subroutines and feeding them data until they refuse even a  
> wafeur
> thin mint. Having this style of battery of tests has often exposed  
> unexpected
> corner cases due to the range of data tried.

The regression test for Perl::Version[1] started out as a little REPL  
that I typed variously formatted version strings into. It captured  
and displayed the output. Every time I noticed an inconsistency I  
probed with some more tests around the same area. Then I dumped all  
the captured inputs and outputs, manually corrected the erroneous  
outputs and wrapped the resulting data with a little test driver.

If I'd written the tests longhand I don't think I'd have been  
anything like as exhaustive.

And if I wanted to port the code to Python or Ruby I'd only have to  
rewrite the 'driver' bit of the test - the data would remain the same.

[1] http://search.cpan.org/src/ANDYA/Perl-Version-v0.0.7/t/ 
10.regression.t

-- 
Andy Armstrong, hexten.net