2 * Copyright 2016 Facebook, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
17 #include <folly/json.h>
19 #include <boost/next_prior.hpp>
20 #include <boost/algorithm/string.hpp>
22 #include <folly/Conv.h>
23 #include <folly/Portability.h>
24 #include <folly/Range.h>
25 #include <folly/String.h>
26 #include <folly/Unicode.h>
27 #include <folly/portability/Constexpr.h>
31 //////////////////////////////////////////////////////////////////////
37 const unsigned char*& p,
38 const unsigned char* const e,
40 /* The following encodings are valid, except for the 5 and 6 byte
44 * 1110xxxx 10xxxxxx 10xxxxxx
45 * 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx
46 * 111110xx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
47 * 1111110x 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
50 auto skip = [&] { ++p; return U'\ufffd'; };
53 if (skipOnError) return skip();
54 throw std::runtime_error("folly::decodeUtf8 empty/invalid string");
57 unsigned char fst = *p;
63 static const uint32_t bitMask[] = {
70 // upper control bits are masked out later
73 if ((fst & 0xC0) != 0xC0) {
74 if (skipOnError) return skip();
75 throw std::runtime_error(to<std::string>("folly::decodeUtf8 i=0 d=", d));
80 for (unsigned int i = 1; i != 3 && p + i < e; ++i) {
81 unsigned char tmp = p[i];
83 if ((tmp & 0xC0) != 0x80) {
84 if (skipOnError) return skip();
85 throw std::runtime_error(
86 to<std::string>("folly::decodeUtf8 i=", i, " tmp=", (uint32_t)tmp));
89 d = (d << 6) | (tmp & 0x3F);
95 // overlong, could have been encoded with i bytes
96 if ((d & ~bitMask[i - 1]) == 0) {
97 if (skipOnError) return skip();
98 throw std::runtime_error(
99 to<std::string>("folly::decodeUtf8 i=", i, " d=", d));
102 // check for surrogates only needed for 3 bytes
104 if ((d >= 0xD800 && d <= 0xDFFF) || d > 0x10FFFF) {
105 if (skipOnError) return skip();
106 throw std::runtime_error(
107 to<std::string>("folly::decodeUtf8 i=", i, " d=", d));
116 if (skipOnError) return skip();
117 throw std::runtime_error("folly::decodeUtf8 encoding length maxed out");
123 unsigned* indentLevel,
124 serialization_opts const* opts)
125 : out_(out), indentLevel_(indentLevel), opts_(*opts) {}
127 void operator()(dynamic const& v) const {
129 case dynamic::DOUBLE:
130 if (!opts_.allow_nan_inf &&
131 (std::isnan(v.asDouble()) || std::isinf(v.asDouble()))) {
132 throw std::runtime_error("folly::toJson: JSON object value was a "
135 toAppend(v.asDouble(), &out_, opts_.double_mode, opts_.double_num_digits);
137 case dynamic::INT64: {
138 auto intval = v.asInt();
139 if (opts_.javascript_safe) {
140 // Use folly::to to check that this integer can be represented
141 // as a double without loss of precision.
142 intval = int64_t(to<double>(intval));
144 toAppend(intval, &out_);
148 out_ += v.asBool() ? "true" : "false";
153 case dynamic::STRING:
154 escapeString(v.asString(), out_, opts_);
156 case dynamic::OBJECT:
163 CHECK(0) << "Bad type " << v.type();
168 void printKV(const std::pair<const dynamic, dynamic>& p) const {
169 if (!opts_.allow_non_string_keys && !p.first.isString()) {
170 throw std::runtime_error("folly::toJson: JSON object key was not a "
178 template <typename Iterator>
179 void printKVPairs(Iterator begin, Iterator end) const {
181 for (++begin; begin != end; ++begin) {
188 void printObject(dynamic const& o) const {
197 if (opts_.sort_keys) {
198 std::vector<std::pair<dynamic, dynamic>> items(
199 o.items().begin(), o.items().end());
200 std::sort(items.begin(), items.end());
201 printKVPairs(items.begin(), items.end());
203 printKVPairs(o.items().begin(), o.items().end());
210 void printArray(dynamic const& a) const {
220 for (auto& val : range(boost::next(a.begin()), a.end())) {
231 void outdent() const {
237 void indent() const {
243 void newline() const {
245 out_ += to<std::string>('\n', std::string(*indentLevel_ * 2, ' '));
249 void mapColon() const {
250 out_ += indentLevel_ ? " : " : ":";
255 unsigned* const indentLevel_;
256 serialization_opts const& opts_;
259 //////////////////////////////////////////////////////////////////////
261 struct ParseError : std::runtime_error {
262 explicit ParseError(int line)
263 : std::runtime_error(to<std::string>("json parse error on line ", line))
266 explicit ParseError(int line, std::string const& context,
267 std::string const& expected)
268 : std::runtime_error(to<std::string>("json parse error on line ", line,
269 !context.empty() ? to<std::string>(" near `", context, '\'')
274 explicit ParseError(std::string const& msg)
275 : std::runtime_error("json parse error: " + msg)
279 // Wraps our input buffer with some helper functions.
281 explicit Input(StringPiece range, json::serialization_opts const* opts)
289 Input(Input const&) = delete;
290 Input& operator=(Input const&) = delete;
292 char const* begin() const { return range_.begin(); }
294 // Parse ahead for as long as the supplied predicate is satisfied,
295 // returning a range of what was skipped.
296 template<class Predicate>
297 StringPiece skipWhile(const Predicate& p) {
298 std::size_t skipped = 0;
299 for (; skipped < range_.size(); ++skipped) {
300 if (!p(range_[skipped])) {
303 if (range_[skipped] == '\n') {
307 auto ret = range_.subpiece(0, skipped);
308 range_.advance(skipped);
313 StringPiece skipDigits() {
314 return skipWhile([] (char c) { return c >= '0' && c <= '9'; });
317 StringPiece skipMinusAndDigits() {
318 bool firstChar = true;
319 return skipWhile([&firstChar] (char c) {
320 bool result = (c >= '0' && c <= '9') || (firstChar && c == '-');
326 void skipWhitespace() {
327 range_ = folly::skipWhitespace(range_);
331 void expect(char c) {
333 throw ParseError(lineNum_, context(),
334 to<std::string>("expected '", c, '\''));
339 std::size_t size() const {
340 return range_.size();
343 int operator*() const {
355 return to<T>(&range_);
356 } catch (std::exception const& e) {
361 bool consume(StringPiece str) {
362 if (boost::starts_with(range_, str)) {
363 range_.advance(str.size());
370 std::string context() const {
371 return range_.subpiece(0, 16 /* arbitrary */).toString();
374 dynamic error(char const* what) const {
375 throw ParseError(lineNum_, context(), what);
378 json::serialization_opts const& getOpts() {
382 void incrementRecursionLevel() {
383 if (currentRecursionLevel_ > opts_.recursion_limit) {
384 error("recursion limit exceeded");
386 currentRecursionLevel_++;
389 void decrementRecursionLevel() {
390 currentRecursionLevel_--;
394 void storeCurrent() {
395 current_ = range_.empty() ? EOF : range_.front();
400 json::serialization_opts const& opts_;
403 unsigned int currentRecursionLevel_{0};
406 class RecursionGuard {
408 explicit RecursionGuard(Input& in) : in_(in) {
409 in_.incrementRecursionLevel();
413 in_.decrementRecursionLevel();
420 dynamic parseValue(Input& in);
421 std::string parseString(Input& in);
422 dynamic parseNumber(Input& in);
424 dynamic parseObject(Input& in) {
428 dynamic ret = dynamic::object;
437 if (in.getOpts().allow_trailing_comma && *in == '}') {
440 if (*in == '\"') { // string
441 auto key = parseString(in);
445 ret.insert(std::move(key), parseValue(in));
446 } else if (!in.getOpts().allow_non_string_keys) {
447 in.error("expected string for object key name");
449 auto key = parseValue(in);
453 ret.insert(std::move(key), parseValue(in));
468 dynamic parseArray(Input& in) {
472 dynamic ret = dynamic::array;
481 if (in.getOpts().allow_trailing_comma && *in == ']') {
484 ret.push_back(parseValue(in));
497 dynamic parseNumber(Input& in) {
498 bool const negative = (*in == '-');
499 if (negative && in.consume("-Infinity")) {
500 if (in.getOpts().parse_numbers_as_strings) {
503 return -std::numeric_limits<double>::infinity();
507 auto integral = in.skipMinusAndDigits();
508 if (negative && integral.size() < 2) {
509 in.error("expected digits after `-'");
512 auto const wasE = *in == 'e' || *in == 'E';
514 constexpr const char* maxInt = "9223372036854775807";
515 constexpr const char* minInt = "-9223372036854775808";
516 constexpr auto maxIntLen = constexpr_strlen(maxInt);
517 constexpr auto minIntLen = constexpr_strlen(minInt);
519 if (*in != '.' && !wasE && in.getOpts().parse_numbers_as_strings) {
523 if (*in != '.' && !wasE) {
524 if (LIKELY(!in.getOpts().double_fallback || integral.size() < maxIntLen) ||
525 (!negative && integral.size() == maxIntLen && integral <= maxInt) ||
526 (negative && integral.size() == minIntLen && integral <= minInt)) {
527 auto val = to<int64_t>(integral);
531 auto val = to<double>(integral);
537 auto end = !wasE ? (++in, in.skipDigits().end()) : in.begin();
538 if (*in == 'e' || *in == 'E') {
540 if (*in == '+' || *in == '-') {
543 auto expPart = in.skipDigits();
546 auto fullNum = range(integral.begin(), end);
547 if (in.getOpts().parse_numbers_as_strings) {
550 auto val = to<double>(fullNum);
554 std::string decodeUnicodeEscape(Input& in) {
555 auto hexVal = [&] (int c) -> uint16_t {
557 c >= '0' && c <= '9' ? c - '0' :
558 c >= 'a' && c <= 'f' ? c - 'a' + 10 :
559 c >= 'A' && c <= 'F' ? c - 'A' + 10 :
560 (in.error("invalid hex digit"), 0));
563 auto readHex = [&]() -> uint16_t {
565 in.error("expected 4 hex digits");
568 uint16_t ret = hexVal(*in) * 4096;
570 ret += hexVal(*in) * 256;
572 ret += hexVal(*in) * 16;
580 * If the value encoded is in the surrogate pair range, we need to
581 * make sure there is another escape that we can use also.
583 uint32_t codePoint = readHex();
584 if (codePoint >= 0xd800 && codePoint <= 0xdbff) {
585 if (!in.consume("\\u")) {
586 in.error("expected another unicode escape for second half of "
589 uint16_t second = readHex();
590 if (second >= 0xdc00 && second <= 0xdfff) {
591 codePoint = 0x10000 + ((codePoint & 0x3ff) << 10) +
594 in.error("second character in surrogate pair is invalid");
596 } else if (codePoint >= 0xdc00 && codePoint <= 0xdfff) {
597 in.error("invalid unicode code point (in range [0xdc00,0xdfff])");
600 return codePointToUtf8(codePoint);
603 std::string parseString(Input& in) {
609 auto range = in.skipWhile(
610 [] (char c) { return c != '\"' && c != '\\'; }
612 ret.append(range.begin(), range.end());
621 case '\"': ret.push_back('\"'); ++in; break;
622 case '\\': ret.push_back('\\'); ++in; break;
623 case '/': ret.push_back('/'); ++in; break;
624 case 'b': ret.push_back('\b'); ++in; break;
625 case 'f': ret.push_back('\f'); ++in; break;
626 case 'n': ret.push_back('\n'); ++in; break;
627 case 'r': ret.push_back('\r'); ++in; break;
628 case 't': ret.push_back('\t'); ++in; break;
629 case 'u': ++in; ret += decodeUnicodeEscape(in); break;
631 in.error(to<std::string>("unknown escape ", *in, " in string").c_str());
636 in.error("unterminated string");
640 * Apparently we're actually supposed to ban all control
641 * characters from strings. This seems unnecessarily
642 * restrictive, so we're only banning zero bytes. (Since the
643 * string is presumed to be UTF-8 encoded it's fine to just
646 in.error("null byte in string");
649 ret.push_back(char(*in));
656 dynamic parseValue(Input& in) {
657 RecursionGuard guard(in);
660 return *in == '[' ? parseArray(in) :
661 *in == '{' ? parseObject(in) :
662 *in == '\"' ? parseString(in) :
663 (*in == '-' || (*in >= '0' && *in <= '9')) ? parseNumber(in) :
664 in.consume("true") ? true :
665 in.consume("false") ? false :
666 in.consume("null") ? nullptr :
667 in.consume("Infinity") ?
668 (in.getOpts().parse_numbers_as_strings ? (dynamic)"Infinity" :
669 (dynamic)std::numeric_limits<double>::infinity()) :
671 (in.getOpts().parse_numbers_as_strings ? (dynamic)"NaN" :
672 (dynamic)std::numeric_limits<double>::quiet_NaN()) :
673 in.error("expected json value");
678 //////////////////////////////////////////////////////////////////////
680 std::string serialize(dynamic const& dyn, serialization_opts const& opts) {
682 unsigned indentLevel = 0;
683 Printer p(ret, opts.pretty_formatting ? &indentLevel : nullptr, &opts);
688 // Escape a string so that it is legal to print it in JSON text.
692 const serialization_opts& opts) {
693 auto hexDigit = [] (uint8_t c) -> char {
694 return c < 10 ? c + '0' : c - 10 + 'a';
699 auto* p = reinterpret_cast<const unsigned char*>(input.begin());
700 auto* q = reinterpret_cast<const unsigned char*>(input.begin());
701 auto* e = reinterpret_cast<const unsigned char*>(input.end());
704 // Since non-ascii encoding inherently does utf8 validation
705 // we explicitly validate utf8 only if non-ascii encoding is disabled.
706 if ((opts.validate_utf8 || opts.skip_invalid_utf8)
707 && !opts.encode_non_ascii) {
708 // to achieve better spatial and temporal coherence
709 // we do utf8 validation progressively along with the
710 // string-escaping instead of two separate passes
712 // as the encoding progresses, q will stay at or ahead of p
715 // as p catches up with q, move q forward
717 // calling utf8_decode has the side effect of
718 // checking that utf8 encodings are valid
719 char32_t v = decodeUtf8(q, e, opts.skip_invalid_utf8);
720 if (opts.skip_invalid_utf8 && v == U'\ufffd') {
721 out.append(u8"\ufffd");
727 if (opts.encode_non_ascii && (*p & 0x80)) {
728 // note that this if condition captures utf8 chars
729 // with value > 127, so size > 1 byte
730 char32_t v = decodeUtf8(p, e, opts.skip_invalid_utf8);
732 out.push_back(hexDigit(uint8_t(v >> 12)));
733 out.push_back(hexDigit((v >> 8) & 0x0f));
734 out.push_back(hexDigit((v >> 4) & 0x0f));
735 out.push_back(hexDigit(v & 0x0f));
736 } else if (*p == '\\' || *p == '\"') {
738 out.push_back(char(*p++));
739 } else if (*p <= 0x1f) {
741 case '\b': out.append("\\b"); p++; break;
742 case '\f': out.append("\\f"); p++; break;
743 case '\n': out.append("\\n"); p++; break;
744 case '\r': out.append("\\r"); p++; break;
745 case '\t': out.append("\\t"); p++; break;
747 // note that this if condition captures non readable chars
748 // with value < 32, so size = 1 byte (e.g control chars).
750 out.push_back(hexDigit((*p & 0xf0) >> 4));
751 out.push_back(hexDigit(*p & 0xf));
762 std::string stripComments(StringPiece jsonC) {
769 } state = State::None;
771 for (size_t i = 0; i < jsonC.size(); ++i) {
772 auto s = jsonC.subpiece(i);
775 if (s.startsWith("/*")) {
776 state = State::InlineComment;
779 } else if (s.startsWith("//")) {
780 state = State::LineComment;
783 } else if (s[0] == '\"') {
784 state = State::InString;
786 result.push_back(s[0]);
788 case State::InString:
790 if (UNLIKELY(s.size() == 1)) {
791 throw std::logic_error("Invalid JSONC: string is not terminated");
793 result.push_back(s[0]);
794 result.push_back(s[1]);
797 } else if (s[0] == '\"') {
800 result.push_back(s[0]);
802 case State::InlineComment:
803 if (s.startsWith("*/")) {
808 case State::LineComment:
810 // skip the line break. It doesn't matter.
815 throw std::logic_error("Unknown comment state");
823 //////////////////////////////////////////////////////////////////////
825 dynamic parseJson(StringPiece range) {
826 return parseJson(range, json::serialization_opts());
831 json::serialization_opts const& opts) {
833 json::Input in(range, &opts);
835 auto ret = parseValue(in);
837 if (in.size() && *in != '\0') {
838 in.error("parsing didn't consume all input");
843 std::string toJson(dynamic const& dyn) {
844 return json::serialize(dyn, json::serialization_opts());
847 std::string toPrettyJson(dynamic const& dyn) {
848 json::serialization_opts opts;
849 opts.pretty_formatting = true;
850 return json::serialize(dyn, opts);
853 //////////////////////////////////////////////////////////////////////
854 // dynamic::print_as_pseudo_json() is implemented here for header
855 // ordering reasons (most of the dynamic implementation is in
856 // dynamic-inl.h, which we don't want to include json.h).
858 void dynamic::print_as_pseudo_json(std::ostream& out) const {
859 json::serialization_opts opts;
860 opts.allow_non_string_keys = true;
861 opts.allow_nan_inf = true;
862 out << json::serialize(*this, opts);
865 void PrintTo(const dynamic& dyn, std::ostream* os) {
866 json::serialization_opts opts;
867 opts.allow_nan_inf = true;
868 opts.allow_non_string_keys = true;
869 opts.pretty_formatting = true;
870 opts.sort_keys = true;
871 *os << json::serialize(dyn, opts);
874 //////////////////////////////////////////////////////////////////////