Handle JSON parsing for Any.

This required a rework of the tokenizer to allow for a "replaying" tokenizer, basically in case the @type value comes after the data itself. This rework is nice in some ways (all the pushback and object depth logic in one place) but is a little fragile in terms of token push-back when using the replay tokenizer. It'll be fine for the scenario we need it for, but we should be careful...

1 files changed, 557 insertions, 489 deletions

diff --git a/csharp/src/Google.Protobuf/JsonTokenizer.cs b/csharp/src/Google.Protobuf/JsonTokenizer.cs
index 6589427a..09a6d43b 100644
--- a/csharp/src/Google.Protobuf/JsonTokenizer.cs
+++ b/csharp/src/Google.Protobuf/JsonTokenizer.cs
@@ -47,32 +47,38 @@ namespace Google.Protobuf
     /// between values. It validates the token stream as it goes - so callers can assume that the
     /// tokens it produces are appropriate. For example, it would never produce "start object, end array."
     /// </para>
+    /// <para>Implementation details: the base class handles single token push-back and </para>
     /// <para>Not thread-safe.</para>
     /// </remarks>
-    internal sealed class JsonTokenizer
+    internal abstract class JsonTokenizer
     {
-        // The set of states in which a value is valid next token.
-        private static readonly State ValueStates = State.ArrayStart | State.ArrayAfterComma | State.ObjectAfterColon | State.StartOfDocument;
-
-        private readonly Stack<ContainerType> containerStack = new Stack<ContainerType>();
-        private readonly PushBackReader reader;
         private JsonToken bufferedToken;
-        private State state;
-        private int objectDepth = 0;
 
         /// <summary>
-        /// Returns the depth of the stack, purely in objects (not collections).
-        /// Informally, this is the number of remaining unclosed '{' characters we have.
+        ///  Creates a tokenizer that reads from the given text reader.
         /// </summary>
-        internal int ObjectDepth { get { return objectDepth; } }
+        internal static JsonTokenizer FromTextReader(TextReader reader)
+        {
+            return new JsonTextTokenizer(reader);
+        }
 
-        internal JsonTokenizer(TextReader reader)
+        /// <summary>
+        /// Creates a tokenizer that first replays the given list of tokens, then continues reading
+        /// from another tokenizer. Note that if the returned tokenizer is "pushed back", that does not push back
+        /// on the continuation tokenizer, or vice versa. Care should be taken when using this method - it was
+        /// created for the sake of Any parsing.
+        /// </summary>
+        internal static JsonTokenizer FromReplayedTokens(IList<JsonToken> tokens, JsonTokenizer continuation)
         {
-            this.reader = new PushBackReader(reader);
-            state = State.StartOfDocument;
-            containerStack.Push(ContainerType.Document);
+            return new JsonReplayTokenizer(tokens, continuation);
         }
 
+        /// <summary>
+        /// Returns the depth of the stack, purely in objects (not collections).
+        /// Informally, this is the number of remaining unclosed '{' characters we have.
+        /// </summary>
+        internal int ObjectDepth { get; private set; }
+
         // TODO: Why do we allow a different token to be pushed back? It might be better to always remember the previous
         // token returned, and allow a parameterless Rewind() method (which could only be called once, just like the current PushBack).
         internal void PushBack(JsonToken token)
@@ -84,11 +90,11 @@ namespace Google.Protobuf
             bufferedToken = token;
             if (token.Type == JsonToken.TokenType.StartObject)
             {
-                objectDepth--;
+                ObjectDepth--;
             }
             else if (token.Type == JsonToken.TokenType.EndObject)
             {
-                objectDepth++;
+                ObjectDepth++;
             }
         }
 
@@ -96,574 +102,636 @@ namespace Google.Protobuf
         /// Returns the next JSON token in the stream. An EndDocument token is returned to indicate the end of the stream,
         /// after which point <c>Next()</c> should not be called again.
         /// </summary>
-        /// <remarks>
-        /// This method essentially just loops through characters skipping whitespace, validating and
-        /// changing state (e.g. from ObjectBeforeColon to ObjectAfterColon)
-        /// until it reaches something which will be a genuine token (e.g. a start object, or a value) at which point
-        /// it returns the token. Although the method is large, it would be relatively hard to break down further... most
-        /// of it is the large switch statement, which sometimes returns and sometimes doesn't.
-        /// </remarks>
+        /// <remarks>This implementation provides single-token buffering, and calls <see cref="NextImpl"/> if there is no buffered token.</remarks>
         /// <returns>The next token in the stream. This is never null.</returns>
         /// <exception cref="InvalidOperationException">This method is called after an EndDocument token has been returned</exception>
         /// <exception cref="InvalidJsonException">The input text does not comply with RFC 7159</exception>
         internal JsonToken Next()
         {
+            JsonToken tokenToReturn;
             if (bufferedToken != null)
             {
-                var ret = bufferedToken;
+                tokenToReturn = bufferedToken;
                 bufferedToken = null;
-                if (ret.Type == JsonToken.TokenType.StartObject)
-                {
-                    objectDepth++;
-                }
-                else if (ret.Type == JsonToken.TokenType.EndObject)
-                {
-                    objectDepth--;
-                }
-                return ret;
             }
-            if (state == State.ReaderExhausted)
+            else
             {
-                throw new InvalidOperationException("Next() called after end of document");
+                tokenToReturn = NextImpl();
             }
-            while (true)
+            if (tokenToReturn.Type == JsonToken.TokenType.StartObject)
             {
-                var next = reader.Read();
-                if (next == null)
-                {
-                    ValidateState(State.ExpectedEndOfDocument, "Unexpected end of document in state: ");
-                    state = State.ReaderExhausted;
-                    return JsonToken.EndDocument;
-                }
-                switch (next.Value)
-                {
-                    // Skip whitespace between tokens
-                    case ' ':
-                    case '\t':
-                    case '\r':
-                    case '\n':
-                        break;
-                    case ':':
-                        ValidateState(State.ObjectBeforeColon, "Invalid state to read a colon: ");
-                        state = State.ObjectAfterColon;
-                        break;
-                    case ',':
-                        ValidateState(State.ObjectAfterProperty | State.ArrayAfterValue, "Invalid state to read a colon: ");
-                        state = state == State.ObjectAfterProperty ? State.ObjectAfterComma : State.ArrayAfterComma;
-                        break;
-                    case '"':
-                        string stringValue = ReadString();
-                        if ((state & (State.ObjectStart | State.ObjectAfterComma)) != 0)
-                        {
-                            state = State.ObjectBeforeColon;
-                            return JsonToken.Name(stringValue);
-                        }
-                        else
-                        {
-                            ValidateAndModifyStateForValue("Invalid state to read a double quote: ");
-                            return JsonToken.Value(stringValue);
-                        }
-                    case '{':
-                        ValidateState(ValueStates, "Invalid state to read an open brace: ");
-                        state = State.ObjectStart;
-                        containerStack.Push(ContainerType.Object);
-                        objectDepth++;
-                        return JsonToken.StartObject;
-                    case '}':
-                        ValidateState(State.ObjectAfterProperty | State.ObjectStart, "Invalid state to read a close brace: ");
-                        PopContainer();
-                        objectDepth--;
-                        return JsonToken.EndObject;
-                    case '[':
-                        ValidateState(ValueStates, "Invalid state to read an open square bracket: ");
-                        state = State.ArrayStart;
-                        containerStack.Push(ContainerType.Array);
-                        return JsonToken.StartArray;
-                    case ']':
-                        ValidateState(State.ArrayAfterValue | State.ArrayStart, "Invalid state to read a close square bracket: ");
-                        PopContainer();
-                        return JsonToken.EndArray;
-                    case 'n': // Start of null
-                        ConsumeLiteral("null");
-                        ValidateAndModifyStateForValue("Invalid state to read a null literal: ");
-                        return JsonToken.Null;
-                    case 't': // Start of true
-                        ConsumeLiteral("true");
-                        ValidateAndModifyStateForValue("Invalid state to read a true literal: ");
-                        return JsonToken.True;
-                    case 'f': // Start of false
-                        ConsumeLiteral("false");
-                        ValidateAndModifyStateForValue("Invalid state to read a false literal: ");
-                        return JsonToken.False;
-                    case '-': // Start of a number
-                    case '0':
-                    case '1':
-                    case '2':
-                    case '3':
-                    case '4':
-                    case '5':
-                    case '6':
-                    case '7':
-                    case '8':
-                    case '9':
-                        double number = ReadNumber(next.Value);
-                        ValidateAndModifyStateForValue("Invalid state to read a number token: ");
-                        return JsonToken.Value(number);
-                    default:
-                        throw new InvalidJsonException("Invalid first character of token: " + next.Value);
-                }
+                ObjectDepth++;
             }
-        }
-
-        private void ValidateState(State validStates, string errorPrefix)
-        {
-            if ((validStates & state) == 0)
+            else if (tokenToReturn.Type == JsonToken.TokenType.EndObject)
             {
-                throw reader.CreateException(errorPrefix + state);
+                ObjectDepth--;
             }
+            return tokenToReturn;
         }
 
         /// <summary>
-        /// Reads a string token. It is assumed that the opening " has already been read.
+        /// Returns the next JSON token in the stream, when requested by the base class. (The <see cref="Next"/> method delegates
+        /// to this if it doesn't have a buffered token.)
         /// </summary>
-        private string ReadString()
-        {
-            var value = new StringBuilder();
-            bool haveHighSurrogate = false;
-            while (true)
-            {
-                char c = reader.ReadOrFail("Unexpected end of text while reading string");
-                if (c < ' ')
-                {
-                    throw reader.CreateException(string.Format(CultureInfo.InvariantCulture, "Invalid character in string literal: U+{0:x4}", (int) c));
-                }
-                if (c == '"')
-                {
-                    if (haveHighSurrogate)
-                    {
-                        throw reader.CreateException("Invalid use of surrogate pair code units");
-                    }
-                    return value.ToString();
-                }
-                if (c == '\\')
-                {
-                    c = ReadEscapedCharacter();
-                }
-                // TODO: Consider only allowing surrogate pairs that are either both escaped,
-                // or both not escaped. It would be a very odd text stream that contained a "lone" high surrogate
-                // followed by an escaped low surrogate or vice versa... and that couldn't even be represented in UTF-8.
-                if (haveHighSurrogate != char.IsLowSurrogate(c))
-                {
-                    throw reader.CreateException("Invalid use of surrogate pair code units");
-                }
-                haveHighSurrogate = char.IsHighSurrogate(c);
-                value.Append(c);
-            }
-        }
+        /// <exception cref="InvalidOperationException">This method is called after an EndDocument token has been returned</exception>
+        /// <exception cref="InvalidJsonException">The input text does not comply with RFC 7159</exception>
+        protected abstract JsonToken NextImpl();
 
         /// <summary>
-        /// Reads an escaped character. It is assumed that the leading backslash has already been read.
+        /// Tokenizer which first exhausts a list of tokens, then consults another tokenizer.
         /// </summary>
-        private char ReadEscapedCharacter()
+        private class JsonReplayTokenizer : JsonTokenizer
         {
-            char c = reader.ReadOrFail("Unexpected end of text while reading character escape sequence");
-            switch (c)
+            private readonly IList<JsonToken> tokens;
+            private readonly JsonTokenizer nextTokenizer;
+            private int nextTokenIndex;
+
+            internal JsonReplayTokenizer(IList<JsonToken> tokens, JsonTokenizer nextTokenizer)
             {
-                case 'n':
-                    return '\n';
-                case '\\':
-                    return '\\';
-                case 'b':
-                    return '\b';
-                case 'f':
-                    return '\f';
-                case 'r':
-                    return '\r';
-                case 't':
-                    return '\t';
-                case '"':
-                    return '"';
-                case '/':
-                    return '/';
-                case 'u':
-                    return ReadUnicodeEscape();
-                default:
-                    throw reader.CreateException(string.Format(CultureInfo.InvariantCulture, "Invalid character in character escape sequence: U+{0:x4}", (int) c));
+                this.tokens = tokens;
+                this.nextTokenizer = nextTokenizer;
             }
-        }
 
-        /// <summary>
-        /// Reads an escaped Unicode 4-nybble hex sequence. It is assumed that the leading \u has already been read.
-        /// </summary>
-        private char ReadUnicodeEscape()
-        {
-            int result = 0;
-            for (int i = 0; i < 4; i++)
+            // FIXME: Object depth not maintained...
+            protected override JsonToken NextImpl()
             {
-                char c = reader.ReadOrFail("Unexpected end of text while reading Unicode escape sequence");
-                int nybble;
-                if (c >= '0' && c <= '9')
+                if (nextTokenIndex >= tokens.Count)
                 {
-                    nybble = c - '0';
+                    return nextTokenizer.Next();
                 }
-                else if (c >= 'a' && c <= 'f')
-                {
-                    nybble = c - 'a' + 10;
-                }
-                else if (c >= 'A' && c <= 'F')
-                {
-                    nybble = c - 'A' + 10;
-                }
-                else
-                {
-                    throw reader.CreateException(string.Format(CultureInfo.InvariantCulture, "Invalid character in character escape sequence: U+{0:x4}", (int) c));
-                }
-                result = (result << 4) + nybble;
+                return tokens[nextTokenIndex++];
             }
-            return (char) result;
         }
 
         /// <summary>
-        /// Consumes a text-only literal, throwing an exception if the read text doesn't match it.
-        /// It is assumed that the first letter of the literal has already been read.
+        /// Tokenizer which does all the *real* work of parsing JSON.
         /// </summary>
-        private void ConsumeLiteral(string text)
+        private sealed class JsonTextTokenizer : JsonTokenizer
         {
-            for (int i = 1; i < text.Length; i++)
+            // The set of states in which a value is valid next token.
+            private static readonly State ValueStates = State.ArrayStart | State.ArrayAfterComma | State.ObjectAfterColon | State.StartOfDocument;
+
+            private readonly Stack<ContainerType> containerStack = new Stack<ContainerType>();
+            private readonly PushBackReader reader;
+            private State state;
+
+            internal JsonTextTokenizer(TextReader reader)
             {
-                char? next = reader.Read();
-                if (next == null)
+                this.reader = new PushBackReader(reader);
+                state = State.StartOfDocument;
+                containerStack.Push(ContainerType.Document);
+            }
+
+            /// <remarks>
+            /// This method essentially just loops through characters skipping whitespace, validating and
+            /// changing state (e.g. from ObjectBeforeColon to ObjectAfterColon)
+            /// until it reaches something which will be a genuine token (e.g. a start object, or a value) at which point
+            /// it returns the token. Although the method is large, it would be relatively hard to break down further... most
+            /// of it is the large switch statement, which sometimes returns and sometimes doesn't.
+            /// </remarks>
+            protected override JsonToken NextImpl()
+            {
+                if (state == State.ReaderExhausted)
                 {
-                    throw reader.CreateException("Unexpected end of text while reading literal token " + text);
+                    throw new InvalidOperationException("Next() called after end of document");
                 }
-                if (next.Value != text[i])
+                while (true)
                 {
-                    throw reader.CreateException("Unexpected character while reading literal token " + text);
+                    var next = reader.Read();
+                    if (next == null)
+                    {
+                        ValidateState(State.ExpectedEndOfDocument, "Unexpected end of document in state: ");
+                        state = State.ReaderExhausted;
+                        return JsonToken.EndDocument;
+                    }
+                    switch (next.Value)
+                    {
+                        // Skip whitespace between tokens
+                        case ' ':
+                        case '\t':
+                        case '\r':
+                        case '\n':
+                            break;
+                        case ':':
+                            ValidateState(State.ObjectBeforeColon, "Invalid state to read a colon: ");
+                            state = State.ObjectAfterColon;
+                            break;
+                        case ',':
+                            ValidateState(State.ObjectAfterProperty | State.ArrayAfterValue, "Invalid state to read a colon: ");
+                            state = state == State.ObjectAfterProperty ? State.ObjectAfterComma : State.ArrayAfterComma;
+                            break;
+                        case '"':
+                            string stringValue = ReadString();
+                            if ((state & (State.ObjectStart | State.ObjectAfterComma)) != 0)
+                            {
+                                state = State.ObjectBeforeColon;
+                                return JsonToken.Name(stringValue);
+                            }
+                            else
+                            {
+                                ValidateAndModifyStateForValue("Invalid state to read a double quote: ");
+                                return JsonToken.Value(stringValue);
+                            }
+                        case '{':
+                            ValidateState(ValueStates, "Invalid state to read an open brace: ");
+                            state = State.ObjectStart;
+                            containerStack.Push(ContainerType.Object);
+                            return JsonToken.StartObject;
+                        case '}':
+                            ValidateState(State.ObjectAfterProperty | State.ObjectStart, "Invalid state to read a close brace: ");
+                            PopContainer();
+                            return JsonToken.EndObject;
+                        case '[':
+                            ValidateState(ValueStates, "Invalid state to read an open square bracket: ");
+                            state = State.ArrayStart;
+                            containerStack.Push(ContainerType.Array);
+                            return JsonToken.StartArray;
+                        case ']':
+                            ValidateState(State.ArrayAfterValue | State.ArrayStart, "Invalid state to read a close square bracket: ");
+                            PopContainer();
+                            return JsonToken.EndArray;
+                        case 'n': // Start of null
+                            ConsumeLiteral("null");
+                            ValidateAndModifyStateForValue("Invalid state to read a null literal: ");
+                            return JsonToken.Null;
+                        case 't': // Start of true
+                            ConsumeLiteral("true");
+                            ValidateAndModifyStateForValue("Invalid state to read a true literal: ");
+                            return JsonToken.True;
+                        case 'f': // Start of false
+                            ConsumeLiteral("false");
+                            ValidateAndModifyStateForValue("Invalid state to read a false literal: ");
+                            return JsonToken.False;
+                        case '-': // Start of a number
+                        case '0':
+                        case '1':
+                        case '2':
+                        case '3':
+                        case '4':
+                        case '5':
+                        case '6':
+                        case '7':
+                        case '8':
+                        case '9':
+                            double number = ReadNumber(next.Value);
+                            ValidateAndModifyStateForValue("Invalid state to read a number token: ");
+                            return JsonToken.Value(number);
+                        default:
+                            throw new InvalidJsonException("Invalid first character of token: " + next.Value);
+                    }
                 }
             }
-        }
 
-        private double ReadNumber(char initialCharacter)
-        {
-            StringBuilder builder = new StringBuilder();
-            if (initialCharacter == '-')
-            {
-                builder.Append("-");
-            }
-            else
-            {
-                reader.PushBack(initialCharacter);
-            }
-            // Each method returns the character it read that doesn't belong in that part,
-            // so we know what to do next, including pushing the character back at the end.
-            // null is returned for "end of text".
-            char? next = ReadInt(builder);
-            if (next == '.')
+            private void ValidateState(State validStates, string errorPrefix)
             {
-                next = ReadFrac(builder);
-            }
-            if (next == 'e' || next == 'E')
-            {
-                next = ReadExp(builder);
-            }
-            // If we read a character which wasn't part of the number, push it back so we can read it again
-            // to parse the next token.
-            if (next != null)
-            {
-                reader.PushBack(next.Value);
+                if ((validStates & state) == 0)
+                {
+                    throw reader.CreateException(errorPrefix + state);
+                }
             }
 
-            // TODO: What exception should we throw if the value can't be represented as a double?
-            try
-            {
-                return double.Parse(builder.ToString(),
-                    NumberStyles.AllowLeadingSign | NumberStyles.AllowDecimalPoint | NumberStyles.AllowExponent,
-                    CultureInfo.InvariantCulture);
-            }
-            catch (OverflowException)
+            /// <summary>
+            /// Reads a string token. It is assumed that the opening " has already been read.
+            /// </summary>
+            private string ReadString()
             {
-                throw reader.CreateException("Numeric value out of range: " + builder);
+                var value = new StringBuilder();
+                bool haveHighSurrogate = false;
+                while (true)
+                {
+                    char c = reader.ReadOrFail("Unexpected end of text while reading string");
+                    if (c < ' ')
+                    {
+                        throw reader.CreateException(string.Format(CultureInfo.InvariantCulture, "Invalid character in string literal: U+{0:x4}", (int) c));
+                    }
+                    if (c == '"')
+                    {
+                        if (haveHighSurrogate)
+                        {
+                            throw reader.CreateException("Invalid use of surrogate pair code units");
+                        }
+                        return value.ToString();
+                    }
+                    if (c == '\\')
+                    {
+                        c = ReadEscapedCharacter();
+                    }
+                    // TODO: Consider only allowing surrogate pairs that are either both escaped,
+                    // or both not escaped. It would be a very odd text stream that contained a "lone" high surrogate
+                    // followed by an escaped low surrogate or vice versa... and that couldn't even be represented in UTF-8.
+                    if (haveHighSurrogate != char.IsLowSurrogate(c))
+                    {
+                        throw reader.CreateException("Invalid use of surrogate pair code units");
+                    }
+                    haveHighSurrogate = char.IsHighSurrogate(c);
+                    value.Append(c);
+                }
             }
-        }
 
-        private char? ReadInt(StringBuilder builder)
-        {
-            char first = reader.ReadOrFail("Invalid numeric literal");
-            if (first < '0' || first > '9')
-            {
-                throw reader.CreateException("Invalid numeric literal");
-            }
-            builder.Append(first);
-            int digitCount;
-            char? next = ConsumeDigits(builder, out digitCount);
-            if (first == '0' && digitCount != 0)
+            /// <summary>
+            /// Reads an escaped character. It is assumed that the leading backslash has already been read.
+            /// </summary>
+            private char ReadEscapedCharacter()
             {
-                throw reader.CreateException("Invalid numeric literal: leading 0 for non-zero value.");
+                char c = reader.ReadOrFail("Unexpected end of text while reading character escape sequence");
+                switch (c)
+                {
+                    case 'n':
+                        return '\n';
+                    case '\\':
+                        return '\\';
+                    case 'b':
+                        return '\b';
+                    case 'f':
+                        return '\f';
+                    case 'r':
+                        return '\r';
+                    case 't':
+                        return '\t';
+                    case '"':
+                        return '"';
+                    case '/':
+                        return '/';
+                    case 'u':
+                        return ReadUnicodeEscape();
+                    default:
+                        throw reader.CreateException(string.Format(CultureInfo.InvariantCulture, "Invalid character in character escape sequence: U+{0:x4}", (int) c));
+                }
             }
-            return next;
-        }
 
-        private char? ReadFrac(StringBuilder builder)
-        {
-            builder.Append('.'); // Already consumed this
-            int digitCount;
-            char? next = ConsumeDigits(builder, out digitCount);
-            if (digitCount == 0)
+            /// <summary>
+            /// Reads an escaped Unicode 4-nybble hex sequence. It is assumed that the leading \u has already been read.
+            /// </summary>
+            private char ReadUnicodeEscape()
             {
-                throw reader.CreateException("Invalid numeric literal: fraction with no trailing digits");
+                int result = 0;
+                for (int i = 0; i < 4; i++)
+                {
+                    char c = reader.ReadOrFail("Unexpected end of text while reading Unicode escape sequence");
+                    int nybble;
+                    if (c >= '0' && c <= '9')
+                    {
+                        nybble = c - '0';
+                    }
+                    else if (c >= 'a' && c <= 'f')
+                    {
+                        nybble = c - 'a' + 10;
+                    }
+                    else if (c >= 'A' && c <= 'F')
+                    {
+                        nybble = c - 'A' + 10;
+                    }
+                    else
+                    {
+                        throw reader.CreateException(string.Format(CultureInfo.InvariantCulture, "Invalid character in character escape sequence: U+{0:x4}", (int) c));
+                    }
+                    result = (result << 4) + nybble;
+                }
+                return (char) result;
             }
-            return next;
-        }
 
-        private char? ReadExp(StringBuilder builder)
-        {
-            builder.Append('E'); // Already consumed this (or 'e')
-            char? next = reader.Read();
-            if (next == null)
+            /// <summary>
+            /// Consumes a text-only literal, throwing an exception if the read text doesn't match it.
+            /// It is assumed that the first letter of the literal has already been read.
+            /// </summary>
+            private void ConsumeLiteral(string text)
             {
-                throw reader.CreateException("Invalid numeric literal: exponent with no trailing digits");
+                for (int i = 1; i < text.Length; i++)
+                {
+                    char? next = reader.Read();
+                    if (next == null)
+                    {
+                        throw reader.CreateException("Unexpected end of text while reading literal token " + text);
+                    }
+                    if (next.Value != text[i])
+                    {
+                        throw reader.CreateException("Unexpected character while reading literal token " + text);
+                    }
+                }
             }
-            if (next == '-' || next == '+')
+
+            private double ReadNumber(char initialCharacter)
             {
-                builder.Append(next.Value);
+                StringBuilder builder = new StringBuilder();
+                if (initialCharacter == '-')
+                {
+                    builder.Append("-");
+                }
+                else
+                {
+                    reader.PushBack(initialCharacter);
+                }
+                // Each method returns the character it read that doesn't belong in that part,
+                // so we know what to do next, including pushing the character back at the end.
+                // null is returned for "end of text".
+                char? next = ReadInt(builder);
+                if (next == '.')
+                {
+                    next = ReadFrac(builder);
+                }
+                if (next == 'e' || next == 'E')
+                {
+                    next = ReadExp(builder);
+                }
+                // If we read a character which wasn't part of the number, push it back so we can read it again
+                // to parse the next token.
+                if (next != null)
+                {
+                    reader.PushBack(next.Value);
+                }
+
+                // TODO: What exception should we throw if the value can't be represented as a double?
+                try
+                {
+                    return double.Parse(builder.ToString(),
+                        NumberStyles.AllowLeadingSign | NumberStyles.AllowDecimalPoint | NumberStyles.AllowExponent,
+                        CultureInfo.InvariantCulture);
+                }
+                catch (OverflowException)
+                {
+                    throw reader.CreateException("Numeric value out of range: " + builder);
+                }
             }
-            else
+
+            private char? ReadInt(StringBuilder builder)
             {
-                reader.PushBack(next.Value);
+                char first = reader.ReadOrFail("Invalid numeric literal");
+                if (first < '0' || first > '9')
+                {
+                    throw reader.CreateException("Invalid numeric literal");
+                }
+                builder.Append(first);
+                int digitCount;
+                char? next = ConsumeDigits(builder, out digitCount);
+                if (first == '0' && digitCount != 0)
+                {
+                    throw reader.CreateException("Invalid numeric literal: leading 0 for non-zero value.");
+                }
+                return next;
             }
-            int digitCount;
-            next = ConsumeDigits(builder, out digitCount);
-            if (digitCount == 0)
+
+            private char? ReadFrac(StringBuilder builder)
             {
-                throw reader.CreateException("Invalid numeric literal: exponent without value");
+                builder.Append('.'); // Already consumed this
+                int digitCount;
+                char? next = ConsumeDigits(builder, out digitCount);
+                if (digitCount == 0)
+                {
+                    throw reader.CreateException("Invalid numeric literal: fraction with no trailing digits");
+                }
+                return next;
             }
-            return next;
-        }
 
-        private char? ConsumeDigits(StringBuilder builder, out int count)
-        {
-            count = 0;
-            while (true)
+            private char? ReadExp(StringBuilder builder)
             {
+                builder.Append('E'); // Already consumed this (or 'e')
                 char? next = reader.Read();
-                if (next == null || next.Value < '0' || next.Value > '9')
+                if (next == null)
+                {
+                    throw reader.CreateException("Invalid numeric literal: exponent with no trailing digits");
+                }
+                if (next == '-' || next == '+')
                 {
-                    return next;
+                    builder.Append(next.Value);
                 }
-                count++;
-                builder.Append(next.Value);
+                else
+                {
+                    reader.PushBack(next.Value);
+                }
+                int digitCount;
+                next = ConsumeDigits(builder, out digitCount);
+                if (digitCount == 0)
+                {
+                    throw reader.CreateException("Invalid numeric literal: exponent without value");
+                }
+                return next;
             }
-        }
 
-        /// <summary>
-        /// Validates that we're in a valid state to read a value (using the given error prefix if necessary)
-        /// and changes the state to the appropriate one, e.g. ObjectAfterColon to ObjectAfterProperty.
-        /// </summary>
-        private void ValidateAndModifyStateForValue(string errorPrefix)
-        {
-            ValidateState(ValueStates, errorPrefix);
-            switch (state)
+            private char? ConsumeDigits(StringBuilder builder, out int count)
             {
-                case State.StartOfDocument:
-                    state = State.ExpectedEndOfDocument;
-                    return;
-                case State.ObjectAfterColon:
-                    state = State.ObjectAfterProperty;
-                    return;
-                case State.ArrayStart:
-                case State.ArrayAfterComma:
-                    state = State.ArrayAfterValue;
-                    return;
-                default:
-                    throw new InvalidOperationException("ValidateAndModifyStateForValue does not handle all value states (and should)");
+                count = 0;
+                while (true)
+                {
+                    char? next = reader.Read();
+                    if (next == null || next.Value < '0' || next.Value > '9')
+                    {
+                        return next;
+                    }
+                    count++;
+                    builder.Append(next.Value);
+                }
             }
-        }
 
-        /// <summary>
-        /// Pops the top-most container, and sets the state to the appropriate one for the end of a value
-        /// in the parent container.
-        /// </summary>
-        private void PopContainer()
-        {
-            containerStack.Pop();
-            var parent = containerStack.Peek();
-            switch (parent)
+            /// <summary>
+            /// Validates that we're in a valid state to read a value (using the given error prefix if necessary)
+            /// and changes the state to the appropriate one, e.g. ObjectAfterColon to ObjectAfterProperty.
+            /// </summary>
+            private void ValidateAndModifyStateForValue(string errorPrefix)
             {
-                case ContainerType.Object:
-                    state = State.ObjectAfterProperty;
-                    break;
-                case ContainerType.Array:
-                    state = State.ArrayAfterValue;
-                    break;
-                case ContainerType.Document:
-                    state = State.ExpectedEndOfDocument;
-                    break;
-                default:
-                    throw new InvalidOperationException("Unexpected container type: " + parent);
+                ValidateState(ValueStates, errorPrefix);
+                switch (state)
+                {
+                    case State.StartOfDocument:
+                        state = State.ExpectedEndOfDocument;
+                        return;
+                    case State.ObjectAfterColon:
+                        state = State.ObjectAfterProperty;
+                        return;
+                    case State.ArrayStart:
+                    case State.ArrayAfterComma:
+                        state = State.ArrayAfterValue;
+                        return;
+                    default:
+                        throw new InvalidOperationException("ValidateAndModifyStateForValue does not handle all value states (and should)");
+                }
             }
-        }
 
-        private enum ContainerType
-        {
-            Document, Object, Array
-        }
-
-        /// <summary>
-        /// Possible states of the tokenizer.
-        /// </summary>
-        /// <remarks>
-        /// <para>This is a flags enum purely so we can simply and efficiently represent a set of valid states
-        /// for checking.</para>
-        /// <para>
-        /// Each is documented with an example,
-        /// where ^ represents the current position within the text stream. The examples all use string values,
-        /// but could be any value, including nested objects/arrays.
-        /// The complete state of the tokenizer also includes a stack to indicate the contexts (arrays/objects).
-        /// Any additional notional state of "AfterValue" indicates that a value has been completed, at which 
-        /// point there's an immediate transition to ExpectedEndOfDocument,  ObjectAfterProperty or ArrayAfterValue.
-        /// </para>
-        /// <para>
-        /// These states were derived manually by reading RFC 7159 carefully.
-        /// </para>
-        /// </remarks>
-        [Flags]
-        private enum State
-        {
-            /// <summary>
-            /// ^ { "foo": "bar" }
-            /// Before the value in a document. Next states: ObjectStart, ArrayStart, "AfterValue"
-            /// </summary>
-            StartOfDocument = 1 << 0,
-            /// <summary>
-            /// { "foo": "bar" } ^
-            /// After the value in a document. Next states: ReaderExhausted
-            /// </summary>
-            ExpectedEndOfDocument = 1 << 1,
-            /// <summary>
-            /// { "foo": "bar" } ^ (and already read to the end of the reader)
-            /// Terminal state.
-            /// </summary>
-            ReaderExhausted = 1 << 2,
-            /// <summary>
-            /// { ^ "foo": "bar" }
-            /// Before the *first* property in an object.
-            /// Next states:
-            /// "AfterValue" (empty object)
-            /// ObjectBeforeColon (read a name)
-            /// </summary>
-            ObjectStart = 1 << 3,
-            /// <summary>
-            /// { "foo" ^ : "bar", "x": "y" }
-            /// Next state: ObjectAfterColon
-            /// </summary>
-            ObjectBeforeColon = 1 << 4,
-            /// <summary>
-            /// { "foo" : ^ "bar", "x": "y" }
-            /// Before any property other than the first in an object.
-            /// (Equivalently: after any property in an object) 
-            /// Next states:
-            /// "AfterValue" (value is simple)
-            /// ObjectStart (value is object)
-            /// ArrayStart (value is array)
-            /// </summary>
-            ObjectAfterColon = 1 << 5,
-            /// <summary>
-            /// { "foo" : "bar" ^ , "x" : "y" }
-            /// At the end of a property, so expecting either a comma or end-of-object
-            /// Next states: ObjectAfterComma or "AfterValue"
-            /// </summary>
-            ObjectAfterProperty = 1 << 6,
-            /// <summary>
-            /// { "foo":"bar", ^ "x":"y" }
-            /// Read the comma after the previous property, so expecting another property.
-            /// This is like ObjectStart, but closing brace isn't valid here
-            /// Next state: ObjectBeforeColon.
-            /// </summary>
-            ObjectAfterComma = 1 << 7,
-            /// <summary>
-            /// [ ^ "foo", "bar" ]
-            /// Before the *first* value in an array.
-            /// Next states:
-            /// "AfterValue" (read a value)
-            /// "AfterValue" (end of array; will pop stack)
-            /// </summary>
-            ArrayStart = 1 << 8,
-            /// <summary>
-            /// [ "foo" ^ , "bar" ]
-            /// After any value in an array, so expecting either a comma or end-of-array
-            /// Next states: ArrayAfterComma or "AfterValue"
-            /// </summary>
-            ArrayAfterValue = 1 << 9,
             /// <summary>
-            /// [ "foo", ^ "bar" ]
-            /// After a comma in an array, so there *must* be another value (simple or complex).
-            /// Next states: "AfterValue" (simple value), StartObject, StartArray
+            /// Pops the top-most container, and sets the state to the appropriate one for the end of a value
+            /// in the parent container.
             /// </summary>
-            ArrayAfterComma = 1 << 10
-        }
-
-        /// <summary>
-        /// Wrapper around a text reader allowing small amounts of buffering and location handling.
-        /// </summary>
-        private class PushBackReader
-        {
-            // TODO: Add locations for errors etc.
-
-            private readonly TextReader reader;
+            private void PopContainer()
+            {
+                containerStack.Pop();
+                var parent = containerStack.Peek();
+                switch (parent)
+                {
+                    case ContainerType.Object:
+                        state = State.ObjectAfterProperty;
+                        break;
+                    case ContainerType.Array:
+                        state = State.ArrayAfterValue;
+                        break;
+                    case ContainerType.Document:
+                        state = State.ExpectedEndOfDocument;
+                        break;
+                    default:
+                        throw new InvalidOperationException("Unexpected container type: " + parent);
+                }
+            }
 
-            internal PushBackReader(TextReader reader)
+            private enum ContainerType
             {
-                // TODO: Wrap the reader in a BufferedReader?
-                this.reader = reader;
+                Document, Object, Array
             }
 
             /// <summary>
-            /// The buffered next character, if we have one.
+            /// Possible states of the tokenizer.
             /// </summary>
-            private char? nextChar;
+            /// <remarks>
+            /// <para>This is a flags enum purely so we can simply and efficiently represent a set of valid states
+            /// for checking.</para>
+            /// <para>
+            /// Each is documented with an example,
+            /// where ^ represents the current position within the text stream. The examples all use string values,
+            /// but could be any value, including nested objects/arrays.
+            /// The complete state of the tokenizer also includes a stack to indicate the contexts (arrays/objects).
+            /// Any additional notional state of "AfterValue" indicates that a value has been completed, at which 
+            /// point there's an immediate transition to ExpectedEndOfDocument,  ObjectAfterProperty or ArrayAfterValue.
+            /// </para>
+            /// <para>
+            /// These states were derived manually by reading RFC 7159 carefully.
+            /// </para>
+            /// </remarks>
+            [Flags]
+            private enum State
+            {
+                /// <summary>
+                /// ^ { "foo": "bar" }
+                /// Before the value in a document. Next states: ObjectStart, ArrayStart, "AfterValue"
+                /// </summary>
+                StartOfDocument = 1 << 0,
+                /// <summary>
+                /// { "foo": "bar" } ^
+                /// After the value in a document. Next states: ReaderExhausted
+                /// </summary>
+                ExpectedEndOfDocument = 1 << 1,
+                /// <summary>
+                /// { "foo": "bar" } ^ (and already read to the end of the reader)
+                /// Terminal state.
+                /// </summary>
+                ReaderExhausted = 1 << 2,
+                /// <summary>
+                /// { ^ "foo": "bar" }
+                /// Before the *first* property in an object.
+                /// Next states:
+                /// "AfterValue" (empty object)
+                /// ObjectBeforeColon (read a name)
+                /// </summary>
+                ObjectStart = 1 << 3,
+                /// <summary>
+                /// { "foo" ^ : "bar", "x": "y" }
+                /// Next state: ObjectAfterColon
+                /// </summary>
+                ObjectBeforeColon = 1 << 4,
+                /// <summary>
+                /// { "foo" : ^ "bar", "x": "y" }
+                /// Before any property other than the first in an object.
+                /// (Equivalently: after any property in an object) 
+                /// Next states:
+                /// "AfterValue" (value is simple)
+                /// ObjectStart (value is object)
+                /// ArrayStart (value is array)
+                /// </summary>
+                ObjectAfterColon = 1 << 5,
+                /// <summary>
+                /// { "foo" : "bar" ^ , "x" : "y" }
+                /// At the end of a property, so expecting either a comma or end-of-object
+                /// Next states: ObjectAfterComma or "AfterValue"
+                /// </summary>
+                ObjectAfterProperty = 1 << 6,
+                /// <summary>
+                /// { "foo":"bar", ^ "x":"y" }
+                /// Read the comma after the previous property, so expecting another property.
+                /// This is like ObjectStart, but closing brace isn't valid here
+                /// Next state: ObjectBeforeColon.
+                /// </summary>
+                ObjectAfterComma = 1 << 7,
+                /// <summary>
+                /// [ ^ "foo", "bar" ]
+                /// Before the *first* value in an array.
+                /// Next states:
+                /// "AfterValue" (read a value)
+                /// "AfterValue" (end of array; will pop stack)
+                /// </summary>
+                ArrayStart = 1 << 8,
+                /// <summary>
+                /// [ "foo" ^ , "bar" ]
+                /// After any value in an array, so expecting either a comma or end-of-array
+                /// Next states: ArrayAfterComma or "AfterValue"
+                /// </summary>
+                ArrayAfterValue = 1 << 9,
+                /// <summary>
+                /// [ "foo", ^ "bar" ]
+                /// After a comma in an array, so there *must* be another value (simple or complex).
+                /// Next states: "AfterValue" (simple value), StartObject, StartArray
+                /// </summary>
+                ArrayAfterComma = 1 << 10
+            }
 
             /// <summary>
-            /// Returns the next character in the stream, or null if we have reached the end.
+            /// Wrapper around a text reader allowing small amounts of buffering and location handling.
             /// </summary>
-            /// <returns></returns>
-            internal char? Read()
+            private class PushBackReader
             {
-                if (nextChar != null)
+                // TODO: Add locations for errors etc.
+
+                private readonly TextReader reader;
+
+                internal PushBackReader(TextReader reader)
                 {
-                    char? tmp = nextChar;
-                    nextChar = null;
-                    return tmp;
+                    // TODO: Wrap the reader in a BufferedReader?
+                    this.reader = reader;
                 }
-                int next = reader.Read();
-                return next == -1 ? null : (char?) next;
-            }
 
-            internal char ReadOrFail(string messageOnFailure)
-            {
-                char? next = Read();
-                if (next == null)
+                /// <summary>
+                /// The buffered next character, if we have one.
+                /// </summary>
+                private char? nextChar;
+
+                /// <summary>
+                /// Returns the next character in the stream, or null if we have reached the end.
+                /// </summary>
+                /// <returns></returns>
+                internal char? Read()
                 {
-                    throw CreateException(messageOnFailure);
+                    if (nextChar != null)
+                    {
+                        char? tmp = nextChar;
+                        nextChar = null;
+                        return tmp;
+                    }
+                    int next = reader.Read();
+                    return next == -1 ? null : (char?) next;
                 }
-                return next.Value;
-            }
 
-            internal void PushBack(char c)
-            {
-                if (nextChar != null)
+                internal char ReadOrFail(string messageOnFailure)
                 {
-                    throw new InvalidOperationException("Cannot push back when already buffering a character");
+                    char? next = Read();
+                    if (next == null)
+                    {
+                        throw CreateException(messageOnFailure);
+                    }
+                    return next.Value;
                 }
-                nextChar = c;
-            }
 
-            /// <summary>
-            /// Creates a new exception appropriate for the current state of the reader.
-            /// </summary>
-            internal InvalidJsonException CreateException(string message)
-            {
-                // TODO: Keep track of and use the location.
-                return new InvalidJsonException(message);
+                internal void PushBack(char c)
+                {
+                    if (nextChar != null)
+                    {
+                        throw new InvalidOperationException("Cannot push back when already buffering a character");
+                    }
+                    nextChar = c;
+                }
+
+                /// <summary>
+                /// Creates a new exception appropriate for the current state of the reader.
+                /// </summary>
+                internal InvalidJsonException CreateException(string message)
+                {
+                    // TODO: Keep track of and use the location.
+                    return new InvalidJsonException(message);
+                }
             }
         }
     }