[ruby-dev:10689] Re: a regex bug?

まつもと ゆきひろです

In message "[ruby-dev:10683] Re: a regex bug?"
    on 00/08/26, Tanaka Akira <[e-mail]> writes:

|ちなみに、(まだ未公開の) Emacs21 にも同様な問題があって、半田さんを通
|して問題を伝えたのですが、Stefan Monnier がさくっと直してしまったよう
|です。

そのパッチを半田さんから送って頂きました。で、それをヒントに
したものが添付のパッチです。

なんとなく動いているようです。しかし、これをさくっと直せるっ
てのは凄いなあ。

--- regex.c	2000/08/25 08:26:03	1.28
+++ regex.c	2000/08/26 16:56:23
@@ -151,2 +151,18 @@
 
+#define ENSURE_FAIL_STACK(n)						\
+  do {									\
+    if (stacke - stackp <= (n)) {					\
+	unsigned char **stackx;						\
+	unsigned int len = stacke - stackb;				\
+	/* if (len > re_max_failures * MAX_NUM_FAILURE_ITEMS)		\
+	   {								\
+	   FREE_VARIABLES();						\
+	   FREE_AND_RETURN(stackb,(-2));				\
+	   }*/								\
+									\
+        /* Roughly double the size of the stack.  */			\
+        EXPAND_FAIL_STACK(stackx, stackb, len);				\
+      }									\
+  } while (0)
+
 /* Get the interface, including the syntax bits.  */
@@ -2071,3 +2087,3 @@
 	  store_jump_n(b, greedy?jump_n:finalize_push_n, laststart + 5,
-		       upper_bound/* - 1*/);
+		       upper_bound - 1);
 	  b += 5;
@@ -2089,3 +2105,3 @@
 	  insert_op_2(set_number_at, laststart, b, b - laststart,
-		      upper_bound/* - 1*/);
+		      upper_bound - 1);
 	  b += 5;
@@ -3331,2 +3347,5 @@
 
+/* I.e., ptr and count.  */
+#define NUM_COUNT_ITEMS 2
+
 /* Individual items aside from the registers.  */
@@ -3340,5 +3359,22 @@
 /* We push this many things on the stack whenever we fail.  */
-#define NUM_FAILURE_ITEMS  (last_used_reg * NUM_REG_ITEMS + NUM_REG_ITEMS)
+#define NUM_FAILURE_ITEMS  (num_failure_counts * NUM_COUNT_ITEMS + last_used_reg * NUM_REG_ITEMS + NUM_REG_ITEMS)
 
+/* This pushes counter information for succeed_n and jump_n */
+#define PUSH_FAILURE_COUNT(ptr)						\
+  do {									\
+    int c;								\
+    EXTRACT_NUMBER(c, ptr);						\
+    ENSURE_FAIL_STACK(2);						\
+    *stackp++ = (unsigned char*)c;					\
+    *stackp++ = (ptr);							\
+    num_failure_counts++;						\
+  } while (0)
 
+#define POP_FAILURE_COUNT()						\
+  do {									\
+    unsigned char *ptr = *--stackp;					\
+    int count = (long)*--stackp;					\
+    STORE_NUMBER(ptr, count);						\
+  } while (0)
+
 /* This pushes most of the information about the current state we will want
@@ -3356,15 +3392,6 @@
 									\
-    if (stacke - stackp <= NUM_FAILURE_ITEMS) {				\
-	unsigned char **stackx;						\
-	unsigned int len = stacke - stackb;				\
-	/* if (len > re_max_failures * MAX_NUM_FAILURE_ITEMS)		\
-	   {								\
-	   FREE_VARIABLES();						\
-	   FREE_AND_RETURN(stackb,(-2));				\
-	   }*/								\
+    ENSURE_FAIL_STACK(NUM_FAILURE_ITEMS);				\
+    *stackp++ = (unsigned char*)num_failure_counts;			\
+    num_failure_counts = 0;						\
 									\
-        /* Roughly double the size of the stack.  */			\
-        EXPAND_FAIL_STACK(stackx, stackb, len);				\
-      }									\
-									\
     /* Now push the info for each of those registers.  */		\
@@ -3386,3 +3413,3 @@
 
-     /* This pops what PUSH_FAILURE_POINT pushes.  */
+/* This pops what PUSH_FAILURE_POINT pushes.  */
 
@@ -3395,2 +3422,7 @@
     stackp -= temp; 		/* Remove the register info.  */	\
+    temp = (long)*--stackp;	/* How many counters pushed.  */	\
+    while (temp--) {							\
+      POP_FAILURE_COUNT();      /* Remove the counter info.  */		\
+    }									\
+    num_failure_counts = 0;	/* Reset num_failure_counts.  */	\
   } while(0)
@@ -3542,2 +3574,4 @@
 
+  int num_failure_counts = 0;
+
   if (regs) {
@@ -3903,3 +3937,3 @@
 	   would have to backtrack because of (as in, e.g., `a*a')
-	   then we can change to pop_failure_jump, because we'll
+	   then we can change to finalize_jump, because we'll
 	   never have to backtrack.
@@ -4027,5 +4061,6 @@
 	/* Originally, this is how many times we HAVE to succeed.  */
-	if (mcnt > 0) {
+	if (mcnt != 0) {
 	  mcnt--;
 	  p += 2;
+	  PUSH_FAILURE_COUNT(p);
 	  STORE_NUMBER_AND_INCR(p, mcnt);
@@ -4033,5 +4068,3 @@
 	}
-	else if (mcnt == 0) {
-	  p[2] = unused;
-	  p[3] = unused;
+	else  {
 	  goto on_failure;
@@ -4045,2 +4078,3 @@
 	  mcnt--;
+	  PUSH_FAILURE_COUNT(p + 2);
 	  STORE_NUMBER(p + 2, mcnt);
@@ -4261,2 +4295,6 @@
 	regstart[this_reg] = *--stackp;
+      }
+      mcnt = (long)*--stackp;
+      while (mcnt--) {
+	POP_FAILURE_COUNT();
       }