Re: [PATCH 5.7 000/121] 5.7.13-rc2 review

[Linus Torvalds <torvalds@linux-foundation.org>]

On Wed, Aug 5, 2020 at 4:20 AM Marc Zyngier <maz@kernel.org> wrote:
>
> I came up with yet another "quality" hack, which gets the job done,
> see below. It is obviously much simpler, but also terribly ugly.

This is effectively what the approach of commit c0842fbc1b18
("random32: move the pseudo-random 32-bit definitions to prandom.h")
was too. It has the prandom include in the middle of random.h - which
makes sense in the context of that patch (because it's where all the
prandom stuff used to live there), but the real secret is that as it
does that, it basically moves the percpu.h include down there too.

This mess has caused me to seriously look at making sparse generate a
"defined here, used here" list, and do a topo-sort on it all, and warn
about unused includes but also about stupidities like this.

But I won't have time to do that until after the merge window, and it
may turn out nastier than expected. But I did add all the include
chaining information to the sparse position logic recently in a fairly
easy to digest form, so my gut feel is that something that doesn't try
to actually optimize or solve the problem - just point out the obvious
errors - might be more reasonably doable than doing something that is
theoretically the RightThing(tm) to do.

IOW, a tool only to help find and figure these things out, rather than
a tool to do all the minimization for you automatically.

We'll see. No promises. And if somebody else beats me to it, I won't complain.

            Linus