Re: Beyond GCC 3.0: Summing Up

[dewar at gnat dot com]

<<That is the problem. I agree we should have a full qualification for
a major relese, even that won't guarantee you a perfect release. We
are talking the bug fix release here. Presumably, changes are
relatively small. Shouldn't put our energy into getting more releases
out rather than making full qualifications? To me, making a bug fix
release itself is one kind of qualification. Many gcc bugs won't be
discovered until the users start using the new gcc. Why not let users
help us make a great gcc, which the gcc developers can't do by
ourselves alone anyway?
>>

First of all, there is no such thing as a "small change" when it comes
to being sure that you have no regressions, you really have to run the
complete test suite. note that this is what I mean by full qualification
in this context, bootstrap on all major targets, and run the test suite
on all major targets to verify that there is no regression at this level.
That should not be so hard (it should, and could, be automated).

Of course there might be regressions which can only be found by field
testing, and you can't do full field testing of every minor release (we
don't really do full field testing of major releases, not in the sense
that I use the term anyway).

Users will not want to spend much time with intermediate releases if they
do not go through at least this level of testing. A bug fix release which
introduces new bugs is a dubious beast!