Re: git: 402dbdd98acc - main - Adjust function definition in arm's mv_common.c to avoid clang 15 warning

On 8/16/22 7:51 AM, Warner Losh wrote:
> On Tue, Aug 16, 2022, 8:00 AM Cy Schubert <Cy.Schubert@cschubert.com> wrote:
> 
>> In message <YvuD7+oK6RZ/DLzH@FreeBSD.org>, Alexey Dokuchaev writes:
>>> On Tue, Aug 16, 2022 at 12:10:04PM +0200, Dimitry Andric wrote:
>>>> ...
>>>> But I think it is better to have the definitions matching the
>>>> declarations exactly. We should sweep through the whole tree and get
>>>> rid of all K&R functions too. I believe Warner wanted to attempt that.
>>>
>>> I won't comment on the technical side of things, but seeing this plethora
>>> of identical commits is not just annoying, but pessimizes blaming as
>> well.
>>> Why can't it all be done in more coarse pieces, if not one commit?
>>
>> Agreed.
>>
> 
> I'd prefer one burst. Cherry picking large commit that have conflicts is a
> real pain. Especially since the pain is different for 12 and 13. Especially
> since things get MFCd at different rates by different people. As someone
> who has had to sort that out multiple times, I know I've spent quite a bit
> more time unwinding one larger commit that was partially merged before I
> got there. It was terrible. It creates a lot of extra work for the mergers.
> Think boot loader that takes a while to settle before a lot of changes are
> mfc'd due to its huge number of combinatorics (hundreds of boot
> combinations) which take a while to have enough testing to know we've
> mitigated the risk as best we can, not all by me or even all in src/stand.
> With lots of commits, the lists are long but easily automated with any
> conflicts being quick and fast to resolve. Larger commits are bigger
> efforts to resolve making it harder to provide good support to old
> branches. And that's already hard enough.
> 
> Also with one burst, it's just a range delete in my email client once.

FWIW, what I did with -Wset-but-unused was to work on them as individual
commits in a branch for a couple of days to fix all the issues.  I then
got the non-trivial cases reviewed and tried to do large blasts of the
remaining ones.  The DRIVER_MODULE sweep was even more this way.  I
think for this one I ended doing 50-commits-at-a-time bursts at the
end for the trivial ones.

-- 
John Baldwin