Should there be a GCC 4.0.1 release quickly?

[Mark Mitchell]

Joe Buck wrote:
> On Thu, Apr 28, 2005 at 07:31:32PM +0200, Steven Bosscher wrote:
> 
>>Yeah, but in this case the patch that introduced the bug was one of
>>the last to go in before the release (it was the fix for PRs 20490
>>and 20929, the patch for that went in on April 17).  So it was more
>>an unfortunate fix than a typical .0 bug.
> 
> 
> Again, that's not surprising.  There's always a risk that a patch can
> create a new bug, and the regression test suite only proves we haven't
> re-broken old bugs.  Late fixes are particularly risky, because of the
> limited testing they get.

Exactly.  The paradox here is that I get a ton of mail right before a 
release with people who want to incude additional patches (which are 
always safe, obviously correct, and fix critical problems), and then I 
often get mail after the release from people who want to do another 
release quickly to fix critical problems (by applying safe, obviously 
correct patches) introduced by the previously applied patches. :-)

I'm not at all surprised that GCC 4.0.0 has bugs, even relatively 
serious ones.  It's a .0 releases with an entirely new optimizer.  If we 
didn't get any bug reports about serious problems I'd just conclude that 
nobody was using it.  I certainly would have been amazed to find 
distributors shipping unpatched 4.0.0 to customers; I know CodeSourcery 
would not!

I'd rather not rush to a 4.0.1 release.  I'm happy to consider moving it 
up a bit, but I'd like to find out what the top five or ten critical 
problems are and try to fix all five or ten, rather than rush out 
another release which fixes one or two of those problems, but still 
leaves people with trouble on the remainder.

So, how about we hang in there another week or so, and see what things 
look like?  In the meantime, I'm trying to plan 3.4.4....

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304