Re: Feeding back fixincludes changes

[Bruce Korb <bkorb at cruzio dot com>]

Richard Henderson wrote:
> 
> On Wed, Nov 29, 2000 at 12:17:08PM -0800, Bruce Korb wrote:
> > It was just too hard to figure out
> > that this construct was inside of a comment.
> 
> I have another amusing idea -- preprocess the header before and
> after the change, and if there is no non-whitespace change, then
> the header must have been safe beforehand.

That might have worked for the rpc issue, but I think it would have
been more work than implementing a special purpose sed that skips
C comments.  Anyone is welcome to play this game.  Just write a
fix application test or fixup routine that follows the appropriate
prototype:

  typedef apply_fix_p_t t_test_proc PARAMS(( tCC* file, tCC* text ));
  typedef void t_fix_proc PARAMS ((const char *, const char *, tFixDesc
*));

plug it into the appropriate fixup/test table:

  #define FIXUP_TABLE \
    _FT_( "char_macro_def",   char_macro_def_fix ) \
    _FT_( "char_macro_use",   char_macro_use_fix ) \
    _FT_( "format",           format_fix )         \
    _FT_( "machine_name",     machine_name_fix )   \
    _FT_( "wrap",             wrap_fix )           \
    _FT_( "gnu_type",         gnu_type_fix )

  #define FIX_TEST_TABLE \
    _FT_( "machine_name",     machine_name_test )

and modify an entry in the hack table, specifying the new
``c_fix'' or ``c_test'', as appropriate:

  /*
   *  Fix non-ansi machine name defines
   */
  fix = {
    hackname  = machine_name;
    c_test    = machine_name;
    c_fix     = machine_name;

    test_text = "/* MACH_DIFF: */\n"
    "#if defined( i386 ) || defined( sparc ) || defined( vax )"
    "\n/* no uniform test, so be careful  :-) */";
  };

voila.