[regexp] correctly advance zero length matches for global/unicode.

test/mjsunit/harmony/unicode-regexp-zero-length.js

+// Copyright 2016 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+// Flags: --harmony-unicode-regexps
+
+var L = "\ud800";
+var T = "\udc00";
+var x = "x";
+
+var r = /()/g;  // Global, but not unicode.
+// Zero-length matches do not advance lastIndex.
+assertEquals(["", ""], r.exec(L + T + L + T));
+assertEquals(0, r.lastIndex);
+r.lastIndex = 1;
+assertEquals(["", ""], r.exec(L + T + L + T));
+assertEquals(1, r.lastIndex);
+
+var u = /()/ug;  // Global and unicode.
+// Zero-length matches do not advance lastIndex.
+assertEquals(["", ""], u.exec(L + T + L + T));
+assertEquals(0, u.lastIndex);
+u.lastIndex = 1;
+assertEquals(["", ""], u.exec(L + T + L + T));
+assertEquals(0, u.lastIndex);
+
+// However, with repeating matches, lastIndex does not matter.
+// We do advance from match to match.
+r.lastIndex = 2;
+assertEquals(x + L + x + T + x + L + x + T + x,
+             (L + T + L + T).replace(r, "x"));
+
+// With unicode flag, we advance code point by code point.
+u.lastIndex = 3;
+assertEquals(x + L + T + x + L + T + x,
+             (L + T + L + T).replace(u, "x"));
+
+// Test that exhausting the global match cache is fine.
+assertEquals((x + L + T).repeat(1000) + x,
+             (L + T).repeat(1000).replace(u, "x"));
+
+// Same thing for RegExp.prototype.match.
+r.lastIndex = 1;
+assertEquals(["","","","",""], (L + T + L + T).match(r));
+r.lastIndex = 2;
+assertEquals(["","","","",""], (L + T + L + T).match(r));
+
+u.lastIndex = 1;
+assertEquals(["","",""], (L + T + L + T).match(u));
+u.lastIndex = 2;
+assertEquals(["","",""], (L + T + L + T).match(u));
+
+var expected = [];
+for (var i = 0; i <= 1000; i++) expected.push("");
+assertEquals(expected, (L + T).repeat(1000).match(u));