"discards qualifiers" with no apparent cause

["L. Peter Deutsch" <ghost at aladdin dot com>]

Environment: gcc 2.95.1 running on:

> Linux oquirrh.math.utah.edu 2.2.5-22 #1 Wed Jun 2 09:17:03 EDT 1999 i686 unknown

Symptom:

Many "discards qualifiers" messages with no apparent cause.  gcc 2.7.2.3
does not produce these messages, so they're caused by something other than
discarding const.  The test case is large and complex: Aladdin Ghostscript
5.91 from ftp://ftp.cs.wisc.edu/ghost/aladdin/test/ghostscript-5.91.tar.gz.
One place it occurs is (twice) on line 55 of gsfunc3.c, which, after
preprocessor expansion, reads:

static   const gc_ptr_element_t      function_ElIn_enum_ptrs   [] =  { { GC_ELT_OBJ, ((int) ( (char *)&((       gs_function_ElIn_t      *)0)->          params.C0      - (char *)((       gs_function_ElIn_t      *)0) ))  } , { GC_ELT_OBJ, ((int) ( (char *)&((       gs_function_ElIn_t      *)0)->          params.C1      - (char *)((       gs_function_ElIn_t      *)0) ))  }   }; static   const gc_struct_data_t           function_ElIn_reloc_ptrs      = {   (sizeof(         function_ElIn_enum_ptrs     ) / sizeof((         function_ElIn_enum_ptrs     )[0]))  ,       0   ,       &    st_function     ,           function_ElIn_enum_ptrs      };     static   const gs_memory_struct_type_t               st_function_ElIn      = { sizeof(          gs_function_ElIn_t     ),           "gs_function_ElIn_t"     , 0, 0, basic_enum_ptrs, basic_reloc_ptrs,     0  , &          function_ElIn_reloc_ptrs      }      ;

Note that the only casts in this mess are applied to pointers derived from
0.  gs_function_ElIn_t is a struct type: its params.C0 and params.C1 members
are declared as const float *, but casting a const float ** (the type of
&xxx->params.C0) to a char * does not discard any qualifiers (unlike casting
a const float *const * to a char *) any more than casting a float ** to a
char * does.  Is that the problem?  (I've run into other compilers that also
don't handle the distinction between const ** and const *const * properly.)

I can change the macro that generates the casts to use const char * rather
than char *, but this still looks like a bug to me, and it may crop up in
other places.

				Thanks -

L. Peter Deutsch         |       Aladdin Enterprises :::: ghost@aladdin.com
203 Santa Margarita Ave. | tel. +1-650-322-0103 (AM only); fax +1-650-322-1734
Menlo Park, CA 94025     |        http://www.cs.wisc.edu/~ghost/index.html