SC issues revised

[mike stump]

> To: Tom Lord <lord@emf.net>
> cc: gcc@gcc.gnu.org
> Date: Tue, 19 Mar 2002 18:41:30 -0500
> From: David Edelsohn <dje@watson.ibm.com>

> 	Discussions about improving the GNU toolchain and coordinating
> that development occur all the time among interested parties.

Take for example, testing.  I have here at my disposal a complete
target lab with OSes and board support packages for easily 41
different boards suitible for full rounds of automated testing,
complete with most all the necessary software to intelligently
schedule and run the testsuites.  I currently do this for the gcc,
g++, binutils, gas, libio and libstdc++ testsuites.  These boards
represent the sweet spot (as we see it) in the current marketplace,
hence a certain amount of relevance to the popular targets.  We don't
currently track top of tree, but we'd like to, and are currently
working to integrate in the top of tree.  I'd like to be able to have
a branch that does track top of tree (because it is more efficient,
better, easier), and have the branch shaken down against the
testsuites and have it report those results back to the usual place.

I'd like to have it intelligently coordinate with the other regression
testers, and publish/report the various regressions.  In part, to
lower our cost, as then _we_ don't have to do this.  Call it
self-serving, if you want.  But, the benefit to gcc is the additional
testing, and testing on more obscure parts, on a more regular and
predictable basis.  The mean age currently for all targets on my
results page is around 4 days.

I don't feel we need changes to the current model to acomplish
more/better testing.  All we need is someone to value it, just one
person, and to contribute.  If not a single person values it, then,
the masses have spoken, it isn't valuable.  SunOS support for example,
not valuable, not tested, doesn't work.  It used to be valuable.