Re: randconfig build error with next-20150316, in samples/kdbus/kdbus-workers

[Michael Ellerman <mpe@ellerman.id.au>]

On Tue, 2015-03-17 at 00:00 +0100, David Herrmann wrote:
> Hi
> 
> On Mon, Mar 16, 2015 at 11:51 PM, Michael Ellerman <mpe@ellerman.id.au> wrote:
> > On Mon, 2015-03-16 at 23:27 +0100, David Herrmann wrote:
> >> The uapi-include only causes the warning, not the build failure.
> >
> > I don't know how you came to that conclusion?
> >
> > It fails looking for linux/compiler.h, which is only included from the kernel
> > headers, never from the exported headers.
> 
> We only include linux/kdbus.h. On sanitized headers, this will include
> linux/types.h -> linux/posix_types.h -> linux/stddef.h.
> If you use the uapi headers, then stddef.h is not sanitized and will
> still include linux/compiler.h (which is removed on sanitized
> headers). Hence, this error only occurs if you include
> uapi/linux/kdbus.h. With sanitized headers in ./usr/, the compiler
> will prefer ./usr/linux/kdbus.h and ./usr/linux/stddef.h, thus never
> including any linux/compiler.h.

Right. But if the uapi include isn't there, you don't get this build failure.
This build failure is caused by the uapi include.

> If you drop -I./include/uapi/, you will get "linux/kdbus.h not found".

Yep, which is exactly correct (unless it's provided by your distro headers).

> The error will be different, but it will not fix anything. However, if
> you run "make headers_install", everything will compile just fine even
> with -I./include/uapi/.

Well probably. But I could probably construct a scenario where something gets
pulled in from uapi, and then you've got a mixture of user & kernel headers
again.

At the end of the day there is no good reason to use include/uapi, I think we
agree on that :)

cheers