Re: release process, showstopper, regression tests

> I think a 3 day period (with a standardized notification message like > "4.0.5 will be released in 2 days") without any veto would be > appropriate to avoid that anybody misses it Yes, I think we will make that better next time. > 2) > also, what qualifies as a showstopper? I would have thought crashes do, > but unfortunately zeev negated in a one-line comment > (http://marc.theaimsgroup.com/?l=php-dev&m=97696875919208&w=2) > > I am simply dissatified having an announce mail underlining the work of > the qa-team and then receiving two bug reports at once regarding an > almost known issue, it would not have been dissatisfying if problems > with ob_gzhandler were totally unknown Yes, I agree that having software with zero bugs would be very very nice. However, in reality we hardly pass "good enough" and for a software like PHP where many people contribute voluntarily, it is even harder to bring quality into the process. That is why I think that it is acceptable to make a cut at some point and only consider defects which affect a large body of users. > 3) > what about a policy forcing devlopers to provide basic regression tests > for their extensions, it most often is no task needing much effort - how > else do we know that specific functions are working in every release > without running tests on all of them... Is there some kind of document describing how to write correct tests? I'm asking this, because I have not seen a single "make test" output yet where some tests don't fail (i.e. dirname). Are the functions or the tests broken in those cases? - Sascha