any value to numerous checks of "#ifdef __GNUC__"?

any value to numerous checks of "#ifdef __GNUC__"?

[Robert P. J. Day]

  there are numerous header files that contain a check of the form

    #ifdef __GNUC__

but <linux/compiler.h> already insists on a GNU C-flavoured compiler:

#if __GNUC__ >= 4
# include <linux/compiler-gcc4.h>
#elif __GNUC__ = 3 && __GNUC_MINOR__ >= 2
# include <linux/compiler-gcc3.h>
#else
# error Sorry, your compiler is too old/not recognized.
#endif

  so is there any point in keeping all those checks around?  is it
even remotely feasible that the kernel might someday support a non-GNU
C compiler, given the numerous gcc extensions that are used?

rday
--

====================================
Robert P. J. Day
Linux Consulting, Training and Annoying Kernel Pedantry:
    Have classroom, will lecture.

http://crashcourse.ca                          Waterloo, Ontario, CANADA
====================================

Re: any value to numerous checks of "#ifdef __GNUC__"?

[walter harms]

Robert P. J. Day wrote:
>   there are numerous header files that contain a check of the form
> 
>     #ifdef __GNUC__
> 
> but <linux/compiler.h> already insists on a GNU C-flavoured compiler:
> 
> #if __GNUC__ >= 4
> # include <linux/compiler-gcc4.h>
> #elif __GNUC__ = 3 && __GNUC_MINOR__ >= 2
> # include <linux/compiler-gcc3.h>
> #else
> # error Sorry, your compiler is too old/not recognized.
> #endif
> 
>   so is there any point in keeping all those checks around?  is it
> even remotely feasible that the kernel might someday support a non-GNU
> C compiler, given the numerous gcc extensions that are used?
> 
> rday
> --
> 

hi,
so far i know the intel compiler can compile a linux kernel.
ntl. this is most likely crap since the compiler is checked much more early.
(or at least should).
I only reason i can think of is that this module is used in other places outside
the linux kernle tree.

re,
 wh

Re: any value to numerous checks of "#ifdef __GNUC__"?

[Robert P. J. Day]

On Fri, 4 Apr 2008, walter harms wrote:

> Robert P. J. Day wrote:
> >   there are numerous header files that contain a check of the form
> >
> >     #ifdef __GNUC__
> >
> > but <linux/compiler.h> already insists on a GNU C-flavoured compiler:
> >
> > #if __GNUC__ >= 4
> > # include <linux/compiler-gcc4.h>
> > #elif __GNUC__ = 3 && __GNUC_MINOR__ >= 2
> > # include <linux/compiler-gcc3.h>
> > #else
> > # error Sorry, your compiler is too old/not recognized.
> > #endif
> >
> >   so is there any point in keeping all those checks around?  is it
> > even remotely feasible that the kernel might someday support a non-GNU
> > C compiler, given the numerous gcc extensions that are used?
> >
> > rday
> > --
> >
>
> hi,
> so far i know the intel compiler can compile a linux kernel.

i know, but even the intel compiler identifies itself as
GNU-compatible by defining __GNUC__.

> ntl. this is most likely crap since the compiler is checked much
> more early. (or at least should).

it is, as i point out above:  linux/compiler.h.

> I only reason i can think of is that this module is used in other
> places outside the linux kernle tree.

perhaps, but unlikely, given the amount of checking that's being done
all over the place:

$ grep -r "#if.*__GNUC__" *

and, philosophically speaking, it's not clear that source in the
kernel tree is responsible for being compatible for being used in
environments *outside* of the tree.

i'm going to go with the position that it's crap and submit a few
patches to start ripping it out.

rday
--


====================================
Robert P. J. Day
Linux Consulting, Training and Annoying Kernel Pedantry:
    Have classroom, will lecture.

http://crashcourse.ca                          Waterloo, Ontario, CANADA
====================================

Re: any value to numerous checks of "#ifdef __GNUC__"?

[Robert P. J. Day]

On Fri, 4 Apr 2008, walter harms wrote:

> hi, so far i know the intel compiler can compile a linux kernel.
> ntl. this is most likely crap since the compiler is checked much
> more early. (or at least should).

actually, those checks could be reasonable for content exported to
userspace, but if it's not exported, i don't see any value to those
tests.

rday
--

====================================
Robert P. J. Day
Linux Consulting, Training and Annoying Kernel Pedantry:
    Have classroom, will lecture.

http://crashcourse.ca                          Waterloo, Ontario, CANADA
====================================

Re: any value to numerous checks of "#ifdef __GNUC__"?

[Darren Jenkins]

On Sat, Apr 5, 2008 at 12:18 AM, Robert P. J. Day <rpjday@crashcourse.ca> wrote:
>  is it
>  even remotely feasible that the kernel might someday support a non-GNU
>  C compiler, given the numerous gcc extensions that are used?

I think people are working on getting tinycc to compile an un-modified
kernel, and I think LLVM is close-ish, also some BSD guys are working
on PCC so it might be able to compile the kernel in the
(distant?)future.

It concerns me a little that GCC is the only OSS compiler that can
compile the kernel.


Darren J.

Re: any value to numerous checks of "#ifdef __GNUC__"?

[Robert P. J. Day]

On Sat, 5 Apr 2008, Darren Jenkins wrote:

> On Sat, Apr 5, 2008 at 12:18 AM, Robert P. J. Day <rpjday@crashcourse.ca> wrote:

> > is it even remotely feasible that the kernel might someday support
> > a non-GNU C compiler, given the numerous gcc extensions that are
> > used?
>
> I think people are working on getting tinycc to compile an
> un-modified kernel, and I think LLVM is close-ish, also some BSD
> guys are working on PCC so it might be able to compile the kernel in
> the (distant?)future.
>
> It concerns me a little that GCC is the only OSS compiler that can
> compile the kernel.

well, as i mentioned earlier, the intel compiler is also officially
approved, but it defines __GNUC__ so it masquerades as gcc, anyway.

and, as a followup to my original question, once i looked closer, it
does look like most of that "#ifdef __GNUC__" checking does get
exported to userspace, quite possibly for exactly the reason darren
mentions -- to support *userspace* non-gcc compilation.  i'm sure
there are *some* superfluous in-kernel __GNUC__ tests, but probably
nowhere near as many as i first suspected.

rday
--

====================================
Robert P. J. Day
Linux Consulting, Training and Annoying Kernel Pedantry:
    Have classroom, will lecture.

http://crashcourse.ca                          Waterloo, Ontario, CANADA
====================================