Unrecog attribute func handling change request + patch

Unrecog attribute func handling change request + patch

[Mumit Khan]

While using a different stdio implementation that defines printf, etc
as macros, I ran into a problem where GCC flags unrecognized attribute
functions as errors. This seems rather too harsh, and I suggest warning
the user instead; after all, this is an extension.

To see the problem, use gas/as.h for example:

  #if (__GNUC__ >= 2) && !defined(VMS)
  /* for use with -Wformat */
  #define PRINTF_LIKE(FCN)  void FCN (const char *format, ...) \
                             __attribute__ ((format (printf, 1, 2)))
  #else 
  /* ... */
  #endif

SFIO (Safe/Fast I/O Library by David G. Korn and Kiem-Phong Vo) defines 
printf to be _stdprintf, so any code using this library and writing code 
like above, and there's lots of it, will fail.

Here's a patch:

Sun Feb  1 17:05:26 1998  Mumit Khan  <khan@xraylith.wisc.edu>
	
	* c-common.c (decl_attributes): Flag unrecognized attribute
	functions as warnings instead of as errors.

Index: c-common.c
===================================================================
RCS file: /usr/local/src/CVSROOT/egcs/gcc/c-common.c,v
retrieving revision 1.1.1.1
diff -u -p -r1.1.1.1 c-common.c
--- c-common.c	1997/12/20 18:10:20	1.1.1.1
+++ c-common.c	1998/02/02 23:00:21
@@ -645,8 +645,8 @@ decl_attributes (node, attributes, prefi
 	      is_scan = 1;
 	    else if (TREE_CODE (format_type) == IDENTIFIER_NODE)
 	      {
-		error ("`%s' is an unrecognized format function type",
-		       IDENTIFIER_POINTER (format_type));
+		warning ("`%s' is an unrecognized format function type",
+		         IDENTIFIER_POINTER (format_type));
 		continue;
 	      }
 	    else

Regards,
Mumit -- khan@xraylith.wisc.edu
http://www.xraylith.wisc.edu/~khan/

Re: Unrecog attribute func handling change request + patch

[Andreas Schwab]

Mumit Khan <khan@xraylith.wisc.edu> writes:

|> While using a different stdio implementation that defines printf, etc
|> as macros, I ran into a problem where GCC flags unrecognized attribute
|> functions as errors. This seems rather too harsh, and I suggest warning
|> the user instead; after all, this is an extension.

|> To see the problem, use gas/as.h for example:

|>   #if (__GNUC__ >= 2) && !defined(VMS)
|>   /* for use with -Wformat */
|>   #define PRINTF_LIKE(FCN)  void FCN (const char *format, ...) \
|>                              __attribute__ ((format (printf, 1, 2)))
|>   #else 
|>   /* ... */
|>   #endif

|> SFIO (Safe/Fast I/O Library by David G. Korn and Kiem-Phong Vo) defines 
|> printf to be _stdprintf, so any code using this library and writing code 
|> like above, and there's lots of it, will fail.

They should be using __attribute__ ((__format__ (__printf__, 1, 2))).

-- 
Andreas Schwab                                      "And now for something
schwab@issan.informatik.uni-dortmund.de              completely different"
schwab@gnu.org

Re: Unrecog attribute func handling change request + patch

[Mumit Khan]

Andreas Schwab <schwab@issan.informatik.uni-dortmund.de> writes:
> 
> They should be using __attribute__ ((__format__ (__printf__, 1, 2))).
> 

Possibly yes, but most existing code take it straight out of the GCC
documentation, where it *doesn't* use __printf__ nor __format__.

In any event, it's simply wrong to flag it as an error.

Mumit

Re: Unrecog attribute func handling change request + patch

[Jeffrey A Law]

  In message < 9802022306.AA14407@modi.xraylith.wisc.edu >you write:
  > While using a different stdio implementation that defines printf, etc
  > as macros, I ran into a problem where GCC flags unrecognized attribute
  > functions as errors. This seems rather too harsh, and I suggest warning
  > the user instead; after all, this is an extension.
  > 
  > To see the problem, use gas/as.h for example:
  > 
  >   #if (__GNUC__ >= 2) && !defined(VMS)
  >   /* for use with -Wformat */
  >   #define PRINTF_LIKE(FCN)  void FCN (const char *format, ...) \
  >                              __attribute__ ((format (printf, 1, 2)))
  >   #else 
  >   /* ... */
  >   #endif
  > 
  > SFIO (Safe/Fast I/O Library by David G. Korn and Kiem-Phong Vo) defines 
  > printf to be _stdprintf, so any code using this library and writing code 
  > like above, and there's lots of it, will fail.
  > 
  > Here's a patch:
  > 
  > Sun Feb  1 17:05:26 1998  Mumit Khan  <khan@xraylith.wisc.edu>
  > 	
  > 	* c-common.c (decl_attributes): Flag unrecognized attribute
  > 	functions as warnings instead of as errors.
I tend to agree that this shouldn't be an error...  I've seen somebody
that disagreed...    I've installed the patch, if we later think it should
be an error, we can always switch it back.

jeff