Re: Probable multiarch related problem in finding header file posix_types_32.h
On Tue, Feb 14, 2012 at 09:29:30AM -0800, Steve Langasek wrote:
> On Tue, Feb 14, 2012 at 12:34:41AM -0800, Russ Allbery wrote:
> > > and goes on an on, repeating the error about posix_types_32.h.
> > > -------------------------------------------------------------------
>
> > > It would help if you could confirm that this is actually an issue caused
> > > by multiarch migration and what kind of changes might help.
>
> > This looks like some sort of more prosaic problem to me. Somehow,
> > posix_types.h was included properly (which means that you have
> > linux-libc-dev installed and the compiler is searching the arch-specific
> > paths for headers), but it can't find posix_types_32.h. On an i386
> > system, both of those files are in linux-libc-dev.
>
> > The complete contents of posix_types.h is:
>
> > # ifdef __i386__
> > # include "posix_types_32.h"
> > # else
> > # include "posix_types_64.h"
> > # endif
>
> > So, for some reason, include of a file with "" is not searching the same
> > directory that the header file itself was found in.
>
> > One of the things that's missing from the above is any hint of what
> > compiler flags are in use. The gcc flag -I- can cause this behavior since
> > it changes the handling of #include "".
>
> Where we've run across similar problems with posix_types.h in the recent
> past, it has indeed been due to the use of "gcc -I-".
Wow, that is a really insane option. However it is documented as
deprecated.
> This is tied to
> multiarch, in that moving linux-libc-dev headers to the multiarch include
> directory triggered the problem; ultimately though I think this is a bug in
> linux-libc-dev for using #include "" here.
"wontfix"
Ben.
> In any case it can usually be worked around by not using -I-.
>
> Cf. bug #637419.
--
Ben Hutchings
We get into the habit of living before acquiring the habit of thinking.
- Albert Camus
Reply to: