1 //===--- JSONParser.cpp - Simple JSON parser ------------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements a JSON parser.
12 //===----------------------------------------------------------------------===//
14 #include "llvm/Support/JSONParser.h"
16 #include "llvm/ADT/Twine.h"
17 #include "llvm/Support/Casting.h"
18 #include "llvm/Support/MemoryBuffer.h"
22 JSONParser::JSONParser(StringRef Input, SourceMgr *SM)
23 : SM(SM), Failed(false) {
24 InputBuffer = MemoryBuffer::getMemBuffer(Input, "JSON");
25 SM->AddNewSourceBuffer(InputBuffer, SMLoc());
26 End = InputBuffer->getBuffer().end();
27 Position = InputBuffer->getBuffer().begin();
30 JSONValue *JSONParser::parseRoot() {
31 if (Position != InputBuffer->getBuffer().begin())
32 report_fatal_error("Cannot resuse JSONParser.");
35 if (errorIfAtEndOfFile("'[' or '{' at start of JSON text"))
39 return new (ValueAllocator.Allocate<JSONArray>(1)) JSONArray(this);
41 return new (ValueAllocator.Allocate<JSONObject>(1)) JSONObject(this);
43 setExpectedError("'[' or '{' at start of JSON text", *Position);
48 bool JSONParser::validate() {
49 JSONValue *Root = parseRoot();
56 bool JSONParser::skip(const JSONAtom &Atom) {
57 switch(Atom.getKind()) {
58 case JSONAtom::JK_Array: return skipContainer(*cast<JSONArray>(&Atom));
59 case JSONAtom::JK_Object: return skipContainer(*cast<JSONObject>(&Atom));
60 case JSONAtom::JK_String: return true;
61 case JSONAtom::JK_KeyValuePair:
62 return skip(*cast<JSONKeyValuePair>(&Atom)->Value);
64 llvm_unreachable("Impossible enum value.");
67 // Sets the current error to:
68 // "expected <Expected>, but found <Found>".
69 void JSONParser::setExpectedError(StringRef Expected, StringRef Found) {
70 SM->PrintMessage(SMLoc::getFromPointer(Position), SourceMgr::DK_Error,
71 "expected " + Expected + ", but found " + Found + ".", ArrayRef<SMRange>());
75 // Sets the current error to:
76 // "expected <Expected>, but found <Found>".
77 void JSONParser::setExpectedError(StringRef Expected, char Found) {
78 setExpectedError(Expected, ("'" + StringRef(&Found, 1) + "'").str());
81 // If there is no character available, returns true and sets the current error
82 // to: "expected <Expected>, but found EOF.".
83 bool JSONParser::errorIfAtEndOfFile(StringRef Expected) {
84 if (Position == End) {
85 setExpectedError(Expected, "EOF");
91 // Sets the current error if the current character is not C to:
92 // "expected 'C', but got <current character>".
93 bool JSONParser::errorIfNotAt(char C, StringRef Message) {
95 std::string Expected =
96 ("'" + StringRef(&C, 1) + "' " + Message).str();
98 setExpectedError(Expected, "EOF");
100 setExpectedError(Expected, *Position);
106 // Forbidding inlining improves performance by roughly 20%.
107 // FIXME: Remove once llvm optimizes this to the faster version without hints.
108 LLVM_ATTRIBUTE_NOINLINE static bool
109 wasEscaped(StringRef::iterator First, StringRef::iterator Position);
111 // Returns whether a character at 'Position' was escaped with a leading '\'.
112 // 'First' specifies the position of the first character in the string.
113 static bool wasEscaped(StringRef::iterator First,
114 StringRef::iterator Position) {
115 assert(Position - 1 >= First);
116 StringRef::iterator I = Position - 1;
117 // We calulate the number of consecutive '\'s before the current position
118 // by iterating backwards through our string.
119 while (I >= First && *I == '\\') --I;
120 // (Position - 1 - I) now contains the number of '\'s before the current
121 // position. If it is odd, the character at 'Positon' was escaped.
122 return (Position - 1 - I) % 2 == 1;
125 // Parses a JSONString, assuming that the current position is on a quote.
126 JSONString *JSONParser::parseString() {
127 assert(Position != End);
128 assert(!isWhitespace());
129 if (errorIfNotAt('"', "at start of string"))
131 StringRef::iterator First = Position + 1;
133 // Benchmarking shows that this loop is the hot path of the application with
134 // about 2/3rd of the runtime cycles. Since escaped quotes are not the common
135 // case, and multiple escaped backslashes before escaped quotes are very rare,
136 // we pessimize this case to achieve a smaller inner loop in the common case.
137 // We're doing that by having a quick inner loop that just scans for the next
138 // quote. Once we find the quote we check the last character to see whether
139 // the quote might have been escaped. If the last character is not a '\', we
140 // know the quote was not escaped and have thus found the end of the string.
141 // If the immediately preceding character was a '\', we have to scan backwards
142 // to see whether the previous character was actually an escaped backslash, or
143 // an escape character for the quote. If we find that the current quote was
144 // escaped, we continue parsing for the next quote and repeat.
145 // This optimization brings around 30% performance improvements.
147 // Step over the current quote.
149 // Find the next quote.
150 while (Position != End && *Position != '"')
152 if (errorIfAtEndOfFile("'\"' at end of string"))
154 // Repeat until the previous character was not a '\' or was an escaped
156 } while (*(Position - 1) == '\\' && wasEscaped(First, Position));
158 return new (ValueAllocator.Allocate<JSONString>())
159 JSONString(StringRef(First, Position - First));
163 // Advances the position to the next non-whitespace position.
164 void JSONParser::nextNonWhitespace() {
167 } while (isWhitespace());
170 // Checks if there is a whitespace character at the current position.
171 bool JSONParser::isWhitespace() {
172 return *Position == ' ' || *Position == '\t' ||
173 *Position == '\n' || *Position == '\r';
176 bool JSONParser::failed() const {
180 // Parses a JSONValue, assuming that the current position is at the first
181 // character of the value.
182 JSONValue *JSONParser::parseValue() {
183 assert(Position != End);
184 assert(!isWhitespace());
187 return new (ValueAllocator.Allocate<JSONArray>(1)) JSONArray(this);
189 return new (ValueAllocator.Allocate<JSONObject>(1)) JSONObject(this);
191 return parseString();
193 setExpectedError("'[', '{' or '\"' at start of value", *Position);
198 // Parses a JSONKeyValuePair, assuming that the current position is at the first
199 // character of the key, value pair.
200 JSONKeyValuePair *JSONParser::parseKeyValuePair() {
201 assert(Position != End);
202 assert(!isWhitespace());
204 JSONString *Key = parseString();
209 if (errorIfNotAt(':', "between key and value"))
213 const JSONValue *Value = parseValue();
217 return new (ValueAllocator.Allocate<JSONKeyValuePair>(1))
218 JSONKeyValuePair(Key, Value);
221 template <> JSONValue *JSONParser::parseElement() {
225 template <> JSONKeyValuePair *JSONParser::parseElement() {
226 return parseKeyValuePair();