Re: [MiNT] Gcc 3.3.6 v 4.0.1

[Alan Hourihane <alanh@fairlite.co.uk>]

On Thu, 2009-02-26 at 09:34 +0100, Vincent Rivière wrote:
> Vincent Rivière wrote:
> > The test calls gmp_obstack_vprintf(), then it turns to call the MiNTLib 
> > in obstack_vprintf() then vprintf().
> > 
> > vfprintf() returns -1, it means failure.
> > Look at the MiNTLib sources at mintlib/stdio/vfprintf.c
> > http://sparemint.org/cgi-bin/cvsweb/mintlib/stdio/vfprintf.c?rev=1.5
> > 
> > That source is quite horrible and not debugger friendly :-(
> > It seems that it fails very soon due to the macro ARGCHECK() (maybe 
> > because a bad FILE status ??)
> 
> I reproduced the bug with a small standalone testcase (not using gmp).
> It passes on Linux, but fails on MiNT.
> So the bug is in the MiNTLib.
> 
> The testcase is attached to this message.
> Compile with :
> gcc ob.c -o ob -g -Wall -D_GNU_SOURCE
> 
> $ ./ob
> FAIL, errno = 37: Bad file descriptor
> 
> Again, the macro ARGCHECK() is hit.
> Unfortunately macros cannot be debugged :-(
> 
> So I guess these versions are used :
> In mintlib/stdio/vfprintf.c:
> # define ARGCHECK(S, Format)				\
>    do							\
>      {							\
>        /* Check file argument for consistence.  */	\
>        if (!__validfp (S) || !S->__mode.__write)		\
> 	{						\
> /*here?*/ __set_errno (EBADF);				\
> 	  return -1;					\
> 	}						\
>        if (Format == NULL)				\
> 	{						\
> 	  __set_errno (EINVAL);				\
> 	  return -1;					\
> 	}						\
>        if (!S->__seen)					\
> 	{						\
> 	  if (__flshfp (S, EOF) == EOF)			\
> 	    return -1;					\
> 	}						\
>      }							\
>     while (0)
> 
> In mintlib/include/stdio.h:
> /* Nonzero if STREAM is a valid stream.
>     STREAM must be a modifiable lvalue (wow, I got to use that term).
>     See stdio/glue.c for what the confusing bit is about.  */
> #define	__validfp(stream)					      \
>    (stream != NULL &&						      \
>     ({ if (stream->__magic == _GLUEMAGIC)			      \
> 	stream = *((struct { int __magic; FILE **__p; } *) stream)->__p; \
>        stream->__magic == _IOMAGIC; }))
> 
> It seems that some black magic is involved here :-(
> No time to dig further...
> The help of an stdio expert would be welcome.

I think this should fix it.....

Alan.

Index: stdio/obstream.c
===================================================================
RCS file: /mint/mintlib/stdio/obstream.c,v
retrieving revision 1.1.1.1
diff -u -r1.1.1.1 obstream.c
--- stdio/obstream.c	12 Oct 2000 10:56:54 -0000	1.1.1.1
+++ stdio/obstream.c	26 Feb 2009 09:22:17 -0000
@@ -122,6 +122,9 @@
 static void
 init_obstream (FILE *stream, struct obstack *obstack)
 {
+  (void) obstack;
+
+  stream->__magic = _IOMAGIC;
   stream->__mode.__write = 1;
   stream->__mode.__read = 1;