Today, I got an email from Tom, my replacement at IQE. He'd found himself needing a new kind of tolerance for Number::Tolerant. I originally wrote Number::Tolerant for IQE's characterization system, and I really enjoyed working on it. Once I left, though, I didn't have much reason to work on it.
So, today Tom sent me that email, and included the code for the new kind of tolerance. It's basically a plus-or-minus offset, but the target value is not the midpoint of the extremes.
I integrated his code (tweaking just a bit), and wrote tests. After writing tests for offset, I ran coverage and found that it was nowhere near the 100% coverage I remebered. I reorganized a bunch of my tests, wrote about 150 more tests, found a few little bugs, and fell back in love with this obscure, weird little distribution.
I can clearly see how it needs to be refactored, now, and I may just do that soon.
Potentially best of all, I believe I've found a bug in Devel::Cover. Maybe I'm wrong, or maybe Paul will decide that the condition required to make the bug appear is not worth addressing. If not, though, we will have identified a new bug in one of the most useful Perl programming tools, all because some engineers at IQE demand weird product specifications!
Tom at IQE, I salute you!
Re:bugreport?
rjbs on 2006-01-29T14:22:35
I've reported the bug to PJ and given him a test case.
Basically, if you define the same package in two files, and each definition of the package has the same sub(s), once you load the second version of the package, coverage of the first package will not be recorded. It will be dropped during coverage analysis as an "extra branch."
I solved this temporarily by reordering my tests, and then by refactoring so that I did not use the same namespace twice. The bug is still there in D::C, though.
