[PATCH] PR 17115: Do not warn for noinline functions

[Giovanni Bajo]

Jason Merrill wrote:

> Hmm, the noinline attribute is currently local to the front ends.
> This seems safe enough,

I was mimicing the way we were handling attribute always_inline a few lines
above. Actually, it looks weird to me that noinline and always_inline are
handled in such a different way (one totally inside the frontends, one in the
middle end?). Or maybe I am still missing something...

Is there a good reason to keep attribute noinline within the frontends?

> but perhaps it would be better to make "user
> explicitly suppressed inlining" another flag in cgraph_local_info,
> and give it its own reason string, and check for that reason in
> expand_call_inline rather than look for the attribute.

I see. I do not have time to do this at this point, though, my backlog is long
enough. If you feel the patch is incorrect, I will add this suggestion to
Bugzilla, and someone can revisit this later. Notice this is also a regression
on 3.3 and 3.4, so maybe my patch makes sense at least there.

Giovanni Bajo