RFD: test requirements for slow platforms

[Mark Mitchell]

Ian Lance Taylor wrote:

> I'm personally reluctant to codify it, because it's really hard to
> codify good judgment.  But if you say in your patch how you tested it,
> the reviewers should be able to consider whether that is sufficient.

I agree.

I always claim that my most valuable contribution to GCC was the
argument I started a decade ago that culminated in us deciding that you
had to run the regression testsuites before check-in.  So, I'm a big fan
of testing, and the general idea that you shouldn't have to deal with
breakage because I was too lazy to test.

On the other hand, I think we can get obsessive about it.  Testing Ada
and Java libraries for every C front-end change seems excessive to me.
Testing is not free.  Similarly, doing a bootstrap on Super-H GNU/Linux
is possible, but it takes a *long* time.  I doubt that's worthwhile most
of the time.  I think we should allow some judgment, and, as you say, we
should provide transparency with respect to which testing was done.

-- 
Mark Mitchell
CodeSourcery
mark@codesourcery.com
(650) 331-3385 x713