egcs-1.0 on mingw32/possibly cygwin32 regression

egcs-1.0 on mingw32/possibly cygwin32 regression

[Andrew Zabolotny]

Hello everybody!

The mingw32 egcs-1.0 port and possibly cygnus-win32 egcs-1.0 (since first
is
heavily based on second) has the following problem:

The program:

---------------
#include <stdio.h>

static __attribute__ ((stdcall)) BlackHole(int a);

void doit()
{
 BlackHole(1);
}

static __attribute__ ((stdcall)) BlackHole(int a)
{
 printf("disappearing function! %d\n", a);
}

int main()
{
 doit();
 return (0);
}
---------------

does not compile correctly with -O3. However, it compiles okay if the
-fkeep-inline-functions or with -fno-inline-functions switch was
specified.
The problem is that the BlackHole function is for some reason marked
as 'inline' but instead of inlining its code the compiler puts a
'call BlackHole@4' anyway, but the actual function code is totally
missing.

This problem is not present with gcc-2.7.2 from cygwin32.

I've did not tracked the problem, but I think it's related to the
ENCODE_SECTION_INFO macro in cygwin32.h (the name of function changes
and compiler detects by name whenever the function is inline??? nonsense)

Greetings,
    _\ndy@teamOS/2

[patch] Re: egcs-1.0 on mingw32/possibly cygwin32 regression

[Mumit Khan]

Hi Andrew,

Here's a patch to undo the effects of ENCODE_SECTION_INFO on win32. 
Currently the i386/cygwin32.h defines ENCODE_SECTION_INFO which adds 
@[NUM] to names that have the "stdcall" attribute, but forgets 
to define STRIP_NAME_ENCODING. (Thanks for debugging the problem!)

Here's your test case (filed Jan 8, 1998):

  #include <stdio.h>

  static __attribute__ ((stdcall)) BlackHole(int a);

  void doit()
  {
   BlackHole(1);
  }

  static __attribute__ ((stdcall)) BlackHole(int a)
  {
   printf("disappearing function! %d\n", a);
  }

  int main()
  {
   doit();
   return (0);
  }

Currently with -O3, BlackHole gets inlined, but there is an (incorrect)
undefined call to BlackHole.
  
  % i386-cygwin32-gcc -O3 -c blackhole.c
  % i386-cygwin32-nm blackhole.o
  00000000 b .bss
  00000000 d .data
  00000000 t .text
	   U _BlackHole@4			<<<<<< Here.
  00000000 t ___gnu_compiled_c
	   U ___main
  0000001c T _doit
  00000028 T _main
  00000000 t gcc2_compiled.

With this patch:

  % i386-cygwin32-nm blackhole.o
  00000000 b .bss
  00000000 d .data
  00000000 t .text
  0000003c t _BlackHole@4			<<<<<< Voila!
  00000000 t ___gnu_compiled_c
	   U ___main
  0000001c T _doit
  00000028 T _main
	   U _printf
  00000000 t gcc2_compiled.

Fri Jan  9 20:36:23 1998  Mumit Khan <khan@xraylith.wisc.edu>

	* i386/cygwin32.h (STRIP_NAME_ENCODING): Define for Win32 to strip
	off the trailing @[NUM] added by ENCODE_SECTION_INFO.

Index: egcs/gcc/config/i386/cygwin32.h
diff -c egcs/gcc/config/i386/cygwin32.h:1.2 egcs/gcc/config/i386/cygwin32.h:1.3
*** egcs/gcc/config/i386/cygwin32.h:1.2	Sat Dec 20 16:34:20 1997
--- egcs/gcc/config/i386/cygwin32.h	Fri Jan  9 20:46:09 1998
***************
*** 150,155 ****
--- 150,178 ----
  while (0)
  #endif
  
+ /* This macro gets just the user-specified name
+    out of the string in a SYMBOL_REF.  Discard
+    trailing @[NUM] encoded by ENCODE_SECTION_INFO. 
+    Do we need the stripping of leading '*'?  */
+ #undef  STRIP_NAME_ENCODING
+ #define STRIP_NAME_ENCODING(VAR,SYMBOL_NAME)				\
+ do {									\
+   char *_p;								\
+   char *_name = ((SYMBOL_NAME) + ((SYMBOL_NAME)[0] == '*'));		\
+   for (_p = _name; *_p && *_p != '@'; ++_p)				\
+     ;									\
+   if (*_p == '@')							\
+     {									\
+       int _len = _p - _name;						\
+       (VAR) = (char *) alloca (_len + 1);				\
+       strncpy ((VAR), _name, _len);					\
+       (VAR)[_len] = '\0';						\
+     }									\
+   else									\
+     (VAR) = _name;							\
+ } while (0)
+       
+ 
  /* Emit code to check the stack when allocating more that 4000
     bytes in one go. */
  
Regards,
Mumit -- khan@xraylith.wisc.edu
http://www.xraylith.wisc.edu/~khan/

Re: [patch] Re: egcs-1.0 on mingw32/possibly cygwin32 regression

[Jeffrey A Law]

  In message < 9801100305.AA09263@modi.xraylith.wisc.edu >you write:
  > Here's a patch to undo the effects of ENCODE_SECTION_INFO on win32. 
  > Currently the i386/cygwin32.h defines ENCODE_SECTION_INFO which adds 
  > @[NUM] to names that have the "stdcall" attribute, but forgets 
  > to define STRIP_NAME_ENCODING. (Thanks for debugging the problem!)
[ ... ]

  > Fri Jan  9 20:36:23 1998  Mumit Khan <khan@xraylith.wisc.edu>
  > 
  > 	* i386/cygwin32.h (STRIP_NAME_ENCODING): Define for Win32 to strip
  > 	off the trailing @[NUM] added by ENCODE_SECTION_INFO.
Thanks.  I've installed this change.

jeff