autoconf for type sizes

[Joseph S. Myers]

On Mon, 12 Mar 2001, Kaveh R. Ghazi wrote:

> 1.  (Optional) is the gcc_AC_C_CHAR_BIT test necessary?  Why not just
>     default to 8 if it doesn't appear in limits.h?  I notice that
>     gcc/glimits.h unconditionally sets it to 8 anyway, so its unlikely
>     that we'll encounter a *host* which sets it to a different value.

This unconditional define to 8 is of course a bug (e.g. c4x).  The
existing limits.h is fragile for odd systems and those where -m options
provide for different type sizes.  I've suggested (e.g. see c/448) that
limits.h should be generated for each multilib, in a similar fashion to
these type size checks, presuming that all choices for -m options
affecting these types that are supported for a given compiler installation
have corresponding multilibs, but someone who understands the multilib
machinery would need to set things up so that a program can be run for
each multilib to generate headers and so that multilib headers get
installed.

-- 
Joseph S. Myers
jsm28@cam.ac.uk