>>Is this opinion backed with actual experience with code coverage on a non-trivial app?
>
>I think I can answer yes to that. I've worked on several non-trivial apps (all on mainframe) and they all got as tested as could possibly be done. That included EVERY compare construct, EVERY BRANCH (these were OS Assembler apps), etc.

For retail software this is either a) not a good ROI or b) the code was not written defensively enough to include code that simply cannot be reached. Probably a good dose of both. I would maintain that it's probably true for custom apps as well.

I know for VFP there's a fair bit of code that I have no way to get to, and even with new code for 8.0 the current devs are no help in getting to some of the stuff I can't execute. A coder more wise than me once said that 90% of good programming is coding for stuff that theoretically should never happen.