Re: [announce] new tree: "fix all build warnings, on all configs"

[Roland Dreier <rdreier@cisco.com>]

 > OTOH, there should be a well-defined work flow to keep this all 
 > manageable: once we know why a warning triggers and it has been 
 > categorized by a human, we should get rid of the warning in some way. 
 > 
 > Applying this patch as-is would be one option. Annotating it with a 
 > specific gcc version would be overkill i think. Ignoring it would be 
 > bad, because there's real value in standardizing on a "no warnings" 
 > build output. Many new warnings get introduced because people do not 
 > notice new warnings amongst the very high baseline noise of the kernel 
 > build.

The specific change I noticed:

 > -						       (1 << MGM_BLCK_LB_BIT));
 > +						       (1U << MGM_BLCK_LB_BIT));

is not a problem to me -- the code is fine either way, and if we're
making an effort to kill all warnings, then I'm OK with merging it.
It's a little unfortunate to add churn due to a gcc bug that is only in
certain 4.3 releases, but this particular case doesn't seem to trigger
in many places, so the cost is low.

However I worry about warnings produced by gcc bugs forcing us to tinker
with correct code.  Maybe it just makes sense to wait and see if we ever
hit a case where a gcc bug forces us to make too many stupid changes,
and figure out what to do if and when that happens.

 - R.