[libfuzzer] moving icu fuzzers to //third_party/icu, additional fuzzers.

fuzzers/icu_break_iterator_fuzzer.cc

+// Copyright 2016 The Chromium Authors. All rights reserved.
+
+#include <stddef.h>
+#include <stdint.h>
+#include <memory>
+#include "third_party/icu/fuzzers/fuzzer_utils.h"
+#include "third_party/icu/source/common/unicode/brkiter.h"
+
+IcuEnvironment* env = new IcuEnvironment();
+
+// Entry point for LibFuzzer.
+extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size) {
+  UErrorCode status = U_ZERO_ERROR;
+  icu::UnicodeString str(reinterpret_cast<const char*>(data), size);

[jungshik at Google, 2016/05/19 07:21:54]

Again, this will test how well UnicodeString can deal with mostly invalid  input
strings but BreakIterator will not be  tested much.  (with valid input strings,
there can be hidden edge cases that may trigger a bug in BreakIterator). 

There should be a helper function to generate valid random strings in UTF-16 or
UTF-8. That way, One way would be to generate a sequence of random numbers ( [0,
0x10FFFF] ) and convert them to UTF-16 or UTF-8 strings. 

+
+  auto rng = CreateRng(data, size);
+  const icu::Locale& locale = GetRandomLocale(&rng);
+
+  std::unique_ptr<icu::BreakIterator> bi;
+
+  switch (rng() % 5) {
+    case 0:
+      bi.reset(icu::BreakIterator::createWordInstance(locale, status));
+      break;
+    case 1:
+      bi.reset(icu::BreakIterator::createLineInstance(locale, status));
+      break;
+    case 2:
+      bi.reset(icu::BreakIterator::createCharacterInstance(locale, status));
+      break;
+    case 3:
+      bi.reset(icu::BreakIterator::createSentenceInstance(locale, status));
+      break;
+    case 4:
+      bi.reset(icu::BreakIterator::createTitleInstance(locale, status));
+      break;
+  }
+  if (U_FAILURE(status)) return 0;
+
+  for (int32_t p = bi->first(); p != icu::BreakIterator::DONE; p = bi->next())
+    if (U_FAILURE(status)) return 0;
+
+  return 0;
+}
+