2.95.2 preprocessing phase calculations (C, not C++)

[Dave Brolley]

The preprocessor must use arithmetic that has "at least the
ranges" specified by in the section referenced by
lib.support.limits. i.e. these ranges may be exceeded.

Dave

Joseph Rivilis wrote:
> 
> I use djgpp 2.95.2 1999102,  with  .c input file.
> 
> Command line options were '-ansi -pedantic', but I think
> following is relevant not only for ANSI subcase.
> 
> It is natural for one to expect (with '-ansi -pedantic' it is a
> must) that preprocessor calculations needed by (acceptable)
> constant expression are the same as they were for this constant
> expression after preprocessing phase (if it is acceptable there
> too).
> 
> For ANSI C it is must because standard simply refers in both
> theese cases to run-time expression evaluation rules, so
> whatever the rules are, they are same.
> 
> GCC for x86 seems to perform preprocessor phase calculations
> as 64 -bit  for constants of  (unsigned) int and long types even
> though int and long are 32 bit wide for the program being
> compiled.
> 
> Below are two constant expressions which demonstrate difference
> between preprocessing phase evaluation and the normal one:
> 
> #if (    ((3U<<31) >> 31) ==  3  )    is true for GCC
> if   (    ((3U<<31) >> 31) ==  1  )    is true for GCC
> #if (    ((3  <<31) >> 31) ==   3  )    is true for GCC
> if   (    ((3  <<31) >> 31) == -1  )    is true for GCC
> 
> Sincerely yours,
> Joseph Rivilis