This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Define wordsize.h macros everywhere


On Wed, 26 Oct 2016, Steve Ellcey wrote:

> diff --git a/bits/wordsize.h b/bits/wordsize.h
> index 9ef0e85..2ebecd6 100644
> --- a/bits/wordsize.h
> +++ b/bits/wordsize.h
> @@ -1 +1,28 @@
>  #error "This file must be written based on the data type sizes of the target"
> +
> +/* The following entries are a template for what defines should be in the
> +   wordsize.h header file for a target.  */
> +
> +/* Bits per word (size of CPU register).  */
> +#define __WORDSIZE

That's not an accurate description.  It's definitely the size of long int 
and pointers, even if registers are larger.

> +/* Set to the size of the f_fsid field in the statvfs and statvfs64 structs.
> +   In most cases this is the same as __WORDSIZE but some 32 bit hosts may set
> +   it to 64.  */
> +#define __SYSCALL_WORDSIZE

If it's so specific to those syscalls, it doesn't belong in 
bits/wordsize.h but in some header specific to those syscalls.  Even the 
*concept* of syscalls doesn't belong in bits/wordsize.h (remember there 
are non-syscall ports such as Hurd).

> +/* Set to 1 if ssize_t is type 'unsigned long' instead of type 'unsigned int'
> +   so that SIZE_MAX is defined as an unsigned long constant instead of an
> +   unsigned int constant.  Set to 0 if ssize_t is 'unsigned int'.  */
> +#define __WORDSIZE32_SIZE_ULONG

Is this about ssize_t as the comment says (twice), or size_t?

> +/* Set to 1 if ptrdiff_t is type 'long' instead of type 'int' so that
> +   PTRDIFF_MIN and PTRDIFF_MAX are defined as long constants instead of
> +   int constants.  Set to 0 if ptrdiff_t is type 'int'.  */
> +#define __WORDSIZE32_PTRDIFF_LONG

For both these macros I think consensus is as Andreas said in 
<https://sourceware.org/ml/libc-alpha/2016-07/msg00251.html> and 
<https://sourceware.org/ml/libc-alpha/2016-07/msg00267.html>.

That is, they should only be defined if __WORDSIZE == 32.  They should be 
tested with #if not #ifdef, but all such tests should only be evaluated if 
__WORDSIZE == 32, so that we get -Wundef errors if any such test gets 
evaluated with __WORDSIZE == 64.

So the comments specifying the semantics of those macros should say that 
they are only defined or used if __WORDSIZE == 32, and the headers should 
only define them under that condition.

-- 
Joseph S. Myers
joseph@codesourcery.com

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]