Architecture specific header files

[Vít Ondruch]

Hi,

Is there some common practice, where to place architecture specific 
header files? From output of the following command, I can't see any such 
place.


$ `gcc -print-prog-name=cc1` -v
ignoring nonexistent directory 
"/usr/lib/gcc/x86_64-redhat-linux/4.8.1/include-fixed"
ignoring nonexistent directory 
"/usr/lib/gcc/x86_64-redhat-linux/4.8.1/../../../../x86_64-redhat-linux/include"
#include "..." search starts here:
#include <...> search starts here:
  /usr/lib/gcc/x86_64-redhat-linux/4.8.1/include
  /usr/local/include
  /usr/include
End of search list.



Actually, the "ignoring nonexistent directory 
"/usr/lib/gcc/x86_64-redhat-linux/4.8.1/../../../../x86_64-redhat-linux/include""

might be usable, but is it allowed to use it? Does it conform to FHS 
(don't think so). Shouldn't it be configured differently?

The point is, that compilation of this simple program:


$ cat > rt.c
#include "ruby.h"

int main(int argc, char **argv)
{
   return 0;
}
^D



fails:


$ gcc rt.c
In file included from /usr/include/ruby.h:33:0,
                  from rt.c:1:
/usr/include/ruby/ruby.h:24:25: fatal error: ruby/config.h: No such file 
or directory
  #include "ruby/config.h"
                          ^
compilation terminated.


This should work out of the box IMO, without any configuration or 
specification of additional paths etc. The problem is, that Ruby ships 
architecture specific config.h header which is not available in GCC's 
default search path and I am wondering how to make this work.


Thanks



Vít

[Petr Pisar]

On 2013-06-25, Vít Ondruch <vondruch(a)redhat.com&gt; wrote:
>  Is there some common practice, where to place architecture specific 
>  header files? From output of the following command, I can't see any such 
>  place.
>  
I dont't think so. SDL moves architecture specific header file into
a private directory:

# Rename SDL_config.h to SDL_config-<arch>.h to avoid file conflicts on
# multilib systems and install SDL_config.h wrapper
mv %{buildroot}/%{_includedir}/SDL/SDL_config.h
%{buildroot}/%{_includedir}/SDL/SDL_config-%{_arch}.h
install -m644 %{SOURCE1} %{buildroot}/%{_includedir}/SDL/SDL_config.h

And installs a wrapper (SOURCE1) to replace the original location. The
wrapper is just a bunch of #if branches:

#if defined(__i386__)
#include "SDL_config-i386.h"
#elif defined(__ia64__)
#include "SDL_config-ia64.h"
[...]
#endif

Though a standardized path would be nice.

-- Petr

[Björn Persson]

Petr Pisar wrote:
>  On 2013-06-25, Vít Ondruch <vondruch(a)redhat.com&gt; wrote:
>  > Is there some common practice, where to place architecture specific 
>  > header files? From output of the following command, I can't see any
>  > such place.
>  >
>  I dont't think so. SDL moves architecture specific header file into
>  a private directory:
>  
>  # Rename SDL_config.h to SDL_config-<arch>.h to avoid file conflicts
>  on # multilib systems and install SDL_config.h wrapper
>  mv %{buildroot}/%{_includedir}/SDL/SDL_config.h
>  %{buildroot}/%{_includedir}/SDL/SDL_config-%{_arch}.h
>  install -m644 %{SOURCE1} %{buildroot}/%{_includedir}/SDL/SDL_config.h
>  
>  And installs a wrapper (SOURCE1) to replace the original location. The
>  wrapper is just a bunch of #if branches:
>  
>  #if defined(__i386__)
>  #include "SDL_config-i386.h"
>  #elif defined(__ia64__)
>  #include "SDL_config-ia64.h"
>  [...]
>  #endif 
As another example, Glib puts an architecture-specific header under
libdir and expects you to use pkg-config to make the compiler find it:

$ rpmls glib2-devel | egrep 'glibconfig.h|glib-2.0.pc'
-rw-r--r--  /usr/lib64/glib-2.0/include/glibconfig.h
-rw-r--r--  /usr/lib64/pkgconfig/glib-2.0.pc
-rw-r--r--  /usr/lib/glib-2.0/include/glibconfig.h
-rw-r--r--  /usr/lib/pkgconfig/glib-2.0.pc
$ pkg-config --cflags glib-2.0 
-I/usr/include/glib-2.0 -I/usr/lib64/glib-2.0/include

Björn Persson

[Vít Ondruch]

Dne 25.6.2013 22:37, Björn Persson napsal(a):
>  Petr Pisar wrote:
> > On 2013-06-25, Vít Ondruch <vondruch(a)redhat.com&gt; wrote:
> >> Is there some common practice, where to place architecture specific
> >> header files? From output of the following command, I can't see any
> >> such place.
> >>
> > I dont't think so. SDL moves architecture specific header file into
> > a private directory:
> >
> > # Rename SDL_config.h to SDL_config-<arch>.h to avoid file conflicts
> > on # multilib systems and install SDL_config.h wrapper
> > mv %{buildroot}/%{_includedir}/SDL/SDL_config.h
> > %{buildroot}/%{_includedir}/SDL/SDL_config-%{_arch}.h
> > install -m644 %{SOURCE1} %{buildroot}/%{_includedir}/SDL/SDL_config.h
> >
> > And installs a wrapper (SOURCE1) to replace the original location. The
> > wrapper is just a bunch of #if branches:
> >
> > #if defined(__i386__)
> > #include "SDL_config-i386.h"
> > #elif defined(__ia64__)
> > #include "SDL_config-ia64.h"
> > [...]
> > #endif
>  As another example, Glib puts an architecture-specific header under
>  libdir and expects you to use pkg-config to make the compiler find it:
>  
Using external tools for configuration, that is what I want to avoid. Of 
course for Ruby, I could use mkmf, which will create suitable Makefile 
for me, or pkgconfig. But why should I create complex makefile or 
configuration script for such simple program I posted in my initial email.


Vít

[Vít Ondruch]

Dne 25.6.2013 15:41, Vít Ondruch napsal(a):
>  Hi,
>
>  Is there some common practice, where to place architecture specific 
>  header files? From output of the following command, I can't see any 
>  such place.
>
>
>  $ `gcc -print-prog-name=cc1` -v
>  ignoring nonexistent directory 
>  "/usr/lib/gcc/x86_64-redhat-linux/4.8.1/include-fixed"
>  ignoring nonexistent directory 
>
> "/usr/lib/gcc/x86_64-redhat-linux/4.8.1/../../../../x86_64-redhat-linux/include"
>  #include "..." search starts here:
>  #include <...> search starts here:
>   /usr/lib/gcc/x86_64-redhat-linux/4.8.1/include
>   /usr/local/include
>   /usr/include
>  End of search list.
>
>
>
>  Actually, the "ignoring nonexistent directory 
>
> "/usr/lib/gcc/x86_64-redhat-linux/4.8.1/../../../../x86_64-redhat-linux/include""
>
>  might be usable, but is it allowed to use it? Does it conform to FHS 
>  (don't think so). Shouldn't it be configured differently?
>
>  The point is, that compilation of this simple program:
>
>
>  $ cat > rt.c
>  #include "ruby.h"
>
>  int main(int argc, char **argv)
>  {
>    return 0;
>  }
>  ^D
>
>
>
>  fails:
>
>
>  $ gcc rt.c
>  In file included from /usr/include/ruby.h:33:0,
>                   from rt.c:1:
>  /usr/include/ruby/ruby.h:24:25: fatal error: ruby/config.h: No such 
>  file or directory
>   #include "ruby/config.h"
>                           ^
>  compilation terminated.
>
>
>  This should work out of the box IMO, without any configuration or 
>  specification of additional paths etc. The problem is, that Ruby ships 
>  architecture specific config.h header which is not available in GCC's 
>  default search path and I am wondering how to make this work.
>
>
>  Thanks
>
>
>
>  Vít 
Since it seems that this is quite common issue and there are more 
packages needing to workaround this issue (Python is not mentioned in 
this thread yet), I took the liberty to open RFE at GCC:

https://bugzilla.redhat.com/show_bug.cgi?id=979403


Vít

[Jakub Jelinek]

On Fri, Jun 28, 2013 at 02:58:12PM +0200, Vít Ondruch wrote:
>  Dne 25.6.2013 15:41, Vít Ondruch napsal(a):
>  >Is there some common practice, where to place architecture
>  >specific header files? From output of the following command, I
>  >can't see any such place. 
gcc doesn't have such location, you'd want a different path
for gcc -m64, different for gcc -m32, different for gcc -mx32, ....
The standard way is to add a wrapper header and include your arch specific
header based on compiler predefined macros.  Say something like:
# ifdef __i386__
#  include <asm/posix_types_32.h>
# elif defined(__ILP32__)
#  include <asm/posix_types_x32.h>
# else
#  include <asm/posix_types_64.h>
# endif
or, if possible, avoid arch specific headers altogether.

	Jakub

[Vít Ondruch]

Dne 28.6.2013 15:04, Jakub Jelinek napsal(a):
>  On Fri, Jun 28, 2013 at 02:58:12PM +0200, Vít Ondruch wrote:
> > Dne 25.6.2013 15:41, Vít Ondruch napsal(a):
> >> Is there some common practice, where to place architecture
> >> specific header files? From output of the following command, I
> >> can't see any such place.
>  gcc doesn't have such location 
Could you please consider to add such location?

Vít

[Vít Ondruch]

Dne 28.6.2013 15:04, Jakub Jelinek napsal(a):
>  On Fri, Jun 28, 2013 at 02:58:12PM +0200, Vít Ondruch wrote:
> > Dne 25.6.2013 15:41, Vít Ondruch napsal(a):
> >> Is there some common practice, where to place architecture
> >> specific header files? From output of the following command, I
> >> can't see any such place.
>  gcc doesn't have such location, you'd want a different path
>  for gcc -m64, different for gcc -m32, different for gcc -mx32, ....
>  The standard way is to add a wrapper header and include your arch specific
>  header based on compiler predefined macros.  Say something like:
>  # ifdef __i386__
>  #  include <asm/posix_types_32.h>
>  # elif defined(__ILP32__)
>  #  include <asm/posix_types_x32.h>
>  # else
>  #  include <asm/posix_types_64.h>
>  # endif
>  or, if possible, avoid arch specific headers altogether.
>
>  	Jakub 
BTW it seems that Debian already does this, although in completely 
different scale: http://wiki.debian.org/Multiarch/TheCaseForMultiarch


Vít