HOST_WIDE_INT, preprocessing, and MIPS

[Mark Mitchell]

Folks --

  On 64-bit mips the preprocessor turns out to be quite severely
broken.  In particular, the following program:

  #if (1L << 32)
  x
  #endif

yields:

  > g++ -mabi=64 -E test.c
  # 1 "test.c"

after preprocessing.  But, since a long is 64-bits in this
environment, this should not occur.  

  The problem is that cccp.c uses HOST_WIDE_INT to do computations,
and mips.h defines this to long, and cccp.c is compiled using the
32-bit object model, in which long is just 32-bits. 

  One possibility is to make HOST_WIDE_INT wider than 32-bits.  But, I
bet this will cause problems in lots of places.

  In cccp.c, we have:

    #ifndef HOST_WIDE_INT
    # if HAVE_INTTYPES_H
    #  include <inttypes.h>
    #  define HOST_WIDE_INT intmax_t
    # else
    #  if (HOST_BITS_PER_LONG <= HOST_BITS_PER_INT && HOST_BITS_PER_LONGLONG <= HOST_BITS_PER_INT)
    #   define HOST_WIDE_INT int
    #  else
    #  if (HOST_BITS_PER_LONGLONG <= HOST_BITS_PER_LONG || ! (defined LONG_LONG_MAX || defined LLONG_MAX))
    #   define HOST_WIDE_INT long
    #  else
    #   define HOST_WIDE_INT long long
    #  endif
    #  endif
    # endif
    #endif
 
  The problem is that HOST_WIDE_INT is being defined in mips.h,
included via config.h, which causes us not to use intmax_t here.
Apparently this is done to provide return types for some functions
declared in mips.h.

  I suggest that we rename the definitions in mips.h to
MIPS_HOST_WIDE_INT, or some such, so that this breakage does not
occur.  That would allow the HAVE_INTTYPES_H test to fire.
Alternatively, we could reorder the cccp.c tests to give
HAVE_INTTYPES_H priority.

  Also, there was discussion recently about removing the use of
inttypes.h from the definition of HOST_WIDE_INT.  I think this
examples shows that it *must* be present.  Since the preprocessor
doesn't know about the object model, it must use the widest type
available on the platform, period.

  Jim, I would like to make the change I suggest to mips.h.  (Namely,
to rename HOST_WIDE_INT to MIPS_HOST_WIDE_INT.)  OK?  

-- 
Mark Mitchell 			mark@markmitchell.com
Mark Mitchell Consulting	http://www.markmitchell.com