Handle UTF-8

This adds full UTF-8 support both on input and output.

Input: read and validate full UTF-8, both Basic Multilingual Plane
and extended characters.
Collate surrogate pairs as specified in RFC4627.
This ensures that UTF-8 strings that reach the application are
always valid, and that invalid UTF-8 fails parsing.

Output: Assume UTF-8 strings provided for output are valid.
The escaping was broken, fix this by not encoding UTF-8 characters with \u.
Writing them to the output stream as-is is the right thing to do.
See https://www.ietf.org/rfc/rfc4627.txt:

    "JSON text SHALL be encoded in Unicode.  The default encoding is UTF-8."

Also add tests for the new functionality.
Fixes #16.

Makefile.am

@@ -3,7 +3,7 @@ ACLOCAL_AMFLAGS = -I build-aux/m4
 .INTERMEDIATE: $(GENBIN)
 
 include_HEADERS = include/univalue.h
-noinst_HEADERS = lib/univalue_escapes.h
+noinst_HEADERS = lib/univalue_escapes.h lib/univalue_utffilter.h
 
 lib_LTLIBRARIES = libunivalue.la
 
@@ -73,6 +73,10 @@ TEST_FILES = \
 	$(TEST_DATA_DIR)/fail35.json \
 	$(TEST_DATA_DIR)/fail36.json \
 	$(TEST_DATA_DIR)/fail37.json \
+	$(TEST_DATA_DIR)/fail38.json \
+	$(TEST_DATA_DIR)/fail39.json \
+	$(TEST_DATA_DIR)/fail40.json \
+	$(TEST_DATA_DIR)/fail41.json \
 	$(TEST_DATA_DIR)/fail3.json \
 	$(TEST_DATA_DIR)/fail4.json \
 	$(TEST_DATA_DIR)/fail5.json \
@@ -83,6 +87,7 @@ TEST_FILES = \
 	$(TEST_DATA_DIR)/pass1.json \
 	$(TEST_DATA_DIR)/pass2.json \
 	$(TEST_DATA_DIR)/pass3.json \
-	$(TEST_DATA_DIR)/round1.json
+	$(TEST_DATA_DIR)/round1.json \
+	$(TEST_DATA_DIR)/round2.json
 
 EXTRA_DIST=$(TEST_FILES) $(GEN_SRCS)

lib/univalue_read.cpp

@@ -6,6 +6,7 @@
 #include <vector>
 #include <stdio.h>
 #include "univalue.h"
+#include "univalue_utffilter.h"
 
 using namespace std;
 
@@ -174,41 +175,31 @@ enum jtokentype getJsonToken(string& tokenVal, unsigned int& consumed,
         raw++;                                // skip "
 
         string valStr;
+        JSONUTF8StringFilter writer(valStr);
 
         while (*raw) {
-            if (*raw < 0x20)
+            if ((unsigned char)*raw < 0x20)
                 return JTOK_ERR;
 
             else if (*raw == '\\') {
                 raw++;                        // skip backslash
 
                 switch (*raw) {
-                case '"':  valStr += "\""; break;
-                case '\\': valStr += "\\"; break;
-                case '/':  valStr += "/"; break;
-                case 'b':  valStr += "\b"; break;
-                case 'f':  valStr += "\f"; break;
-                case 'n':  valStr += "\n"; break;
-                case 'r':  valStr += "\r"; break;
-                case 't':  valStr += "\t"; break;
+                case '"':  writer.push_back('\"'); break;
+                case '\\': writer.push_back('\\'); break;
+                case '/':  writer.push_back('/'); break;
+                case 'b':  writer.push_back('\b'); break;
+                case 'f':  writer.push_back('\f'); break;
+                case 'n':  writer.push_back('\n'); break;
+                case 'r':  writer.push_back('\r'); break;
+                case 't':  writer.push_back('\t'); break;
 
                 case 'u': {
                     unsigned int codepoint;
                     if (hatoui(raw + 1, raw + 1 + 4, codepoint) !=
                                raw + 1 + 4)
                         return JTOK_ERR;
-
-                    if (codepoint <= 0x7f)
-                        valStr.push_back((char)codepoint);
-                    else if (codepoint <= 0x7FF) {
-                        valStr.push_back((char)(0xC0 | (codepoint >> 6)));
-                        valStr.push_back((char)(0x80 | (codepoint & 0x3F)));
-                    } else if (codepoint <= 0xFFFF) {
-                        valStr.push_back((char)(0xE0 | (codepoint >> 12)));
-                        valStr.push_back((char)(0x80 | ((codepoint >> 6) & 0x3F)));
-                        valStr.push_back((char)(0x80 | (codepoint & 0x3F)));
-                    }
-
+                    writer.push_back_u(codepoint);
                     raw += 4;
                     break;
                     }
@@ -226,11 +217,13 @@ enum jtokentype getJsonToken(string& tokenVal, unsigned int& consumed,
             }
 
             else {
-                valStr += *raw;
+                writer.push_back(*raw);
                 raw++;
             }
         }
 
+        if (!writer.finalize())
+            return JTOK_ERR;
         tokenVal = valStr;
         consumed = (raw - rawStart);
         return JTOK_STRING;

lib/univalue_utffilter.h

@@ -0,0 +1,119 @@
+// Copyright 2016 Wladimir J. van der Laan
+// Distributed under the MIT software license, see the accompanying
+// file COPYING or http://www.opensource.org/licenses/mit-license.php.
+#ifndef UNIVALUE_UTFFILTER_H
+#define UNIVALUE_UTFFILTER_H
+
+#include <string>
+
+/**
+ * Filter that generates and validates UTF-8, as well as collates UTF-16
+ * surrogate pairs as specified in RFC4627.
+ */
+class JSONUTF8StringFilter
+{
+public:
+    JSONUTF8StringFilter(std::string &s):
+        str(s), is_valid(true), codepoint(0), state(0), surpair(0)
+    {
+    }
+    // Write single 8-bit char (may be part of UTF-8 sequence)
+    void push_back(unsigned char ch)
+    {
+        if (state == 0) {
+            if (ch < 0x80) // 7-bit ASCII, fast direct pass-through
+                str.push_back(ch);
+            else if (ch < 0xc0) // Mid-sequence character, invalid in this state
+                is_valid = false;
+            else if (ch < 0xe0) { // Start of 2-byte sequence
+                codepoint = (ch & 0x1f) << 6;
+                state = 6;
+            } else if (ch < 0xf0) { // Start of 3-byte sequence
+                codepoint = (ch & 0x0f) << 12;
+                state = 12;
+            } else if (ch < 0xf8) { // Start of 4-byte sequence
+                codepoint = (ch & 0x07) << 18;
+                state = 18;
+            } else // Reserved, invalid
+                is_valid = false;
+        } else {
+            if ((ch & 0xc0) != 0x80) // Not a continuation, invalid
+                is_valid = false;
+            state -= 6;
+            codepoint |= (ch & 0x3f) << state;
+            if (state == 0)
+                push_back_u(codepoint);
+        }
+    }
+    // Write codepoint directly, possibly collating surrogate pairs
+    void push_back_u(unsigned int codepoint)
+    {
+        if (state) // Only accept full codepoints in open state
+            is_valid = false;
+        if (codepoint >= 0xD800 && codepoint < 0xDC00) { // First half of surrogate pair
+            if (surpair) // Two subsequent surrogate pair openers - fail
+                is_valid = false;
+            else
+                surpair = codepoint;
+        } else if (codepoint >= 0xDC00 && codepoint < 0xE000) { // Second half of surrogate pair
+            if (surpair) { // Open surrogate pair, expect second half
+                // Compute code point from UTF-16 surrogate pair
+                append_codepoint(0x10000 | ((surpair - 0xD800)<<10) | (codepoint - 0xDC00));
+                surpair = 0;
+            } else // Second half doesn't follow a first half - fail
+                is_valid = false;
+        } else {
+            if (surpair) // First half of surrogate pair not followed by second - fail
+                is_valid = false;
+            else
+                append_codepoint(codepoint);
+        }
+    }
+    // Check that we're in a state where the string can be ended
+    // No open sequences, no open surrogate pairs, etc
+    bool finalize()
+    {
+        if (state || surpair)
+            is_valid = false;
+        return is_valid;
+    }
+private:
+    std::string &str;
+    bool is_valid;
+    // Current UTF-8 decoding state
+    unsigned int codepoint;
+    int state; // Top bit to be filled in for next UTF-8 byte, or 0
+
+    // Keep track of the following state to handle the following section of
+    // RFC4627:
+    //
+    //    To escape an extended character that is not in the Basic Multilingual
+    //    Plane, the character is represented as a twelve-character sequence,
+    //    encoding the UTF-16 surrogate pair.  So, for example, a string
+    //    containing only the G clef character (U+1D11E) may be represented as
+    //    "\uD834\uDD1E".
+    //
+    //  Two subsequent \u.... may have to be replaced with one actual codepoint.
+    unsigned int surpair; // First half of open UTF-16 surrogate pair, or 0
+
+    void append_codepoint(unsigned int codepoint)
+    {
+        if (codepoint <= 0x7f)
+            str.push_back((char)codepoint);
+        else if (codepoint <= 0x7FF) {
+            str.push_back((char)(0xC0 | (codepoint >> 6)));
+            str.push_back((char)(0x80 | (codepoint & 0x3F)));
+        } else if (codepoint <= 0xFFFF) {
+            str.push_back((char)(0xE0 | (codepoint >> 12)));
+            str.push_back((char)(0x80 | ((codepoint >> 6) & 0x3F)));
+            str.push_back((char)(0x80 | (codepoint & 0x3F)));
+        } else if (codepoint <= 0x1FFFFF) {
+            str.push_back((char)(0xF0 | (codepoint >> 18)));
+            str.push_back((char)(0x80 | ((codepoint >> 12) & 0x3F)));
+            str.push_back((char)(0x80 | ((codepoint >> 6) & 0x3F)));
+            str.push_back((char)(0x80 | (codepoint & 0x3F)));
+        }
+    }
+};
+
+#endif

lib/univalue_write.cpp

@@ -8,8 +8,6 @@
 #include "univalue.h"
 #include "univalue_escapes.h"
 
-// TODO: Using UTF8
-
 using namespace std;
 
 static string json_escape(const string& inS)
@@ -23,15 +21,8 @@ static string json_escape(const string& inS)
 
         if (escStr)
             outS += escStr;
-
-        else if (ch < 0x80)
+        else
             outS += ch;
-
-        else { // TODO handle UTF-8 properly
-            char tmpesc[16];
-            sprintf(tmpesc, "\\u%04x", ch);
-            outS += tmpesc;
-        }
     }
 
     return outS;

test/fail38.json

@@ -0,0 +1 @@
+["\ud834"]

test/fail39.json

@@ -0,0 +1 @@
+["\udd61"]

test/fail40.json

@@ -0,0 +1 @@
+["���"]

test/fail41.json

@@ -0,0 +1 @@
+["�"]

test/round2.json

@@ -0,0 +1 @@
+["a§■𐎒𝅘𝅥𝅯"]

test/unitester.cpp

@@ -22,6 +22,7 @@ string srcdir(JSON_TEST_SRC);
 static bool test_failed = false;
 
 #define d_assert(expr) { if (!(expr)) { test_failed = true; fprintf(stderr, "%s failed\n", filename.c_str()); } }
+#define f_assert(expr) { if (!(expr)) { test_failed = true; fprintf(stderr, "%s failed\n", __func__); } }
 
 static std::string rtrim(std::string s)
 {
@@ -108,6 +109,10 @@ static const char *filenames[] = {
         "fail35.json",
         "fail36.json",
         "fail37.json",
+        "fail38.json",               // invalid unicode: only first half of surrogate pair
+        "fail39.json",               // invalid unicode: only second half of surrogate pair
+        "fail40.json",               // invalid unicode: broken UTF-8
+        "fail41.json",               // invalid unicode: unfinished UTF-8
         "fail3.json",
         "fail4.json",                // extra comma
         "fail5.json",
@@ -119,14 +124,40 @@ static const char *filenames[] = {
         "pass2.json",
         "pass3.json",
         "round1.json",              // round-trip test
+        "round2.json",              // unicode
 };
 
+// Test \u handling
+void unescape_unicode_test()
+{
+    UniValue val;
+    bool testResult;
+    // Escaped ASCII (quote)
+    testResult = val.read("[\"\\u0022\"]");
+    f_assert(testResult);
+    f_assert(val[0].get_str() == "\"");
+    // Escaped Basic Plane character, two-byte UTF-8
+    testResult = val.read("[\"\\u0191\"]");
+    f_assert(testResult);
+    f_assert(val[0].get_str() == "\xc6\x91");
+    // Escaped Basic Plane character, three-byte UTF-8
+    testResult = val.read("[\"\\u2191\"]");
+    f_assert(testResult);
+    f_assert(val[0].get_str() == "\xe2\x86\x91");
+    // Escaped Supplementary Plane character U+1d161
+    testResult = val.read("[\"\\ud834\\udd61\"]");
+    f_assert(testResult);
+    f_assert(val[0].get_str() == "\xf0\x9d\x85\xa1");
+}
+
 int main (int argc, char *argv[])
 {
     for (unsigned int fidx = 0; fidx < ARRAY_SIZE(filenames); fidx++) {
         runtest_file(filenames[fidx]);
     }
 
+    unescape_unicode_test();
+
     return test_failed ? 1 : 0;
 }