Implement Linux sys_string_conversions using the system APIs.

base/sys_string_conversions_linux.cc

Index: base/sys_string_conversions_linux.cc
diff --git a/base/sys_string_conversions_linux.cc b/base/sys_string_conversions_linux.cc
index 7661d8730b92705592f02fed2cc82054df0bc3a2..118f0ac48494e2048079457f1f6b094ab2f21b33 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 +25,120 @@ 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 buffer since calling wcrtomb with an output of NULL does not
+    // calculate the output length.
+    char buf[16];
+    // Skip NULLs to avoid wcrtomb's special handling of them.
+    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;
+        break;
+      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 zeroed.
+        break;
+      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;

[Evan Martin, 2009/06/26 15:19:26]

Style frowns upon abbreviated variable names.  "mbstate" woulda worked.

[Dean McNamee, 2009/06/26 15:23:07]

Yeah, that's the style they give in the man page so I just copied it.  You're
probably right though.

+
+  // 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 byte.
+        break;
+      default:
+        i += res;
+        break;
+    }
+  }
+
+  return out;
 }
 
 }  // namespace base