https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98734
--- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #5)
> Note, always_inline is not a guarantee out of line copy will not be emitted.
> Either because address of the always_inline function is taken and not
> optimized in all places into a direct call, or when using
> -fkeep-inline-functions.
> So, not emitting -Wpsabi in those cases looks wrong (sure, perhaps it could
> be somehow restricted to the cases where the function body is actually
> emitted, so for DECL_DISREGARD_INLINE_LIMITS don't warn when processing the
> function declaration and repeat the diagnostics for
> DECL_DISREGARD_INLINE_LIMITS functions when emitting the body.
> Though wonder if that would cover the always_inline,gnu_inline extern inline
> cases where we emit a call to the external definition.
I think the warning should be emitted only when the function body is emitted
and is either address taken or has non-local linkage.
Note always-inline is
DECL_DISREGARD_INLINE_LIMITS && lookup_attribute ("always_inline", ...
only for those inlining is guaranteed (apart from the "bugs" with
address-taking that we fail to diagnose).
I suppose cgraph_node->can_change_signature might be an indicator that the ABI
(and thus related warnings) is not relevant.
