Fix that "ss" in look-behind causes syntax error

Fixes #92.

This fix was ported from oniguruma:
kkos/oniguruma@257082d

Author: k-takata

regcomp.c

@@ -3264,6 +3264,14 @@ setup_subexp_call(Node* node, ScanEnv* env)
 }
 #endif
 
+#define IN_ALT          (1<<0)
+#define IN_NOT          (1<<1)
+#define IN_REPEAT       (1<<2)
+#define IN_VAR_REPEAT   (1<<3)
+#define IN_CALL         (1<<4)
+#define IN_RECCALL      (1<<5)
+#define IN_LOOK_BEHIND  (1<<6)
+
 /* divide different length alternatives in look-behind.
   (?<=A|B) ==> (?<=A)|(?<=B)
   (?<!A|B) ==> (?<!A)(?<!B)
@@ -3560,24 +3568,29 @@ expand_case_fold_string_alt(int item_num, OnigCaseFoldCodeItem items[],
   return ONIGERR_MEMORY;
 }
 
-static int
-expand_case_fold_string(Node* node, regex_t* reg)
-{
 #define THRESHOLD_CASE_FOLD_ALT_FOR_EXPANSION  8
 
+static int
+expand_case_fold_string(Node* node, regex_t* reg, int state)
+{
   int r, n, len, alt_num;
   int varlen = 0;
+  int is_in_look_behind;
   UChar *start, *end, *p;
   Node *top_root, *root, *snode, *prev_node;
   OnigCaseFoldCodeItem items[ONIGENC_GET_CASE_FOLD_CODES_MAX_NUM];
-  StrNode* sn = NSTR(node);
+  StrNode* sn;
 
   if (NSTRING_IS_AMBIG(node)) return 0;
 
+  sn = NSTR(node);
+
   start = sn->s;
   end   = sn->end;
   if (start >= end) return 0;
 
+  is_in_look_behind = (state & IN_LOOK_BEHIND) != 0;
+
   r = 0;
   top_root = root = prev_node = snode = NULL_NODE;
   alt_num = 1;
@@ -3593,7 +3606,7 @@ expand_case_fold_string(Node* node, regex_t* reg)
     len = enclen(reg->enc, p, end);
 
     varlen = is_case_fold_variable_len(n, items, len);
-    if (n == 0 || varlen == 0) {
+    if (n == 0 || varlen == 0 || is_in_look_behind) {
       if (IS_NULL(snode)) {
 	if (IS_NULL(root) && IS_NOT_NULL(prev_node)) {
           onig_node_free(top_root);
@@ -3854,13 +3867,6 @@ setup_comb_exp_check(Node* node, int state, ScanEnv* env)
 }
 #endif
 
-#define IN_ALT        (1<<0)
-#define IN_NOT        (1<<1)
-#define IN_REPEAT     (1<<2)
-#define IN_VAR_REPEAT (1<<3)
-#define IN_CALL       (1<<4)
-#define IN_RECCALL    (1<<5)
-
 /* setup_tree does the following work.
  1. check empty loop. (set qn->target_empty_info)
  2. expand ignore-case in char class.
@@ -3902,7 +3908,7 @@ setup_tree(Node* node, regex_t* reg, int state, ScanEnv* env)
 
   case NT_STR:
     if (IS_IGNORECASE(reg->options) && !NSTRING_IS_RAW(node)) {
-      r = expand_case_fold_string(node, reg);
+      r = expand_case_fold_string(node, reg, state);
     }
     break;
 
@@ -4145,7 +4151,7 @@ setup_tree(Node* node, regex_t* reg, int state, ScanEnv* env)
 	  if (r < 0) return r;
 	  if (r > 0) return ONIGERR_INVALID_LOOK_BEHIND_PATTERN;
 	  if (NTYPE(node) != NT_ANCHOR) goto restart;
-	  r = setup_tree(an->target, reg, state, env);
+	  r = setup_tree(an->target, reg, (state | IN_LOOK_BEHIND), env);
 	  if (r != 0) return r;
 	  r = setup_look_behind(node, reg, env);
 	}
@@ -4158,7 +4164,8 @@ setup_tree(Node* node, regex_t* reg, int state, ScanEnv* env)
 	  if (r < 0) return r;
 	  if (r > 0) return ONIGERR_INVALID_LOOK_BEHIND_PATTERN;
 	  if (NTYPE(node) != NT_ANCHOR) goto restart;
-	  r = setup_tree(an->target, reg, (state | IN_NOT), env);
+	  r = setup_tree(an->target, reg, (state | IN_NOT | IN_LOOK_BEHIND),
+			 env);
 	  if (r != 0) return r;
 	  r = setup_look_behind(node, reg, env);
 	}

testpy.py

@@ -1172,6 +1172,28 @@ def main():
         x2("(?i)(?<=\u0149)a", "\u02bcna", 2, 3)    # with look-behind
         # Other Unicode tests
         x2("\\x{25771}", "\U00025771", 0, 1)
+    x2("(?i:ss)", "ss", 0, 2)
+    x2("(?i:ss)", "Ss", 0, 2)
+    x2("(?i:ss)", "SS", 0, 2)
+    if is_unicode_encoding(onig_encoding):
+        x2("(?i:ss)", "\u017fS", 0, 2)  # LATIN SMALL LETTER LONG S
+        x2("(?i:ss)", "s\u017f", 0, 2)
+        x2("(?i:ss)", "\u00df", 0, 1)   # LATIN SMALL LETTER SHARP S
+        x2("(?i:ss)", "\u1e9e", 0, 1)   # LATIN CAPITAL LETTER SHARP S
+    x2("(?i:xssy)", "xssy", 0, 4)
+    x2("(?i:xssy)", "xSsy", 0, 4)
+    x2("(?i:xssy)", "xSSy", 0, 4)
+    if is_unicode_encoding(onig_encoding):
+        x2("(?i:xssy)", "x\u017fSy", 0, 4)
+        x2("(?i:xssy)", "xs\u017fy", 0, 4)
+        x2("(?i:xssy)", "x\u00dfy", 0, 3)
+        x2("(?i:xssy)", "x\u1e9ey", 0, 3)
+        x2("(?i:\u00df)", "ss", 0, 2)
+        x2("(?i:\u00df)", "SS", 0, 2)
+        x2("(?i:[\u00df])", "ss", 0, 2)
+        x2("(?i:[\u00df])", "SS", 0, 2)
+    x2("(?i)(?<!ss)z", "qqz", 2, 3)     # Issue #92
+    x2("(?i)(?<!xss)z", "qqz", 2, 3)
     x2("[0-9-a]+", " 0123456789-a ", 1, 13)     # same as [0-9\-a]
     x2("[0-9-\\s]+", " 0123456789-a ", 0, 12)   # same as [0-9\-\s]
     n("[0-9-a]", "", syn=onigmo.ONIG_SYNTAX_GREP, err=onigmo.ONIGERR_UNMATCHED_RANGE_SPECIFIER_IN_CHAR_CLASS)