Re: Re: On PEAR quality issues and natural selection

Lukas Smith wrote:

So what I am saying is this: Its alot of work to really cover your API. Its a worthwhile cause of course. Generally however reality shows me that most wierd combinations I dont even think up. So in the end I write a test case for every bug found and so my test suite slowly improves.

I completely agree on the balancing act. My only point is that it is important to let users know whether they can rely upon a package on their system, i.e. can they set up and run test cases to make sure things work? If not, I think it's fair to consider stability lower, just as you've been doing with MDB2, even though it works quite well. In other words, a high barrier to the "stable" label is a good thing, even if it delays the golden label a bit. In the end, a good solution to this issue will be found for each individual package, and I hope for the whole of PEAR. I'm sure that convergence will happen as I describe in one way or another. PHPUnit, .phpt, and simpletest are all beginning examples of attempts to converge regression testing, and it will only go further as more experience is gained. Greg