RE: [Vxl-maintainers] changes to explicitly sized integer types in vxl_config.h

[Ian S. <ian...@st...>]

OK,

There appears to be some disquiet about getting rid of vxl_.?int_8 and
vxl_sint_.+, so  I will not get rid of anything, and later today just add
vxl_byte, and vxl_sbyte to vxl_config.

If someone else wants to argue for getting rid of vxl_.?int_8 or
vxl_sint_.+, then we (Manchester University) would be happy for them to do
so.

Ian.

> -----Original Message-----
> From: Frederik Schaffalitzky [mailto:fs...@sy...]
> Sent: Friday, October 04, 2002 2:16 AM
> To: Ian Scott
> Cc: Vxl-maintainers (E-mail)
> Subject: RE: [Vxl-maintainers] changes to explicitly sized
> integer types
> in vxl_config.h
>
>
> On Thu, 3 Oct 2002, Ian Scott wrote:
>
> > add vxl_byte:
> > 	typedeffed to unsigned char 8-bit char platforms.
> > 	vil2 will use this instead of vil2_byte.
> >
> > add vxl_sbyte:
> > 	typedeffed to signed char 8-bit char platforms.
>
> Sure, why not. That would be handy.
>
>
> > deprecate vxl_int_8, vxl_sint_8, vxl_uint_8:
> > 	(As Peter pointed out) these are normally defined
> > 	to be char so
> > 		vcl_cout << (vxl_int_8)(65) << vcl_endl;
> > 	displays 'A' not 65. This is not int behaviour.
> > 	I also prefer vxl_byte over vxl_uint_8, and for
> > 	consistency vxl_sbyte over vxl_int_8.
>
> The vxl_*_* typedefs denote machine words of various sizes.
> To call 8-bit
> words "bytes" may be helpful at times but it is an inconsistent naming
> scheme.  One way in which a consistent naming is useful is that the
> typedefs can then be used in macros. For a contrived example:
>   #define macro(n) \
>   do { \
>     std::cout << "trying " << n << "-bit integers" << std::endl; \
>     typedef vxl_int_##n> int_type; \
>     ... etc ...
>   } while (false)
>
> Then one can do
>   macro(8);
>   macro(16);
>   macro(32);
> etc. This would not work if the 8-bit versions are renamed to "byte".
>
> Macros are also the reason why having "uint" and "sint" is
> useful. E.g.
>   #define macro(z, n) \
>   do { \
>     typedef vxl_##z##int_##8 type; \
>     std::vector<type> values; \
>     ... etc ...
>   } while(false)
>
> which can used as in
>   // implement signed case
>   macro(s, 8);
>   macro(s, 16);
>   macro(s, 32);
>   // implement unsigned case
>   macro(u, 8);
>   macro(u, 16);
>   macro(u, 32);
>
> My point is that having consistently and explicitly named
> typedefs around
> can be useful. That does not preclude having, also useful, typedefs
>   typedef vxl_uint_8 vxl_byte; // assume unsigned
> and
>   typedef vxl_int_8  vxl_sint_8;
>   typedef vxl_int_16 vxl_sint_16;
>   typedef vxl_int_32 vxl_sint_32;
> for those who prefer to not see the "s"s but to eliminate the
> other ones
> altogether seems a bit pedantic, and not useful in any way.
>
> Why not just add some more typedefs but leave the existing
> ones as they
> are? Why must the existing ones be eliminated?
>