Manuel Klimek | 76f1301 | 2011-12-16 13:09:10 +0000 | [diff] [blame] | 1 | //===--- JsonParser.cpp - Simple JSON parser ------------------------------===// |
| 2 | // |
| 3 | // The LLVM Compiler Infrastructure |
| 4 | // |
| 5 | // This file is distributed under the University of Illinois Open Source |
| 6 | // License. See LICENSE.TXT for details. |
| 7 | // |
| 8 | //===----------------------------------------------------------------------===// |
| 9 | // |
| 10 | // This file implements a JSON parser. |
| 11 | // |
| 12 | //===----------------------------------------------------------------------===// |
| 13 | |
| 14 | #include "llvm/Support/JSONParser.h" |
| 15 | |
| 16 | #include "llvm/ADT/Twine.h" |
| 17 | #include "llvm/Support/Casting.h" |
| 18 | |
| 19 | namespace llvm { |
| 20 | |
| 21 | JSONParser::JSONParser(StringRef Input) |
| 22 | : Input(Input), Position(Input.begin()) {} |
| 23 | |
| 24 | JSONValue *JSONParser::parseRoot() { |
| 25 | if (Position != Input.begin()) |
| 26 | report_fatal_error("Cannot resuse JSONParser."); |
| 27 | if (isWhitespace()) |
| 28 | nextNonWhitespace(); |
| 29 | if (errorIfAtEndOfFile("'[' or '{' at start of JSON text")) |
| 30 | return 0; |
| 31 | switch (*Position) { |
| 32 | case '[': |
| 33 | return new (ValueAllocator.Allocate<JSONArray>(1)) JSONArray(this); |
| 34 | case '{': |
| 35 | return new (ValueAllocator.Allocate<JSONObject>(1)) JSONObject(this); |
| 36 | default: |
| 37 | setExpectedError("'[' or '{' at start of JSON text", *Position); |
| 38 | return 0; |
| 39 | } |
| 40 | } |
| 41 | |
| 42 | bool JSONParser::validate() { |
| 43 | return parseRoot()->skip(); |
| 44 | } |
| 45 | |
| 46 | // Sets the current error to: |
| 47 | // "Error while parsing JSON: expected <Expected>, but found <Found>". |
| 48 | void JSONParser::setExpectedError(StringRef Expected, StringRef Found) { |
| 49 | ErrorMessage = ("Error while parsing JSON: expected " + |
| 50 | Expected + ", but found " + Found + ".").str(); |
| 51 | } |
| 52 | |
| 53 | // Sets the current error to: |
| 54 | // "Error while parsing JSON: expected <Expected>, but found <Found>". |
| 55 | void JSONParser::setExpectedError(StringRef Expected, char Found) { |
| 56 | setExpectedError(Expected, StringRef(&Found, 1)); |
| 57 | } |
| 58 | |
| 59 | // If there is no character available, returns true and sets the current error |
| 60 | // to: "Error while parsing JSON: expected <Expected>, but found EOF.". |
| 61 | bool JSONParser::errorIfAtEndOfFile(StringRef Expected) { |
| 62 | if (Position == Input.end()) { |
| 63 | setExpectedError(Expected, "EOF"); |
| 64 | return true; |
| 65 | } |
| 66 | return false; |
| 67 | } |
| 68 | |
| 69 | // Sets the current error if the current character is not C to: |
| 70 | // "Error while parsing JSON: expected 'C', but got <current character>". |
| 71 | bool JSONParser::errorIfNotAt(char C, StringRef Message) { |
| 72 | if (Position == Input.end() || *Position != C) { |
| 73 | std::string Expected = |
| 74 | ("'" + StringRef(&C, 1) + "' " + Message).str(); |
| 75 | if (Position == Input.end()) |
| 76 | setExpectedError(Expected, "EOF"); |
| 77 | else |
| 78 | setExpectedError(Expected, *Position); |
| 79 | return true; |
| 80 | } |
| 81 | return false; |
| 82 | } |
| 83 | |
| 84 | // Forbidding inlining improves performance by roughly 20%. |
| 85 | // FIXME: Remove once llvm optimizes this to the faster version without hints. |
| 86 | LLVM_ATTRIBUTE_NOINLINE static bool |
| 87 | wasEscaped(StringRef::iterator First, StringRef::iterator Position); |
| 88 | |
| 89 | // Returns whether a character at 'Position' was escaped with a leading '\'. |
| 90 | // 'First' specifies the position of the first character in the string. |
| 91 | static bool wasEscaped(StringRef::iterator First, |
| 92 | StringRef::iterator Position) { |
| 93 | assert(Position - 1 >= First); |
| 94 | StringRef::iterator I = Position - 1; |
| 95 | // We calulate the number of consecutive '\'s before the current position |
| 96 | // by iterating backwards through our string. |
| 97 | while (I >= First && *I == '\\') --I; |
| 98 | // (Position - 1 - I) now contains the number of '\'s before the current |
| 99 | // position. If it is odd, the character at 'Positon' was escaped. |
| 100 | return (Position - 1 - I) % 2 == 1; |
| 101 | } |
| 102 | |
| 103 | // Parses a JSONString, assuming that the current position is on a quote. |
| 104 | JSONString *JSONParser::parseString() { |
| 105 | assert(Position != Input.end()); |
| 106 | assert(!isWhitespace()); |
| 107 | if (errorIfNotAt('"', "at start of string")) |
| 108 | return 0; |
| 109 | StringRef::iterator First = Position + 1; |
| 110 | |
| 111 | // Benchmarking shows that this loop is the hot path of the application with |
| 112 | // about 2/3rd of the runtime cycles. Since escaped quotes are not the common |
| 113 | // case, and multiple escaped backslashes before escaped quotes are very rare, |
| 114 | // we pessimize this case to achieve a smaller inner loop in the common case. |
| 115 | // We're doing that by having a quick inner loop that just scans for the next |
| 116 | // quote. Once we find the quote we check the last character to see whether |
| 117 | // the quote might have been escaped. If the last character is not a '\', we |
| 118 | // know the quote was not escaped and have thus found the end of the string. |
| 119 | // If the immediately preceding character was a '\', we have to scan backwards |
| 120 | // to see whether the previous character was actually an escaped backslash, or |
| 121 | // an escape character for the quote. If we find that the current quote was |
| 122 | // escaped, we continue parsing for the next quote and repeat. |
| 123 | // This optimization brings around 30% performance improvements. |
| 124 | do { |
| 125 | // Step over the current quote. |
| 126 | ++Position; |
| 127 | // Find the next quote. |
| 128 | while (Position != Input.end() && *Position != '"') |
| 129 | ++Position; |
| 130 | if (errorIfAtEndOfFile("\" at end of string")) |
| 131 | return 0; |
| 132 | // Repeat until the previous character was not a '\' or was an escaped |
| 133 | // backslash. |
| 134 | } while (*(Position - 1) == '\\' && wasEscaped(First, Position)); |
| 135 | |
| 136 | return new (ValueAllocator.Allocate<JSONString>()) |
| 137 | JSONString(StringRef(First, Position - First)); |
| 138 | } |
| 139 | |
| 140 | |
| 141 | // Advances the position to the next non-whitespace position. |
| 142 | void JSONParser::nextNonWhitespace() { |
| 143 | do { |
| 144 | ++Position; |
| 145 | } while (isWhitespace()); |
| 146 | } |
| 147 | |
| 148 | // Checks if there is a whitespace character at the current position. |
| 149 | bool JSONParser::isWhitespace() { |
| 150 | return Position != Input.end() && (*Position == ' ' || *Position == '\t' || |
| 151 | *Position == '\n' || *Position == '\r'); |
| 152 | } |
| 153 | |
| 154 | bool JSONParser::failed() const { |
| 155 | return !ErrorMessage.empty(); |
| 156 | } |
| 157 | |
| 158 | std::string JSONParser::getErrorMessage() const { |
| 159 | return ErrorMessage; |
| 160 | } |
| 161 | |
| 162 | bool JSONAtom::skip() const { |
| 163 | switch (MyKind) { |
| 164 | case JK_Array: return cast<JSONArray>(this)->skip(); |
| 165 | case JK_Object: return cast<JSONObject>(this)->skip(); |
| 166 | case JK_String: return cast<JSONString>(this)->skip(); |
| 167 | case JK_KeyValuePair: return cast<JSONKeyValuePair>(this)->skip(); |
| 168 | } |
| 169 | llvm_unreachable("Impossible enum value."); |
| 170 | } |
| 171 | |
| 172 | // Parses a JSONValue, assuming that the current position is at the first |
| 173 | // character of the value. |
| 174 | JSONValue *JSONParser::parseValue() { |
| 175 | assert(Position != Input.end()); |
| 176 | assert(!isWhitespace()); |
| 177 | switch (*Position) { |
| 178 | case '[': |
| 179 | return new (ValueAllocator.Allocate<JSONArray>(1)) JSONArray(this); |
| 180 | case '{': |
| 181 | return new (ValueAllocator.Allocate<JSONObject>(1)) JSONObject(this); |
| 182 | case '"': |
| 183 | return parseString(); |
| 184 | default: |
| 185 | setExpectedError("'[', '{' or '\"' at start of value", *Position); |
| 186 | return 0; |
| 187 | } |
| 188 | } |
| 189 | |
| 190 | // Parses a JSONKeyValuePair, assuming that the current position is at the first |
| 191 | // character of the key, value pair. |
| 192 | JSONKeyValuePair *JSONParser::parseKeyValuePair() { |
| 193 | assert(Position != Input.end()); |
| 194 | assert(!isWhitespace()); |
| 195 | |
| 196 | JSONString *Key = parseString(); |
| 197 | if (Key == 0) |
| 198 | return 0; |
| 199 | |
| 200 | nextNonWhitespace(); |
| 201 | if (errorIfNotAt(':', "between key and value")) |
| 202 | return 0; |
| 203 | |
| 204 | nextNonWhitespace(); |
| 205 | const JSONValue *Value = parseValue(); |
| 206 | if (Value == 0) |
| 207 | return 0; |
| 208 | |
| 209 | return new (ValueAllocator.Allocate<JSONKeyValuePair>(1)) |
| 210 | JSONKeyValuePair(Key, Value); |
| 211 | } |
| 212 | |
| 213 | template <> JSONValue *JSONParser::parseElement() { |
| 214 | return parseValue(); |
| 215 | } |
| 216 | |
| 217 | template <> JSONKeyValuePair *JSONParser::parseElement() { |
| 218 | return parseKeyValuePair(); |
| 219 | } |
| 220 | |
| 221 | } // end namespace llvm |