[klibc] Re: udevinfo output broken

[Greg KH]

On Tue, Jan 27, 2004 at 10:54:31PM +0100, Olaf Hering wrote:
>  On Tue, Jan 27, Olaf Hering wrote:
> 
> >  On Tue, Jan 27, Kay Sievers wrote:
> > 
> > > On Tue, Jan 27, 2004 at 09:06:51PM +0100, Olaf Hering wrote:
> > > > udevinfo writes to fd 0, all the output is lost somehow. 
> > > > 
> > > > olaf at ibook:~> i="`/sbin/udevinfo -r 2>&1`"
> > > > /dev/
> > > 
> > > Seems like a klibc feature :)
> > > libc works as expected.
> > 
> > _fwrite gets 0x1 as f, calls fileno() which decrements it and passes it
> > to write().
> 
> No idea if this is correct:
> 
> --- klibc-0.98/klibc/include/stdio.h    2004-01-22 21:56:49.000000000 +0100
> +++ klibc/klibc/include/stdio.h   2004-01-27 22:50:27.000000000 +0100
> @@ -15,9 +15,9 @@
>  struct _IO_file;
>  typedef struct _IO_file FILE;
>  
> -#define stdin  ((FILE *)0)
> -#define stdout ((FILE *)1)
> -#define stderr ((FILE *)2)
> +#define stdin  ((FILE *)1)
> +#define stdout ((FILE *)2)
> +#define stderr ((FILE *)3)
>  
>  #ifndef EOF
>  # define EOF (-1)
> 

I don't know about your change, but I traced the regression down to this
recent recent change in klibc:


diff -Nru a/klibc/klibc/include/stdio.h b/klibc/klibc/include/stdio.h
--- a/klibc/klibc/include/stdio.h	Tue Jan 27 14:00:35 2004
+++ b/klibc/klibc/include/stdio.h	Tue Jan 27 14:00:35 2004
@@ -31,15 +31,22 @@
 #define SEEK_CUR 1
 #define SEEK_END 2
 
+/*
+ * Convert between a FILE * and a file descriptor.  We don't actually
+ * have any in-memory data, so we just abuse the pointer itself to
+ * hold the data.  Note, however, that for file descriptors, -1 is
+ * error and 0 is a valid value; for FILE *, NULL (0) is error and
+ * non-NULL are valid.
+ */
 static __inline__ int fileno(FILE *__f)
 {
   /* This should really be intptr_t, but size_t should be the same size */
-  return (int)(size_t)__f;
+  return (int)(size_t)__f - 1;
 }
 
 static __inline__ FILE * __create_file(int __fd)
 {
-  return (FILE *)(size_t)__fd;
+  return (FILE *)(size_t)(__fd + 1);
 }
 
 __extern FILE *fopen(const char *, const char *);


Peter, any reason for doing this?  It seems to change the way stdout
works.

thanks,

greg k-h