****************************************************************


- cut and paste grammar from 
  http://www.w3.org/TR/2008/WD-rif-bld-20080730/#EBNF_for_the_Rule_Language

- Change EBNF to BNF

   - replace parenthesized phrases with named phrases, as needed:
       (RULE | Group)                    RULE_or_Group
       (Name '->' TERM)                  Name_arrow_TERM
       (TERM '->' TERM)                  TERM_arrow_TERM
       (Frame | 'And' '(' Frame* ')')    Frame_or_AndFrame

   - replace Kleene operators:
        '* '      '_star '
        '? '      '_opt '
        '+ '      '_plus '

   - add Kleene operator rules (run grammar_gen.py)

- Remove parens...

- Split some disjunctive productions into multiple productions

    (just a random judgement call?)

- Trivial syntax changes

   - left align the text

   - turn "::=" into ":"

   - wrap lines in rule 

         def p_X_N(t):
            'X : ...'
            pass
     
      (where N is branch N -- if there's more than one)

      IE use:
         python grammar_gen.py  < grammar > gend

****************************************************************


lots of dirty work....
see all the commented bits!


****************************************************************



# EXISTING TEST CASES don't quote URIs in Prefix, even given close-paren


****************
   in bld 4:

        x:pd[dc:publisher -> http://www.w3.org/

    BAD IDEA.   BARE_IRI

****************
  
    two reduce/reduce conflicts around name_arrow_term_star

      1.  solved by making the lexer find NAME_ARROW

      2.  solved by making it name_arrow_term_plus, since
          it was ambiguous which kind of uniterm it was when
	  both were absent.

****************

    shift/reduce:

       -- integrate the _opt logic into the main meta_opt rule

       -- remove external frames

       -- different keyword for external terms and external formulas

              state 109

                without this, we have:

               (18) FORMULA -> IRIMETA_opt . KW_External LPAREN Atom RPAREN
              vs
               (20) ATOMIC -> IRIMETA_opt . Equal
               (27) Equal -> . TERM EQUALS TERM
               (34) TERM -> . IRIMETA_opt KW_External LPAREN Expr RPAREN

               So when we get a KW_External, we don't know whether
               we're on the first branch or the second one; on the second,
               we're skipping a null IRIMETA_opt

       -- an ATOM keyword to prefix atoms

              without it, it's unclear to what the metadata applies.

              (* ... *)  p(a) = q(a)

              which should it be?
                p
                p(a)
                p(a)=q(a)

          (* metadata 
          ***
          stuff
          *)



       -- a metasep keyword

              state 2

               given this:  (* a:b 

               was there an omitted metadata oportunity in there?
               if it turns out the a:b is the beginning of a frame,
               then there was.  If it turns out to be the ID in the
	       metadata, then there was not.    Of course, we don't
               care about missed metadata opportunities.


      ALL S/R conflicts are against this rule:
                rule 53 (IRIMETA_opt -> .)
      so that should be okay to go either way.

****************


bld-4:

   (* "http://sample.org"^^rif:iri x:pd[dc:publish...

   (* <a> <b>[...] *)
   (* <b>[...] *)


  -- quoting strings
  -- no bare IRIs  (even in prefix and base)
  -- do allow null prefix (somehow)
  -- _:foo as shortcut for local?   or SOMETHING


****************************************************************

The only thing I think I needed to mess with on BLD documents
is at the annotation frame objects are LOCALNAMES and I don't
allow those.

****************************************************************

Need:
    - run on wiki test cases
    - lots of little syntax tests
    - XML reading, writing
    - write a PLY -> html, for nicely seeing the resulting grammar
      (maybe remove the useless productions?)



****************************************************************

Where the RIF XML syntax differs from pure striping:

     - maybe in annotations, depending how you model them
         i gues "id",and "meta", work fine.   Nothing special there.
     - slot
     - args
     - const
     - var

   var->  something with text is taken as an implicit "label" ?
   args->   ordered=yes is like parsetype=collection
   slots->  same -- it's list-valued-per slot, BUT there are ALSO
            multiple slots.

                  type = [TERM]     (list of terms?)

   Const.

        so it's where we have  type=   and   ordered=yes
        and Var with its weird having-content.


<Var>x</Var>

      <Var><name>x</name></Var>

<Const type="x">y</Const>

      <Const>
          <datatype>x</datatype>
          <lexrep>y</lexrep>
      </Const>

<args ordered="yes">
     a b c
</args>

      <args>
          <Sequence>
              a b c
          </Sequence>
      <args>

<slot ordered="yes">
    a b
</slot>

      <slot>
           <Sequence>
             a b
           </Sequence>
      </slot>

   OR

      <slot>
          <Slot>
            <key>a</key>
            <value>b</value>
          </Slot>
      </slot>

?

Better go with "Sequence", I think.
   in the RIF model.

****************************************************************


Do we want barewords to be local names?  Or is that a poor use of them.
Better to have 
   Prefix(: <foo>)
?