Comment 20 for bug 488354

(In reply to comment #14)
> The undeniable advantage of the C code is that the compiler will just use the
> right construct depending on the target architecture. No need to #ifdef
> __thumb__ or whatever.

Oh, certainly. C code is always preferable to assembly for that very reason, unless you need to do something that C code can't express efficiently (though that isn't the case here).

I'm still not entirely convinced about the SP-fiddling though. It seems unsafe to me. For instance, how do you know that the stack_space array is at the bottom of your frame? C doesn't guarantee anything there (even though you declare it last). Yes, it probably works, but it feels unsafe. That's why the original was in assembly, so it was clear what was happening.

In the case of writing to SP in the asm block: Yes, the compiler knows you've clobbered it, but C mandates that you have a stack and I'd be surprised if the compiler copes well with you doing that. If it isn't using a frame pointer, for example, it won't know how to wind back the stack pointer. Perhaps writing to SP in this way forces it to use a frame pointer, but none of that behaviour is documented (as far as I know) so it's likely to break between even minor compiler revisions.

I like the C implementation, but it does make me feel nervous :-)