More .so hell

[Dan Kegel]

On 2/9/06, securehell@hushmail.com <securehell@hushmail.com> wrote:
> This might be more appropriate for a GCC list

Yes, it would.

> I have a header file that includes the code for a funcion and is
> declared __forceinline on a different (Windows) platform. I am
> trying to get it to build on Linux and have tried __forceinline,
> __inline__, inline, and __attribute__((always_inline)) but gcc
> 3.4.4 doesn't seem to compile it as the header is called several
> times by various .c files.

http://lkml.org/lkml/2006/1/6/84 suggests

#define __always_inline		inline __attribute__((always_inline))

> For example: __attribute__(allways_inline)) gave me this:
> file.h:64: sorry, unimplemented: inlining failed in call to
> 'your_function': function not inlinable

That might make sense if you had the __attribute__((always_inline))
but forgot to also say inline.

See also
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14950
which says a similar problem (perhaps the same one)
is fixed in gcc-4.0 but won't be fixed in 3.4.

If that doesn't help enough, please post a minimal test
case.

There are quite a few other bugzilla entries about
inlining that make good reading, too.
- Dan

--
Wine for Windows ISVs: http://kegel.com/wine/isv