[PATCH] Fix PR c/9262

[Eric Botcazou]

> Eh.  I think I'd rather you did the split the other way around:
> create a flag_inlining_on_rtl which is frobbed by the front ends that
> want tree inlining, and leave flag_inline_functions named what it is
> right now.

The rationale was to keep the changes local to a single file, so as to
minimize the impact on the whole compiler. I thought this rationale was in
accordance with the current situation, because flag_inline_functions should
already have been renamed. Call that over-interpretation... or laziness :-)

> Also, does this effect mean that the front ends that use tree inlining
> ignore -finline-functions?

No, flag_inline_functions is correctly set by -finline-functions and taken
into account by the front-ends that use tree inlining. The problem is that
they unset it (rightfully of course) before the ouput of the assembly file
(in c-opts.c for example).

- Eric