At Wed, 11 Mar 2015 10:46:29 +0100, Arnd Bergmann wrote:
On Wednesday 11 March 2015 07:11:18 Takashi Iwai wrote:
At Wed, 11 Mar 2015 03:22:04 +0200, Mikko Rapeli wrote:
On Tue, Feb 17, 2015 at 07:27:38AM +0100, Takashi Iwai wrote:
At Tue, 17 Feb 2015 00:05:44 +0100, Mikko Rapeli wrote:
The DECLARE_BITMAP macro is not available in userspace headers. Fixes userspace compile error: error: expected specifier-qualifier-list before ‘DECLARE_BITMAP’
It's nonsense. This results in an incompatible structure, thus ABI would be broken completely (actually this will break the compile of ld10k1).
None of the exported headers after 'make headers_install' have definition of DECLARE_BITMAP macro. It is defined in include/linux/types.h which is different from include/uapi/linux/types.h and missing this definition and a few other things.
One option would be add DECLARE_BITMAP macro to include/uapi/linux/types.h and add include/linux/bitops.h to uapi.
Thoughts?
Are there any other headers like that? If this is the only one, leave it as is. The only program that reads this are some alsa-tools ones and they have already own DECLARE_BITMAP() definition. Adding the extra definition here will even break the compilation out of sudden.
I think it's a worthy goal to have the header files be compilable standalone,
In general yes, but this case is very minor issue: - the file in question is for a hardware device-specific data definition, - there are only two programs read this file, both can be built properly, - and the device and the programs are very old, modifying such need extra care.
but I don't think we should make the DECLARE_BITMAP() macro globally visible in user space, in particular because it will clash with every instance in which user space has a macro of the same name.
What we could do here is to add a private copy of the macro to emu10k1.h under a different name, such as __EMU10K1_DECLARE_BITMAP().
Yes, it's a better option.
thanks,
Takashi