2 * Copyright 2014 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/Range.h>
24 #include <folly/String.h>
25 #include <folly/Unicode.h>
29 //////////////////////////////////////////////////////////////////////
35 const unsigned char*& p,
36 const unsigned char* const e,
38 /* The following encodings are valid, except for the 5 and 6 byte
42 * 1110xxxx 10xxxxxx 10xxxxxx
43 * 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx
44 * 111110xx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
45 * 1111110x 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
48 auto skip = [&] { ++p; return U'\ufffd'; };
51 if (skipOnError) return skip();
52 throw std::runtime_error("folly::decodeUtf8 empty/invalid string");
55 unsigned char fst = *p;
61 static const uint32_t bitMask[] = {
68 // upper control bits are masked out later
71 if ((fst & 0xC0) != 0xC0) {
72 if (skipOnError) return skip();
73 throw std::runtime_error(to<std::string>("folly::decodeUtf8 i=0 d=", d));
78 for (unsigned int i = 1; i != 3 && p + i < e; ++i) {
79 unsigned char tmp = p[i];
81 if ((tmp & 0xC0) != 0x80) {
82 if (skipOnError) return skip();
83 throw std::runtime_error(
84 to<std::string>("folly::decodeUtf8 i=", i, " tmp=", (uint32_t)tmp));
87 d = (d << 6) | (tmp & 0x3F);
93 // overlong, could have been encoded with i bytes
94 if ((d & ~bitMask[i - 1]) == 0) {
95 if (skipOnError) return skip();
96 throw std::runtime_error(
97 to<std::string>("folly::decodeUtf8 i=", i, " d=", d));
100 // check for surrogates only needed for 3 bytes
102 if ((d >= 0xD800 && d <= 0xDFFF) || d > 0x10FFFF) {
103 if (skipOnError) return skip();
104 throw std::runtime_error(
105 to<std::string>("folly::decodeUtf8 i=", i, " d=", d));
114 if (skipOnError) return skip();
115 throw std::runtime_error("folly::decodeUtf8 encoding length maxed out");
119 explicit Printer(fbstring& out,
120 unsigned* indentLevel,
121 serialization_opts const* opts)
123 , indentLevel_(indentLevel)
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<fbstring>('\n', fbstring(*indentLevel_ * 2, ' '));
249 void mapColon() const {
250 out_ += indentLevel_ ? " : " : ":";
255 unsigned* const indentLevel_;
256 serialization_opts const& opts_;
259 // Wraps our input buffer with some helper functions.
261 explicit Input(StringPiece range, json::serialization_opts const* opts)
269 Input(Input const&) = delete;
270 Input& operator=(Input const&) = delete;
272 char const* begin() const { return range_.begin(); }
274 // Parse ahead for as long as the supplied predicate is satisfied,
275 // returning a range of what was skipped.
276 template<class Predicate>
277 StringPiece skipWhile(const Predicate& p) {
278 std::size_t skipped = 0;
279 for (; skipped < range_.size(); ++skipped) {
280 if (!p(range_[skipped])) {
283 if (range_[skipped] == '\n') {
287 auto ret = range_.subpiece(0, skipped);
288 range_.advance(skipped);
293 StringPiece skipDigits() {
294 return skipWhile([] (char c) { return c >= '0' && c <= '9'; });
297 StringPiece skipMinusAndDigits() {
298 bool firstChar = true;
299 return skipWhile([&firstChar] (char c) {
300 bool result = (c >= '0' && c <= '9') || (firstChar && c == '-');
306 void skipWhitespace() {
307 range_ = folly::skipWhitespace(range_);
311 void expect(char c) {
313 throw ParseError(lineNum_, context(),
314 to<std::string>("expected '", c, '\''));
319 std::size_t size() const {
320 return range_.size();
323 int operator*() const {
335 return to<T>(&range_);
336 } catch (std::exception const& e) {
341 bool consume(StringPiece str) {
342 if (boost::starts_with(range_, str)) {
343 range_.advance(str.size());
350 std::string context() const {
351 return range_.subpiece(0, 16 /* arbitrary */).toString();
354 dynamic error(char const* what) const {
355 throw ParseError(lineNum_, context(), what);
358 json::serialization_opts const& getOpts() {
363 void storeCurrent() {
364 current_ = range_.empty() ? EOF : range_.front();
369 json::serialization_opts const& opts_;
374 dynamic parseValue(Input& in);
375 fbstring parseString(Input& in);
376 dynamic parseNumber(Input& in);
378 dynamic parseObject(Input& in) {
382 dynamic ret = dynamic::object;
391 if (in.getOpts().allow_trailing_comma && *in == '}') {
394 if (*in == '\"') { // string
395 auto key = parseString(in);
399 ret.insert(std::move(key), parseValue(in));
400 } else if (!in.getOpts().allow_non_string_keys) {
401 in.error("expected string for object key name");
403 auto key = parseValue(in);
407 ret.insert(std::move(key), parseValue(in));
422 dynamic parseArray(Input& in) {
435 if (in.getOpts().allow_trailing_comma && *in == ']') {
438 ret.push_back(parseValue(in));
451 dynamic parseNumber(Input& in) {
452 bool const negative = (*in == '-');
454 if (in.consume("-Infinity")) {
455 return -std::numeric_limits<double>::infinity();
459 auto integral = in.skipMinusAndDigits();
460 if (negative && integral.size() < 2) {
461 in.error("expected digits after `-'");
464 auto const wasE = *in == 'e' || *in == 'E';
465 if (*in != '.' && !wasE) {
466 auto val = to<int64_t>(integral);
471 auto end = !wasE ? (++in, in.skipDigits().end()) : in.begin();
472 if (*in == 'e' || *in == 'E') {
474 if (*in == '+' || *in == '-') {
477 auto expPart = in.skipDigits();
480 auto fullNum = range(integral.begin(), end);
482 auto val = to<double>(fullNum);
486 fbstring decodeUnicodeEscape(Input& in) {
487 auto hexVal = [&] (char c) -> unsigned {
488 return c >= '0' && c <= '9' ? c - '0' :
489 c >= 'a' && c <= 'f' ? c - 'a' + 10 :
490 c >= 'A' && c <= 'F' ? c - 'A' + 10 :
491 (in.error("invalid hex digit"), 0);
494 auto readHex = [&]() -> uint16_t {
496 in.error("expected 4 hex digits");
499 uint16_t ret = hexVal(*in) * 4096;
501 ret += hexVal(*in) * 256;
503 ret += hexVal(*in) * 16;
511 * If the value encoded is in the surrogate pair range, we need to
512 * make sure there is another escape that we can use also.
514 uint32_t codePoint = readHex();
515 if (codePoint >= 0xd800 && codePoint <= 0xdbff) {
516 if (!in.consume("\\u")) {
517 in.error("expected another unicode escape for second half of "
520 uint16_t second = readHex();
521 if (second >= 0xdc00 && second <= 0xdfff) {
522 codePoint = 0x10000 + ((codePoint & 0x3ff) << 10) +
525 in.error("second character in surrogate pair is invalid");
527 } else if (codePoint >= 0xdc00 && codePoint <= 0xdfff) {
528 in.error("invalid unicode code point (in range [0xdc00,0xdfff])");
531 return codePointToUtf8(codePoint);
534 fbstring parseString(Input& in) {
540 auto range = in.skipWhile(
541 [] (char c) { return c != '\"' && c != '\\'; }
543 ret.append(range.begin(), range.end());
552 case '\"': ret.push_back('\"'); ++in; break;
553 case '\\': ret.push_back('\\'); ++in; break;
554 case '/': ret.push_back('/'); ++in; break;
555 case 'b': ret.push_back('\b'); ++in; break;
556 case 'f': ret.push_back('\f'); ++in; break;
557 case 'n': ret.push_back('\n'); ++in; break;
558 case 'r': ret.push_back('\r'); ++in; break;
559 case 't': ret.push_back('\t'); ++in; break;
560 case 'u': ++in; ret += decodeUnicodeEscape(in); break;
561 default: in.error(to<fbstring>("unknown escape ", *in,
562 " in string").c_str());
567 in.error("unterminated string");
571 * Apparently we're actually supposed to ban all control
572 * characters from strings. This seems unnecessarily
573 * restrictive, so we're only banning zero bytes. (Since the
574 * string is presumed to be UTF-8 encoded it's fine to just
577 in.error("null byte in string");
587 dynamic parseValue(Input& in) {
589 return *in == '[' ? parseArray(in) :
590 *in == '{' ? parseObject(in) :
591 *in == '\"' ? parseString(in) :
592 (*in == '-' || (*in >= '0' && *in <= '9')) ? parseNumber(in) :
593 in.consume("true") ? true :
594 in.consume("false") ? false :
595 in.consume("null") ? nullptr :
596 in.consume("Infinity") ? std::numeric_limits<double>::infinity() :
597 in.consume("NaN") ? std::numeric_limits<double>::quiet_NaN() :
598 in.error("expected json value");
603 //////////////////////////////////////////////////////////////////////
605 fbstring serialize(dynamic const& dyn, serialization_opts const& opts) {
607 unsigned indentLevel = 0;
608 Printer p(ret, opts.pretty_formatting ? &indentLevel : nullptr, &opts);
613 // Escape a string so that it is legal to print it in JSON text.
614 void escapeString(StringPiece input,
616 const serialization_opts& opts) {
617 auto hexDigit = [] (int c) -> char {
618 return c < 10 ? c + '0' : c - 10 + 'a';
621 out.reserve(out.size() + input.size() + 2);
624 auto* p = reinterpret_cast<const unsigned char*>(input.begin());
625 auto* q = reinterpret_cast<const unsigned char*>(input.begin());
626 auto* e = reinterpret_cast<const unsigned char*>(input.end());
629 // Since non-ascii encoding inherently does utf8 validation
630 // we explicitly validate utf8 only if non-ascii encoding is disabled.
631 if ((opts.validate_utf8 || opts.skip_invalid_utf8)
632 && !opts.encode_non_ascii) {
633 // to achieve better spatial and temporal coherence
634 // we do utf8 validation progressively along with the
635 // string-escaping instead of two separate passes
637 // as the encoding progresses, q will stay at or ahead of p
640 // as p catches up with q, move q forward
642 // calling utf8_decode has the side effect of
643 // checking that utf8 encodings are valid
644 char32_t v = decodeUtf8(q, e, opts.skip_invalid_utf8);
645 if (opts.skip_invalid_utf8 && v == U'\ufffd') {
646 out.append("\ufffd");
652 if (opts.encode_non_ascii && (*p & 0x80)) {
653 // note that this if condition captures utf8 chars
654 // with value > 127, so size > 1 byte
655 char32_t v = decodeUtf8(p, e, opts.skip_invalid_utf8);
657 out.push_back(hexDigit(v >> 12));
658 out.push_back(hexDigit((v >> 8) & 0x0f));
659 out.push_back(hexDigit((v >> 4) & 0x0f));
660 out.push_back(hexDigit(v & 0x0f));
661 } else if (*p == '\\' || *p == '\"') {
664 } else if (*p <= 0x1f) {
666 case '\b': out.append("\\b"); p++; break;
667 case '\f': out.append("\\f"); p++; break;
668 case '\n': out.append("\\n"); p++; break;
669 case '\r': out.append("\\r"); p++; break;
670 case '\t': out.append("\\t"); p++; break;
672 // note that this if condition captures non readable chars
673 // with value < 32, so size = 1 byte (e.g control chars).
675 out.push_back(hexDigit((*p & 0xf0) >> 4));
676 out.push_back(hexDigit(*p & 0xf));
687 fbstring stripComments(StringPiece jsonC) {
694 } state = State::None;
696 for (size_t i = 0; i < jsonC.size(); ++i) {
697 auto s = jsonC.subpiece(i);
700 if (s.startsWith("/*")) {
701 state = State::InlineComment;
704 } else if (s.startsWith("//")) {
705 state = State::LineComment;
708 } else if (s[0] == '\"') {
709 state = State::InString;
711 result.push_back(s[0]);
713 case State::InString:
715 if (UNLIKELY(s.size() == 1)) {
716 throw std::logic_error("Invalid JSONC: string is not terminated");
718 result.push_back(s[0]);
719 result.push_back(s[1]);
722 } else if (s[0] == '\"') {
725 result.push_back(s[0]);
727 case State::InlineComment:
728 if (s.startsWith("*/")) {
733 case State::LineComment:
735 // skip the line break. It doesn't matter.
740 throw std::logic_error("Unknown comment state");
748 //////////////////////////////////////////////////////////////////////
750 dynamic parseJson(StringPiece range) {
751 return parseJson(range, json::serialization_opts());
756 json::serialization_opts const& opts) {
758 json::Input in(range, &opts);
760 auto ret = parseValue(in);
762 if (in.size() && *in != '\0') {
763 in.error("parsing didn't consume all input");
768 fbstring toJson(dynamic const& dyn) {
769 return json::serialize(dyn, json::serialization_opts());
772 fbstring toPrettyJson(dynamic const& dyn) {
773 json::serialization_opts opts;
774 opts.pretty_formatting = true;
775 return json::serialize(dyn, opts);
778 //////////////////////////////////////////////////////////////////////
779 // dynamic::print_as_pseudo_json() is implemented here for header
780 // ordering reasons (most of the dynamic implementation is in
781 // dynamic-inl.h, which we don't want to include json.h).
783 void dynamic::print_as_pseudo_json(std::ostream& out) const {
784 json::serialization_opts opts;
785 opts.allow_non_string_keys = true;
786 opts.allow_nan_inf = true;
787 out << json::serialize(*this, opts);
790 //////////////////////////////////////////////////////////////////////