why "no new warnings" is good. was: regclass changes reverted

[Robert Lipe]

rth> They are responsible for the bootstrap comparison
rth> failures that folks have been reporting for x86.

jh> See the mailing list - there is already patch to fix the problem. It
jh> was caused by using uninitialized variable.  I am just in the middle

Remember the dust I kicked up a few weeks ago suggesting a policy that
all new code be required to introduce no new warnings?  Though that
conversation fizzled out, this serves as a good reminder why that policy
should be implemented.

There's already one that says changes must bootstrap; perhaps this
didn't break all targets or that uninitialized variable just happened to
be correctly initialized on the test systems.

RJL