Re: exact integral types

On Sat, 19 Oct 2002, Warren Turkal wrote:

> Here is a patch that should make Xmd.h use C99 when available.

> --- /usr/include/X11/Xmd.h	2002-10-16 15:14:38.000000000 -0500
> +++ Xmd.h	2002-10-19 14:23:33.000000000 -0500
> @@ -47,6 +47,7 @@
>  ******************************************************************/
>  #ifndef XMD_H
>  #define XMD_H 1
> +
>  /* $Xorg: Xmd.h,v 1.4 2001/02/09 02:03:22 xorgcvs Exp $ */
>  /*
>   *  Xmd.h: MACHINE DEPENDENT DECLARATIONS.
> @@ -101,6 +102,36 @@
>  #define SIZEOF(x) sz_/**/x
>  #endif /* if ANSI C compiler else not */
>
> +#if defined(__STDC_VERSION__) && (__STDC_VERSION__ == 199901L)
                                                      ^^
This should be                                        >=

> +
> +#include <stdint.h>
> +
> +typedef uint8_t  CARD8;
> +typedef uint16_t CARD16;
> +typedef uint32_t CARD32;
> +typedef uint64_t CARD64;
> +
> +typedef int8_t  INT8;
> +typedef int16_t INT16;
> +typedef int32_t INT32;
> +typedef int64_t INT64;
> +
> +typedef CARD8 BYTE;
> +typedef CARD8 BOOL;
> +
> +typedef CARD32 BITS32;
> +typedef CARD16 BITS16;
> +
> +/*is there any reason why B32 and B16 cannot always be defined?*/
> +# if defined(WORD64)
> +#define B32 :32
> +#define B16 :16
> +# else
> +#define B32
> +#define B16
> +# endif
> +
> +#else //__STDC_VERSION__
>  /*
>   * Bitfield suffixes for the protocol structure elements, if you
>   * need them.  Note that bitfields are not guarranteed to be signed
> @@ -165,6 +196,8 @@
>  #define BOOL		CARD8
>  #endif /* __EMX__ */
>
> +#endif // __STDC_VERSION__
> +
>  /*
>   * definitions for sign-extending bitfields on 64-bit architectures
>   */

I have spent the time to review this change and found that it does not
work (at least not as intended).

First, a minor problem is that even GCC 3.2 #define's __STDC_VERSION__
only if compiling with -std=c99 (or -std=gnu99), but not with -ansi (the
default we use).  I am given to understand that this will change in some
future GCC.

A larger issue is that, on some 32-bit systems, this ends up typedef'ing
CARD32 as 'unsigned int', and on others, as 'unsigned long'.   For reasons
I won't get into here, the X sources currently need CARD32 to be
typedef'ed as 'unsigned long' for all 32-bit environments.  The same
issue, without 'unsigned' occurs with INT32.

Bottom line:  The X sources are not yet ready for C99, and a lot more work
than this relatively simple change is needed to make them so.

Marc.

+----------------------------------+-----------------------------------+
|  Marc Aurele La France           |  work:   1-780-492-9310           |
|  Computing and Network Services  |  fax:    1-780-492-1729           |
|  352 General Services Building   |  email:  [email protected]          |
|  University of Alberta           +-----------------------------------+
|  Edmonton, Alberta               |                                   |
|  T6G 2H1                         |     Standard disclaimers apply    |
|  CANADA                          |                                   |
+----------------------------------+-----------------------------------+
XFree86 Core Team member.  ATI driver and X server internals.