Re: Checkin Policy

[Mark Mitchell <mark at markmitchell dot com>]

>>>>> "Joern" == Joern Rennecke <amylaar@cygnus.co.uk> writes:

    Joern> C++ has test regressions so often and so quickly chainging
    Joern> that it's hard to use the C++ testsuite for regression

That's a fair criticism.  But, I don't think it's as fair now as it
has been in the past.  We've committed ourselves to a policy of no new
tests with unexpected results, which means that the set of passing
tests should remain stable.  I, for one, never check in a change that
causes test results to change.  (In other words, I check in test-suite
changes along with any code changes so that results remain stable.)  I
believe that the other C++ developers make a good faith effort to do
the same.

So, I suggest that you take the time to run the C++ testsuite as
well.  If you notice the C++ test results changing over time, please
complain; that will encourage the C++ developers (myself included) to
make sure we're not breaking the testsuite.  

And, you could always compare the output from the version with your
change to a version without your change.  Lots of recent test-suite
results area avilable in the archives.

    Joern> testing of the gcc optimizers.  Note that I did do an x86
    Joern> native bootstrap and an SH cross and checked for C
    Joern> regressions.

Obviously, these are quite comprehensive tests.  That's why I
specifically mentioned in my mail that I meant to particular criticism
of the change in question, or your procedures for checking it in.

Never-the-less, I don't think that running the C++ tests is a very
large additional overhead, if you were willing to do all that you did
above.  So, perhaps you would be willing to run the C++ tests too in
the future?

-- 
Mark Mitchell 			mark@markmitchell.com
Mark Mitchell Consulting	http://www.markmitchell.com