Skip to contentSkip to conversation

macro in linux/compiler.h pollutes gcc __attribute__ namespace

DISCUSSION Timmy Douglas 3 participants 5 messages May 4, 2005 – May 5, 2005

lore
Share & tools
lore
#0 lore raw

(I'm not subscribed so please CC me replies that you want me to reply to.)

Recently I've found a problem with emacs where gcc optimizes a function to be inline where it shouldn't be. The emacs developers use a macro like this:

#define NO_INLINE __attribute__((noinline))

that would normally work fine but when we compile the file with NO_INLINE, the -E output looks like:

static void __attribute__(())
x_error_quitter (display, error)

     Display *display;
     XErrorEvent *error;

{
char buf[256], buf1[356];

...etc

I've realized that this file includes linux/compiler.h which does:

139

   140  #ifndef noinline
   141  #define noinline
   142  #endif

143

which causes __atribute__((noinline)) to change into __attribute__(()). I'm not sure how linux developers keep a function from getting inlined, but I'm hoping someone will consider removing or changing this macro.

#1 lore raw
Quoted 31 lines · Timmy Douglas

(I'm not subscribed so please CC me replies that you want me to reply to.)

Recently I've found a problem with emacs where gcc optimizes a function to be inline where it shouldn't be. The emacs developers use a macro like this:

#define NO_INLINE __attribute__((noinline))

that would normally work fine but when we compile the file with NO_INLINE, the -E output looks like:

static void __attribute__(())
x_error_quitter (display, error)

     Display *display;
     XErrorEvent *error;

{
char buf[256], buf1[356];

...etc

I've realized that this file includes linux/compiler.h which does:

   139
   140  #ifndef noinline
   141  #define noinline
   142  #endif
   143
Quoted 4 lines

which causes __atribute__((noinline)) to change into __attribute__(()). I'm not sure how linux developers keep a function from getting inlined, but I'm hoping someone will consider removing or changing this macro.

The right question to be asking is why is emacs including kernel headers?

--
Brian Gerst

#2 lore raw
Quoted 15 lines · Brian Gerst
Quoted 7 lines · Timmy Douglas

(I'm not subscribed so please CC me replies that you want me to reply to.)
Recently I've found a problem with emacs where gcc optimizes a function to be inline where it shouldn't be. The emacs developers use a macro like this:
[snip]
I've realized that this file includes linux/compiler.h which does:

Quoted 6 lines
   139
   140  #ifndef noinline
   141  #define noinline
   142  #endif
   143

[snip]

Quoted 1 line

The right question to be asking is why is emacs including kernel headers?

I'm guessing it goes sort of like this:

signal.h -> bits/sigcontext.h -> asm/sigcontext.h -> linux/compiler.h

#3 lore raw
Quoted 3 lines · Timmy Douglas

I'm guessing it goes sort of like this:

signal.h -> bits/sigcontext.h -> asm/sigcontext.h -> linux/compiler.h

Installing headers directly from the kernel has been deprecated for quite a while. Please look for the "linux-kernel-headers" package in whatever your package management system is. It has a set of cleaned headers. IIRC, there were some proposals a while back for how to fix this and make a set of headers useable from userspace, but nothing substantial ever got done.

Cheers,
Kyle Moffett

-----BEGIN GEEK CODE BLOCK-----
Version: 3.12
GCM/CS/IT/U d- s++: a18 C++++>$ UB/L/X/*++++(+)>$ P+++(++++)>$ L++++(+++) E W++(+) N+++(++) o? K? w--- O? M++ V? PS+() PE+(-) Y+ PGP+++ t+(+++) 5 X R? tv-(--) b++++(++) DI+ D+ G e->++++$ h!*()>++$ r !y?(-) ------END GEEK CODE BLOCK------

#4 lore raw
Quoted 11 lines · Kyle Moffett
Quoted 3 lines · Timmy Douglas

I'm guessing it goes sort of like this:

signal.h -> bits/sigcontext.h -> asm/sigcontext.h -> linux/compiler.h

Installing headers directly from the kernel has been deprecated for quite a while. Please look for the "linux-kernel-headers" package in whatever your package management system is. It has a set of cleaned headers. IIRC, there were some proposals a while back for how to fix this and make a set of headers useable from userspace, but nothing substantial ever got done.

Thanks for the input. It seems this is a gentoo bug then because it leaves this #define in their version of the source (even though it removes some other things). I don't see this problem on a debian box I have though. Too bad this wasn't standardized on.

Thanks again.