Re: [0/6] v10: stack branch merge to trunk

["Joseph S. Myers" <joseph at codesourcery dot com>]

On Thu, 31 Jul 2008, Andreas Tobler wrote:

> So, I'd have expected a call from you for testing your branch on different
> OS's which have the base x86 cpu as default before merging to trunk. I'm
> willing, and I DO testing on all of my machine park cpu's if I'm asked.
> 
> The blame is not only against you, but also to the reviewers.
> 
> I'd like to propose a merging criteria, merging from a branch to trunk is only
> allowed if a set of primary and secondary targets passed bootstrap and
> testing. If not, the branch has to stay branch until this criteria is
> fulfilled.

A branch is just one way of developing a patch.  The criteria should be 
based on the risks associated with the particular patch (as judged in the 
associated review), not on where it was developed.  For some patches one 
target will suffice, some need several.  In some cases the variation 
should be between different CPUs, in some cases between targets for the 
same CPU.  In some cases benchmarks or compile-time performance results 
may be a desirable part of the validation.  For graphite I expressed a 
concern for testing on several hosts rather than several targets; if 
gcc-in-cxx merges to trunk in future then testing it builds with a non-GNU 
C++ compiler would be an appropriate requirement.  You can't produce 
one-size-fits-all requirements for branch merges based simply on the fact 
that a branch was used.

-- 
Joseph S. Myers
joseph@codesourcery.com