[Toybox] [PATCH] Fix devmem build with clang.

[Rob Landley]

On 11/8/24 08:10, enh wrote:
> hmm... thinking about this more we don't actually need this condition
> 
>        else if (sizeof(long)==8 && bytes==8) data = *(unsigned long *)p;
> 
> because we already check that we don't let you set bytes to 8 if you're on
> ILP32 (which is good, because _that_ would be a bug).
> 
> so we can just write
> 
>        else data = *(unsigned long *)p;

It was a compile time optimization to let dead code elimination yank a 
tiny amount of code because sizeof(long) is a compile type constant. I 
mostly don't do that level of micromanagement anymore because busybox 
exists and compilers sometimes get smarter. Well, they change their 
optimizations around a lot. Somewhat randomly. And get lucky sometimes.

> here (and in the other "copy" of this if above) and make this go away with
> less code? if we do that, we can remove the QUIET from data completely
> (because at least with `make defconfig` gcc 13 doesn't complain about this
> code at all).

I already applied the patch.

And I should try the "commit to a branch then delete the branch" trick 
to close https://github.com/landley/toybox/issues/520 now I've lost web 
login access.

I hadn't gotten around to that until you poked me because last I checked 
LLVM was setting __gnuc__ so this didn't bite in NDK testing, and I 
thought they were talking about qnx or something. Which I should deal 
with before cutting a release but isn't otherwise time critical because 
I don't have a test environment for it so never know when I break it (or 
fix it)...

I'm guessing one of the updates finally yanked that define?

> and if we do _that_, then i wonder whether we should keep clang on the
> "no-op QUIET" path until/unless we really do need it, to force us to
> continue to look at and think about these things? there are very few uses
> of QUIET, and it does seem like something we should avoid...

It would be nice if gcc would just stop producing unsilenceable false 
positives because it can't understand the relationship between two 
variables. Alas, 7 year time horizon also means building with older gcc, 
and a LOT of cross compilers aren't the latest and greatest because 
updating just causes pain when upstream doesn't know you exist and isn't 
happy to hear from you...

Anyway, commit 542f62759a779

Rob