[PATCH] Fix PR c/8518

[PATCH] Fix PR c/8518

[Eric Botcazou]

Hi,

This is a high-priority PR, regression from 3.0.x on gcc-3_2-branch and 
mainline. The compiler segfaults when building the initial CFG, but I think 
the problem comes from the front-end: the compiler gets confused because of 
the redefinition of an extern inline function:

extern int inline bar () { return 0; }
static int inline bar () { return foo(); }

bar() is deferred twice and is expanded twice by expand_deferred_fns() at the 
end of the file.

On the contrary, in the following slighly modified situation:

extern int inline bar () { return 0; }
static int baz () { return bar(); }
static int inline bar () { return foo(); }

bar() is deferred only once because the compiler authoritatively outlines the 
second definition of bar(), and the first bar() is thus never expanded.

The proposed fix is to even the things, i.e to authoritatively outline the 
second definition of bar() in the first situation too. Bootstrapped/regtested 
on i586-redhat-linux-gnu (c,c++,objc,f77).

Ok for mainline and for gcc-3_2-branch when it reopens ?

-- 
Eric Botcazou


2002-11-17  Eric Botcazou  <ebotcazou@libertysurf.fr>

	PR c/8518
	* c-decl.c (duplicate_decls): Outline the second definition
	of an extern inline function in all cases.


2002-11-17  Eric Botcazou  <ebotcazou@libertysurf.fr>

	* gcc.c-torture/compile/20021117-1.c: New test
	* gcc.c-torture/compile/20021117-2.c: New test


Index: c-decl.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/c-decl.c,v
retrieving revision 1.353
diff -c -r1.353 c-decl.c
*** c-decl.c	10 Nov 2002 16:24:24 -0000	1.353
--- c-decl.c	16 Nov 2002 21:59:43 -0000
***************
*** 1474,1482 ****
  	 inline, make sure we emit debug info for the inline before we
  	 throw it away, in case it was inlined into a function that hasn't
  	 been written out yet.  */
!       if (new_is_definition && DECL_INITIAL (olddecl) && TREE_USED 
(olddecl))
  	{
! 	  (*debug_hooks->outlining_inline_function) (olddecl);
  
  	  /* The new defn must not be inline.  */
  	  DECL_INLINE (newdecl) = 0;
--- 1474,1483 ----
  	 inline, make sure we emit debug info for the inline before we
  	 throw it away, in case it was inlined into a function that hasn't
  	 been written out yet.  */
!       if (new_is_definition && DECL_INITIAL (olddecl))
  	{
! 	  if (TREE_USED (olddecl))
! 	    (*debug_hooks->outlining_inline_function) (olddecl);
  
  	  /* The new defn must not be inline.  */
  	  DECL_INLINE (newdecl) = 0;


--- /dev/null   Thu Apr 11 16:25:15 2002
+++ gcc.c-torture/compile/20021117-1.c  Sun Nov 17 02:21:30 2002
@@ -0,0 +1,9 @@
+/* PR c/8518 */
+/* Contributed by Volker Reichelt. */
+
+/* Verify that GCC doesn't get confused by the
+   redefinition of an extern inline function. */
+
+extern int inline foo () { return 0; }
+extern int inline bar () { return 0; }
+static int inline bar () { return foo(); }


--- /dev/null   Thu Apr 11 16:25:15 2002
+++ gcc.c-torture/compile/20021117-2.c  Sun Nov 17 02:21:57 2002
@@ -0,0 +1,9 @@
+/* PR c/8518 */
+/* Contributed by Volker Reichelt. */
+
+/* Verify that GCC doesn't get confused by the
+   redefinition of an extern inline function. */
+
+extern int inline foo () { return 0; }
+extern int inline bar () { return 0; }
+static int bar () { return foo(); }

Re: [PATCH] Fix PR c/8518

[Richard Henderson]

On Sun, Nov 17, 2002 at 02:23:19AM +0100, Eric Botcazou wrote:
> ... the compiler gets confused because of 
> the redefinition of an extern inline function:
> 
> extern int inline bar () { return 0; }
> static int inline bar () { return foo(); }

Incidentally, this seems like error material.  The redefinition of an
extern inline function should be neither static nor inline itself.

> 	PR c/8518
> 	* c-decl.c (duplicate_decls): Outline the second definition
> 	of an extern inline function in all cases.
> 	* gcc.c-torture/compile/20021117-1.c: New test
> 	* gcc.c-torture/compile/20021117-2.c: New test

I guess this is ok though.


r~

Re: [PATCH] Fix PR c/8518

[Geoff Keating]

Richard Henderson <rth@redhat.com> writes:

> On Sun, Nov 17, 2002 at 02:23:19AM +0100, Eric Botcazou wrote:
> > ... the compiler gets confused because of 
> > the redefinition of an extern inline function:
> > 
> > extern int inline bar () { return 0; }
> > static int inline bar () { return foo(); }
> 
> Incidentally, this seems like error material.  The redefinition of an
> extern inline function should be neither static nor inline itself.

I'd argue that the redefinition of any function should be an error.

-- 
- Geoffrey Keating <geoffk@geoffk.org>

Re: [PATCH] Fix PR c/8518

[Richard Henderson]

On Tue, Nov 19, 2002 at 03:40:45PM -0800, Geoff Keating wrote:
> I'd argue that the redefinition of any function should be an error.

extern inline is special, since it means "this function will
be defined out of line somewhere else".


r~

Re: [PATCH] Fix PR c/8518

[Geoff Keating]

> Date: Tue, 19 Nov 2002 15:51:36 -0800
> From: Richard Henderson <rth@redhat.com>
> Cc: gcc-patches@gcc.gnu.org
> Mail-Followup-To: Richard Henderson <rth@redhat.com>,
> 	Geoff Keating <geoffk@geoffk.org>, gcc-patches@gcc.gnu.org
> Content-Disposition: inline
> User-Agent: Mutt/1.4i
> X-OriginalArrivalTime: 19 Nov 2002 23:51:40.0507 (UTC) FILETIME=[9B192AB0:01C29026]
> 
> On Tue, Nov 19, 2002 at 03:40:45PM -0800, Geoff Keating wrote:
> > I'd argue that the redefinition of any function should be an error.
> 
> extern inline is special, since it means "this function will
> be defined out of line somewhere else".

The way that ISO C handles this is that you write

inline int foo(void) { return 1; }  // in header
extern int foo(void);  // in .c

and the compiler's supposed to produce an out-of-line copy of the
function.

-- 
- Geoffrey Keating <geoffk@geoffk.org>

Re: [PATCH] Fix PR c/8518

[Richard Henderson]

On Tue, Nov 19, 2002 at 04:13:31PM -0800, Geoff Keating wrote:
> The way that ISO C handles this is that you write
> 
> inline int foo(void) { return 1; }  // in header
> extern int foo(void);  // in .c
> 
> and the compiler's supposed to produce an out-of-line copy of the
> function.

That is indeed a better solution.  Doesn't help with the
existing GCC extension though.  :-(


r~

Re: [PATCH] Fix PR c/8518

[Eric Botcazou]

> > 	PR c/8518
> > 	* c-decl.c (duplicate_decls): Outline the second definition
> > 	of an extern inline function in all cases.
> > 	* gcc.c-torture/compile/20021117-1.c: New test
> > 	* gcc.c-torture/compile/20021117-2.c: New test
>
> I guess this is ok though.

Applied. OK for the branch too ?

-- 
Eric Botcazou

Re: [PATCH] Fix PR c/8518

[Richard Henderson]

On Wed, Nov 20, 2002 at 12:00:57PM +0100, Eric Botcazou wrote:
> Applied. OK for the branch too ?

Once it's open.


r~