Syncing platform-specific bits with generic code (was: Mark inputsbeing ZERO as unlikely __mul in powerpc)

[Siddhesh Poyarekar <siddhesh dot poyarekar at gmail dot com>]

On 20 February 2013 21:59, Carlos O'Donell <carlos@redhat.com> wrote:
> If you want you can kick-off a question regarding concensus:
> "If performance is not degraded, can patches that synchronize
>  machine copies of code with generic copies of code be committed
>  without review?"

That would definitely be more convenient.  In fact, I was wondering if
we could follow the gdb-style definition of Obvious Changes:

"All maintainers listed in this file, including the Write After Approval
developers, are allowed to check in obvious fixes.

An "obvious fix" means that there is no possibility that anyone will
disagree with the change.

A good mental test is "will the person who hates my work the most be
able to find fault with the change" - if so, then it's not obvious and
needs to be posted first. :-)

Something like changing or bypassing an interface is _not_ an obvious
fix, since such a change without discussion will result in
instantaneous and loud complaints.

For documentation changes, about the only kind of fix that is obvious
is correction of a typo or bad English usage."

I know it could lead to more situations where reviewers disagree on a
change being obvious and look for changes to the commit, etc. but
that's not that bad is it?  I had committed changes to gdb that some
maintainers did not like and all it took to resolve the situation was
someone to sound that opinion, a discussion and a commit revert.

The above text seems to cover most of the bits we already have in the
Consensus wiki page.  In addition, things like fixing trivial build
warnings also get included.

-- 
http://siddhesh.in