ac_cv_sizeof_X, et al.

[Philip Prindeville]

[ As per Ralf's suggestion, starting a new thread on a tangential issue... ]

I'm looking at sox-12.17.9 and seeing in their "configure" file (which 
claims to be generated via Autoconf 2.59) from their configure.in file 
and seeing:

ac_cv_sizeof_x="$ac_cv_sizeof_char:$ac_cv_sizeof_short"
ac_cv_sizeof_X="$ac_cv_sizeof_x:$ac_cv_sizeof_int"
ac_cv_sizeof_X="$ac_cv_sizeof_X:$ac_cv_sizeof_voidp:$ac_cv_sizeof_long"
if test "$ac_cv_header_stdint" = "stddef.h" ; then
#   we must guess all the basic types. Apart from byte-adressable system,
# there a few 32-bit-only dsp-systems. nibble-addressable systems are way off.
   cat >>$ac_stdint_h <<EOF
/* ------------ BITSPECIFIC INTTYPES SECTION --------------- */
EOF
   t="typedefs for a"
   case "$ac_cv_sizeof_X" in
    1:2:2:2:4) echo "$as_me:$LINENO: result: ..adding $t normal 16-bit system" 
>&5
echo "${ECHO_T}..adding $t normal 16-bit system" >&6
               cat >>$ac_stdint_h <<EOF
/*              a normal 16-bit system                       */
typedef unsigned char   uint8_t;
typedef unsigned short  uint16_t;
typedef unsigned long   uint32_t;
#ifndef __int8_t_defined
#define __int8_t_defined
typedef          char    int8_t;
typedef          short   int16_t;
typedef          long    int32_t;
#endif
EOF
;;
...


and wondering what is going on with the ac_cv_sizeof_X variable?

On my cross-compile platform (a Geode LX700), I'm getting:

1:2:4::4


as the value. Apparently we didn't define the size of a void *... (not 
sure why there's both charp/char_p and voidp/void_p combinations out 
there... would be nice to have uniformity.)

Seems that, had we forced ac_cv_sizeof_voidp=4 in the autoconf 
invocation, this would have forced:

    1:2:4:4:4) echo "$as_me:$LINENO: result: ..adding $t normal 32-bit system" 
>&5
echo "${ECHO_T}..adding $t normal 32-bit system" >&6
               cat >>$ac_stdint_h <<EOF
/*              a normal 32-bit system                       */
typedef unsigned char   uint8_t;
typedef unsigned short  uint16_t;
typedef unsigned int    uint32_t;
#ifndef __int8_t_defined
#define __int8_t_defined
typedef          char    int8_t;
typedef          short   int16_t;
typedef          int     int32_t;
#endif
EOF
;;


instead.

I did a quick search, and it seems that other packages also use this 
ac_cv_sizeof_X convention.

Is this encouraged?  Discouraged?

It seems to me that more useful would be the converse, i.e. 
--with-nominal-32bit-sizes or whatever forcing all the standard sizes 
(instead of having to specify each).

I'm also not sure why there's "ac_cv_sizeof_long" and 
"ac_cv_sizeof_long_int", etc. except perhaps to make the cross-compiling 
distro-manager's life hellish.  :-)

Likewise with the "ac_cv_sizeof_charp" versus "ac_cv_sizeof_char_p", and 
"ac_cv_sizeof_char" versus "ac_cv_sizeof_unsigned_char" (are these 
*ever* different???).

Is any of this covered somewhere where a distro manager can get up to 
speed on the usual pitfalls when adding a new package to a 
cross-compiled build?

And is there a "autoconf-lint" to check for using broken names, like 
"ac_cv_sizeof_long_long_unsigned_int" instead of whatever is preferred?

Thanks,

-Philip