On Mon, Apr 13, 2020 at 03:40:55PM -0700, Linus Torvalds wrote:
> On Mon, Apr 13, 2020 at 12:32 PM Linus Torvalds
> <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> >
> > I _feel_ like the fix to that should be that the only thing that
> > creates the actual symbol is the label definition, and that the goto
> > should only ever use the 'ident' and we'd tie the two together later.
>
> Actually, how about something like this?
>
> I've not signed off on these patches, and the commit logs are
> questionable, but part of that is that the two first ones are just
> quick-and-dirty versions of your rename cleanups.
>
> The third patch is the serious one, which shows what I think might be
> the solution to the odd scoping rules for labels.
>
> Basically, we always scope labels - but if a label is _used_ but not
> defined in an inner label, when we close the label scope, we move it
> out to the next level.
>
> But a defined label is never moved out, and when we define it, we
> require that any previous use was in the same scope (where "same
> scope" might have been an inner scope that was moved out).
>
> I think it gets the semantics right, and it's actually fairly simple.
>
> But it has very little testing, so this is more of a "how about
> something like this" than a serious submission.
>
> If you test it, and fix up the warnings and error cases (like the
> other patches in your series did), you are more than welcome to take
> credit and authorship for this.
>
> I just felt that the best way to describe (and do _some_ testing) my
> idea was to have a quick implementation to show what I mean.
>
> And by "_some_ testing" I literally mean "almost no testing at all". I
> didn't even run this on the kernel tree. I just used one stipid small
> test-case for this, and when it gave the warning I wanted, I said
> "good enough" and sent this email out ;)
I like the idea. I just gave it a very quick test with sparse's
"make check" (it covers a lot of simple but corner/dirty cases that
the kernel may/should not have). It seemed to pass all the tests but
the ones using __label__. For exemple, things like this complain:
{
__label__ l;
l:
goto l;
}
I'll look more at it tommorow as it's a bit late here. I just fear
that __label__ will spoil things here or at least complicate them.
-- Luc
