Re: testing of CVS Autoconf

[Ralf Wildenhues]

Hi Paul,

* Paul Eggert wrote on Wed, Mar 15, 2006 at 08:05:27PM CET:
> 
> My gnulib additions ran into a problem when I actually tried to use
> them on GNU tar.  The problem is that, even with the changes, GNU tar
> still assumes that 1 << (sizeof (int) * CHAR_BIT - 1) is valid.  But
> this isn't correct in general: you need to know how many value bits
> (i.e., non-padding bits) are in the int.  (A counterexample is the
> Unisys 2200 series, which as I understand it is still produced and
> maintained, and which supports POSIX.) 

Is there a simulator or a cross target of GCC available that would make
it possible to test this system, or another one where int has padding
bits or is otherwise funny?  Alternatively, does anyone have access to
such a system and could do those tests?

> The gnulib additions calculate the number of value bits but don't make
> them available to 'tar'.  So I would like to add that.

I assume you are talking about AC_C_TYPE_RANGE_INTEGER here, right?
What would tar need?  The value of $ac_shiftbits and $ac_signbits
as `#define's?

Cheers,
Ralf