Implement Linux sys_string_conversions using the system APIs.

BUG=14826

Review URL: http://codereview.chromium.org/140062


git-svn-id: svn://svn.chromium.org/chrome/trunk/src@19239 0039d316-1c4b-4281-b951-d872f2087c98

2 files changed, 219 insertions, 4 deletions

diff --git a/base/sys_string_conversions_linux.cc b/base/sys_string_conversions_linux.cc
index 7661d87..ff1d3c3 100644
--- a/base/sys_string_conversions_linux.cc
+++ b/base/sys_string_conversions_linux.cc
@@ -4,6 +4,8 @@
 
 #include "base/sys_string_conversions.h"
 
+#include <wchar.h>
+
 #include "base/string_piece.h"
 #include "base/string_util.h"
 
@@ -23,13 +25,116 @@ std::wstring SysUTF8ToWide(const StringPiece& utf8) {
 }
 
 std::string SysWideToNativeMB(const std::wstring& wide) {
-  // TODO(evanm): we can't assume Linux is UTF-8.
-  return SysWideToUTF8(wide);
+  mbstate_t ps;
+
+  // Calculate the number of multi-byte characters.  We walk through the string
+  // without writing the output, counting the number of multi-byte characters.
+  size_t num_out_chars = 0;
+  memset(&ps, 0, sizeof(ps));
+  for (size_t i = 0; i < wide.size(); ++i) {
+    const wchar_t src = wide[i];
+    // Use a temp buf since a output of NULL does not do what we want.
+    char buf[16];
+    // We don't want wcrtomb to do it's funkiness for embedded NULLs.
+    size_t res = src ? wcrtomb(buf, src, &ps) : 0;
+    switch (res) {
+      // Handle any errors and return an empty string.
+      case -1:
+        return std::string();
+        break;
+      case 0:
+        // We hit an embedded null byte, keep going.
+        ++num_out_chars;
+      default:
+        num_out_chars += res;
+        break;
+    }
+  }
+
+  if (num_out_chars == 0)
+    return std::string();
+
+  std::string out;
+  out.resize(num_out_chars);
+
+  // We walk the input string again, with |i| tracking the index of the
+  // wide input, and |j| tracking the multi-byte output.
+  memset(&ps, 0, sizeof(ps));
+  for (size_t i = 0, j = 0; i < wide.size(); ++i) {
+    const wchar_t src = wide[i];
+    // We don't want wcrtomb to do it's funkiness for embedded NULLs.
+    size_t res = src ? wcrtomb(&out[j], src, &ps) : 0;
+    switch (res) {
+      // Handle any errors and return an empty string.
+      case -1:
+        return std::string();
+        break;
+      case 0:
+        // We hit an embedded null byte, keep going.
+        ++j;  // Output is already 0.
+      default:
+        j += res;
+        break;
+    }
+  }
+
+  return out;
 }
 
 std::wstring SysNativeMBToWide(const StringPiece& native_mb) {
-  // TODO(evanm): we can't assume Linux is UTF-8.
-  return SysUTF8ToWide(native_mb);
+  mbstate_t ps;
+
+  // Calculate the number of wide characters.  We walk through the string
+  // without writing the output, counting the number of wide characters.
+  size_t num_out_chars = 0;
+  memset(&ps, 0, sizeof(ps));
+  for (size_t i = 0; i < native_mb.size(); ) {
+    const char* src = native_mb.data() + i;
+    size_t res = mbrtowc(NULL, src, native_mb.size() - i, &ps);
+    switch (res) {
+      // Handle any errors and return an empty string.
+      case -2:
+      case -1:
+        return std::wstring();
+        break;
+      case 0:
+        // We hit an embedded null byte, keep going.
+        i += 1;  // Fall through.
+      default:
+        i += res;
+        ++num_out_chars;
+        break;
+    }
+  }
+
+  if (num_out_chars == 0)
+    return std::wstring();
+
+  std::wstring out;
+  out.resize(num_out_chars);
+
+  memset(&ps, 0, sizeof(ps));  // Clear the shift state.
+  // We walk the input string again, with |i| tracking the index of the
+  // multi-byte input, and |j| tracking the wide output.
+  for (size_t i = 0, j = 0; i < native_mb.size(); ++j) {
+    const char* src = native_mb.data() + i;
+    wchar_t* dst = &out[j];
+    size_t res = mbrtowc(dst, src, native_mb.size() - i, &ps);
+    switch (res) {
+      // Handle any errors and return an empty string.
+      case -2:
+      case -1:
+        return std::wstring();
+        break;
+      case 0:
+        i += 1;  // Skip null, fall through.
+      default:
+        i += res;
+        break;
+    }
+  }
+
+  return out;
 }
 
 }  // namespace base
diff --git a/base/sys_string_conversions_unittest.cc b/base/sys_string_conversions_unittest.cc
index 2f6e3e4e..4834f00 100644
--- a/base/sys_string_conversions_unittest.cc
+++ b/base/sys_string_conversions_unittest.cc
@@ -3,6 +3,7 @@
 // found in the LICENSE file.
 
 #include "base/string_piece.h"
+#include "base/string_util.h"
 #include "base/sys_string_conversions.h"
 #include "testing/gtest/include/gtest/gtest.h"
 
@@ -66,3 +67,112 @@ TEST(SysStrings, SysUTF8ToWide) {
 
   EXPECT_EQ(expected_null, SysUTF8ToWide(utf8_null));
 }
+
+// We assume the test is running in a UTF8 locale.
+TEST(SysStrings, SysWideToNativeMB) {
+  using base::SysWideToNativeMB;
+  EXPECT_EQ("Hello, world", SysWideToNativeMB(L"Hello, world"));
+  EXPECT_EQ("\xe4\xbd\xa0\xe5\xa5\xbd", SysWideToNativeMB(L"\x4f60\x597d"));
+
+  // >16 bits
+  EXPECT_EQ("\xF0\x90\x8C\x80", SysWideToNativeMB(kSysWideOldItalicLetterA));
+
+  // Error case. When Windows finds a UTF-16 character going off the end of
+  // a string, it just converts that literal value to UTF-8, even though this
+  // is invalid.
+  //
+  // This is what XP does, but Vista has different behavior, so we don't bother
+  // verifying it:
+  //EXPECT_EQ("\xE4\xBD\xA0\xED\xA0\x80zyxw",
+  //          SysWideToNativeMB(L"\x4f60\xd800zyxw"));
+
+  // Test embedded NULLs.
+  std::wstring wide_null(L"a");
+  wide_null.push_back(0);
+  wide_null.push_back('b');
+
+  std::string expected_null("a");
+  expected_null.push_back(0);
+  expected_null.push_back('b');
+
+  EXPECT_EQ(expected_null, SysWideToNativeMB(wide_null));
+}
+
+// We assume the test is running in a UTF8 locale.
+TEST(SysStrings, SysNativeMBToWide) {
+  using base::SysNativeMBToWide;
+  EXPECT_EQ(L"Hello, world", SysNativeMBToWide("Hello, world"));
+  EXPECT_EQ(L"\x4f60\x597d", SysNativeMBToWide("\xe4\xbd\xa0\xe5\xa5\xbd"));
+  // >16 bits
+  EXPECT_EQ(kSysWideOldItalicLetterA, SysNativeMBToWide("\xF0\x90\x8C\x80"));
+
+  // Error case. When Windows finds an invalid UTF-8 character, it just skips
+  // it. This seems weird because it's inconsistent with the reverse conversion.
+  //
+  // This is what XP does, but Vista has different behavior, so we don't bother
+  // verifying it:
+  //EXPECT_EQ(L"\x4f60zyxw", SysNativeMBToWide("\xe4\xbd\xa0\xe5\xa5zyxw"));
+
+  // Test embedded NULLs.
+  std::string utf8_null("a");
+  utf8_null.push_back(0);
+  utf8_null.push_back('b');
+
+  std::wstring expected_null(L"a");
+  expected_null.push_back(0);
+  expected_null.push_back('b');
+
+  EXPECT_EQ(expected_null, SysNativeMBToWide(utf8_null));
+}
+
+static const wchar_t* const kConvertRoundtripCases[] = {
+  L"Google Video",
+  // "网页 图片 资讯更多 »"
+  L"\x7f51\x9875\x0020\x56fe\x7247\x0020\x8d44\x8baf\x66f4\x591a\x0020\x00bb",
+  //  "Παγκόσμιος Ιστός"
+  L"\x03a0\x03b1\x03b3\x03ba\x03cc\x03c3\x03bc\x03b9"
+  L"\x03bf\x03c2\x0020\x0399\x03c3\x03c4\x03cc\x03c2",
+  // "Поиск страниц на русском"
+  L"\x041f\x043e\x0438\x0441\x043a\x0020\x0441\x0442"
+  L"\x0440\x0430\x043d\x0438\x0446\x0020\x043d\x0430"
+  L"\x0020\x0440\x0443\x0441\x0441\x043a\x043e\x043c",
+  // "전체서비스"
+  L"\xc804\xccb4\xc11c\xbe44\xc2a4",
+
+  // Test characters that take more than 16 bits. This will depend on whether
+  // wchar_t is 16 or 32 bits.
+#if defined(WCHAR_T_IS_UTF16)
+  L"\xd800\xdf00",
+  // ?????  (Mathematical Alphanumeric Symbols (U+011d40 - U+011d44 : A,B,C,D,E)
+  L"\xd807\xdd40\xd807\xdd41\xd807\xdd42\xd807\xdd43\xd807\xdd44",
+#elif defined(WCHAR_T_IS_UTF32)
+  L"\x10300",
+  // ?????  (Mathematical Alphanumeric Symbols (U+011d40 - U+011d44 : A,B,C,D,E)
+  L"\x11d40\x11d41\x11d42\x11d43\x11d44",
+#endif
+};
+
+
+TEST(SysStrings, SysNativeMBAndWide) {
+  for (size_t i = 0; i < arraysize(kConvertRoundtripCases); ++i) {
+    std::wstring wide = kConvertRoundtripCases[i];
+    std::wstring trip = base::SysNativeMBToWide(base::SysWideToNativeMB(wide));
+    EXPECT_EQ(wide.size(), trip.size());
+    EXPECT_EQ(wide, trip);
+  }
+
+  // We assume our test is running in UTF-8, so double check through ICU.
+  for (size_t i = 0; i < arraysize(kConvertRoundtripCases); ++i) {
+    std::wstring wide = kConvertRoundtripCases[i];
+    std::wstring trip = base::SysNativeMBToWide(WideToUTF8(wide));
+    EXPECT_EQ(wide.size(), trip.size());
+    EXPECT_EQ(wide, trip);
+  }
+
+  for (size_t i = 0; i < arraysize(kConvertRoundtripCases); ++i) {
+    std::wstring wide = kConvertRoundtripCases[i];
+    std::wstring trip = UTF8ToWide(base::SysWideToNativeMB(wide));
+    EXPECT_EQ(wide.size(), trip.size());
+    EXPECT_EQ(wide, trip);
+  }
+}