Re: ac_cv_sizeof_X, et al.

[Philip Prindeville]

Eric Blake wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> According to Philip Prindeville on 7/14/2008 5:27 PM:
>
> | I'm looking at sox-12.17.9 and seeing in their "configure" file (which
> |  claims to be generated via Autoconf 2.59)
>
> Newer autoconf versions are able to accurately determine type sizes when
> cross-compiling, making life much easier for porting to a
> cross-compilation platform.  You may want to consider putting pressure on
> the sox maintainers to use a newer autoconf version for their next 
> release.
>
> |
> | 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.  :-)
>
> Unfortunately, the C standard specifies several alternative spellings for
> the same intrinsic type, so you have to worry about all of the 
> possibilities.

How hard would it be to parse out the type, and generate a warning for 
anything other than the "canonical" name for a given intrinsic type?

unsigned long int => unsigned long

(actually, do you even need to differentiate between "unsigned" and signed?

> |
> | Likewise with the "ac_cv_sizeof_charp" versus "ac_cv_sizeof_char_p",
>
> That may (although I haven't checked) stem from a difference in 'char*'
> vs. 'char *', where the space is being converted to _.

That should be easy enough to standardize, right?

-Philip


> | and
> | "ac_cv_sizeof_char" versus "ac_cv_sizeof_unsigned_char" (are these
> | *ever* different???).
>
> C requires sizeof(char), sizeof(signed char), and sizeof(unsigned 
> char) to
> all be equal to 1.  Any compiler where this is not true is broken.  Thus,
> any package that has a configure test for this is wasting time.  A recent
> thread was posted proposing adding a warning if the user does
> AC_CHECK_SIZEOF([char]), but no one has implemented something that works
> on all the variant spellings.