take account of hide property in rule flattening (fixes #8)

brag/brag/codegen/flatten.rkt

@@ -133,8 +133,22 @@
 
 
 ;; Given a pattern, return a key appropriate for a hash.
+;;
+;; In the `ragg` days this used `syntax->datum` only.
+;; The problem is that with cuts & splices in the mix, it creates ambiguity:
+;; e.g., the pattern (/"," foo)* and ("," foo)* differ only in the 'hide syntax property
+;; so `syntax->datum` does not capture their differences.
+;; That means they produced the same hash key,
+;; which meant they produced the same inferred pattern. Which is wrong.
+;; So we adjust the key to take account of the 'hide property
+;; by "lifting" it into the datum with cons.
+;; Then the pattern-inference process treats them separately.
 (define (pattern->hash-key a-pat)
-  (syntax->datum a-pat))
+  (let loop ([x a-pat])
+    (let ([maybe-stx-list (syntax->list x)])
+      (if maybe-stx-list
+          (cons (syntax-property x 'hide) (map loop maybe-stx-list))
+          (syntax->datum x)))))
 
 
 ;; Returns true if the pattern looks primitive

brag/brag/examples/cutter.rkt

@@ -0,0 +1,4 @@
+#lang brag
+top : expr (/"," expr)*
+expr : "x" | list
+list : "(" expr ("," expr)* ")"

brag/brag/test/test-cutter.rkt

@@ -0,0 +1,9 @@
+#lang racket/base
+(require brag/examples/cutter
+         brag/support
+         rackunit)
+
+;; related to rule-flattening problem
+(check-equal?
+ (parse-to-datum (list "(" "x" "," "x" ")"))
+ '(top (expr (list "(" (expr "x") "," (expr "x") ")"))))