Re: Make more use of idx_t

[Paul Eggert]

On 12/6/20 3:32 AM, Bruno Haible wrote:
> Paul Eggert wrote:
> > Those all look good to me.
>
> OK, I pushed them.

When I looked into this a bit more when hacking on canonicalize-lgpl, I 
realized I spoke too soon. First, there's some cruft left over from when 
idx_t could be unsigned. More important, defining IDX_WIDTH to be 
PTRDIFF_WIDTH - 1 is confusing as this makes idx_t different from all other 
signed types in that IDX_WIDTH does not count the sign bit. I can see why it 
doesn't count the sign bit (the sign bit is always zero) but I can also see 
why it should (shift counts are limited by the *_WIDTH macros). If we're 
going to have a macro that stands for PTRDIFF_WIDTH - 1 it'd probably be 
better to use a different naming convention, to avoid that confusion. 
However, nobody's using any such macro now and possibly we'll never need it 
so in the meantime I merely substituted a comment for the #define in the 
attached patch.

Hope this is OK; further comments welcome of course.

0001-idx-simplify-IDX_MAX-remove-IDX_WIDTH.patch
Description: Text Data