Write a filter "yucca" (Yetanother Unicode Compiler-Compiler Abomination) which takes a more yacc-like grammar like my older "tacc" parser, e.g.

keywordargument : ID EQ boolean             { $$ = MakeBinary (NodeTag.arg, $1, $3); }
                | ID EQ NUMBER              { $$ = MakeBinary (NodeTag.arg, $1, $3); }
                | ID EQ size_vector         { $$ = MakeBinary (NodeTag.arg, $1, $3); }
                | ID EQ vector              { $$ = MakeBinary (NodeTag.arg, $1, $3); }
                | ID EQ point_2d            { $$ = MakeBinary (NodeTag.arg, $1, $3); }
                | ID EQ stripped_string     { $$ = MakeBinary (NodeTag.arg, $1, $3); }
                ;

I guess really all I need to do is add the $$ and $n symbols.  ${nn} for >= 10.  Maybe ${name} for named args, or ${name:num} for multiple occurances of same phrase.


Also need a syntax to handle grouping, as in P<rule> = <item>  ( "," <item> )* ;

- maybe ${item:2[n]} - :2 is second explicit occurance of item, n is index of element of multiple instances of that subphrase.
Unfortunately the linked list structure means that it doesn't map to an array element but to <n> links down a linked list.
Equivalent to

P<rule> = <item> <rule_2> ;
P<rule_2> = "," <item> <rule_2>, ;

Which suggests:
 P<rule> = <item>  ( "," <item> )* ;
   |       |       |  |     |
   $$      $1      $2 |     |
                      |     |
                      $2_$1 |
                            |
                            $2_$2

$2_$2 could also be referred to as $2_${item}



Add << in emacs filter for .g files to convert to guillemets.


Add an alternative syntax for regexps that doesn't use unicode guillemets, for ease of editing.