2 * Copyright 2012 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/Range.h"
23 #include "folly/Unicode.h"
24 #include "folly/Conv.h"
28 //////////////////////////////////////////////////////////////////////
33 char32_t decodeUtf8(const char*& p, const char* const e) {
34 /* The following encodings are valid, except for the 5 and 6 byte
38 * 1110xxxx 10xxxxxx 10xxxxxx
39 * 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx
40 * 111110xx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
41 * 1111110x 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx 10xxxxxx
45 throw std::runtime_error("folly::decodeUtf8 empty/invalid string");
48 unsigned char fst = *p;
54 static const uint32_t bitMask[] = {
61 // upper control bits are masked out later
64 if ((fst & 0xC0) != 0xC0) {
65 throw std::runtime_error(
66 to<std::string>("folly::decodeUtf8 i=0 d=", d));
71 for (unsigned int i = 1; i != 3 && p + i < e; ++i) {
72 unsigned char tmp = p[i];
74 if ((tmp & 0xC0) != 0x80) {
75 throw std::runtime_error(
76 to<std::string>("folly::decodeUtf8 i=", i, " tmp=", (uint32_t)tmp));
79 d = (d << 6) | (tmp & 0x3F);
85 // overlong, could have been encoded with i bytes
86 if ((d & ~bitMask[i - 1]) == 0) {
87 throw std::runtime_error(
88 to<std::string>("folly::decodeUtf8 i=", i, " d=", d));
91 // check for surrogates only needed for 3 bytes
93 if ((d >= 0xD800 && d <= 0xDFFF) || d > 0x10FFFF) {
94 throw std::runtime_error(
95 to<std::string>("folly::decodeUtf8 i=", i, " d=", d));
104 throw std::runtime_error("folly::decodeUtf8 encoding length maxed out");
107 // Escape a string so that it is legal to print it in JSON text.
108 void escapeString(StringPiece input,
110 const serialization_opts& opts) {
111 auto hexDigit = [] (int c) -> char {
112 return c < 10 ? c + '0' : c - 10 + 'a';
115 out.reserve(out.size() + input.size() + 2);
118 const char* p = input.begin();
119 const char* q = input.begin();
120 const char* const e = input.end();
123 // Since non-ascii encoding inherently does utf8 validation
124 // we explicitly validate utf8 only if non-ascii encoding is disabled.
125 if (opts.validate_utf8 && !opts.encode_non_ascii) {
126 // to achieve better spatial and temporal coherence
127 // we do utf8 validation progressively along with the
128 // string-escaping instead of two separate passes
130 // as the encoding progresses, q will stay at or ahead of p
133 // as p catches up with q, move q forward
135 // calling utf8_decode has the side effect of
136 // checking that utf8 encodings are valid
141 if (opts.encode_non_ascii && (*p & 0x80)) {
142 char32_t v = decodeUtf8(p, e);
144 out.push_back(hexDigit(v >> 12));
145 out.push_back(hexDigit((v >> 8) & 0x0f));
146 out.push_back(hexDigit((v >> 4) & 0x0f));
147 out.push_back(hexDigit(v & 0x0f));
148 } else if (*p == '\\' || *p == '\"') {
151 } else if (*p <= 0x1f) {
153 case '\b': out.append("\\b"); p++; break;
154 case '\f': out.append("\\f"); p++; break;
155 case '\n': out.append("\\n"); p++; break;
156 case '\r': out.append("\\r"); p++; break;
157 case '\t': out.append("\\t"); p++; break;
159 // note that this if condition captures both control characters
160 // and extended ascii characters
162 out.push_back(hexDigit((*p & 0xf0) >> 4));
163 out.push_back(hexDigit(*p & 0xf));
175 explicit Printer(fbstring& out,
176 unsigned* indentLevel,
177 serialization_opts const* opts)
179 , indentLevel_(indentLevel)
183 void operator()(dynamic const& v) const {
185 case dynamic::DOUBLE:
186 toAppend(v.asDouble(), &out_);
188 case dynamic::INT64: {
189 auto intval = v.asInt();
190 if (opts_.javascript_safe) {
191 // Use folly::to to check that this integer can be represented
192 // as a double without loss of precision.
193 intval = int64_t(to<double>(intval));
195 toAppend(intval, &out_);
199 out_ += v.asBool() ? "true" : "false";
204 case dynamic::STRING:
205 escapeString(v.asString(), out_, opts_);
207 case dynamic::OBJECT:
214 CHECK(0) << "Bad type " << v.type();
219 void printKV(const std::pair<dynamic, dynamic>& p) const {
220 if (!opts_.allow_non_string_keys && !p.first.isString()) {
221 throw std::runtime_error("folly::toJson: JSON object key was not a "
229 void printObject(dynamic const& o) const {
238 auto it = o.items().begin();
240 for (++it; it != o.items().end(); ++it) {
250 void printArray(dynamic const& a) const {
260 for (auto& val : makeRange(boost::next(a.begin()), a.end())) {
271 void outdent() const {
277 void indent() const {
283 void newline() const {
285 out_ += to<fbstring>('\n', fbstring(*indentLevel_ * 2, ' '));
289 void mapColon() const {
290 out_ += indentLevel_ ? " : " : ":";
295 unsigned* const indentLevel_;
296 serialization_opts const& opts_;
299 //////////////////////////////////////////////////////////////////////
301 struct ParseError : std::runtime_error {
302 explicit ParseError(int line)
303 : std::runtime_error(to<std::string>("json parse error on line ", line))
306 explicit ParseError(int line, std::string const& context,
307 std::string const& expected)
308 : std::runtime_error(to<std::string>("json parse error on line ", line,
309 !context.empty() ? to<std::string>(" near `", context, '\'')
314 explicit ParseError(std::string const& what)
315 : std::runtime_error("json parse error: " + what)
319 // Wraps our input buffer with some helper functions.
321 explicit Input(StringPiece range)
328 Input(Input const&) = delete;
329 Input& operator=(Input const&) = delete;
331 char const* begin() const { return range_.begin(); }
333 // Parse ahead for as long as the supplied predicate is satisfied,
334 // returning a range of what was skipped.
335 template<class Predicate>
336 StringPiece skipWhile(const Predicate& p) {
337 std::size_t skipped = 0;
338 for (; skipped < range_.size(); ++skipped) {
339 if (!p(range_[skipped])) {
342 if (range_[skipped] == '\n') {
346 auto ret = range_.subpiece(0, skipped);
347 range_.advance(skipped);
352 StringPiece skipDigits() {
353 return skipWhile([] (char c) { return c >= '0' && c <= '9'; });
356 void skipWhitespace() {
357 // Spaces other than ' ' characters are less common but should be
358 // checked. This configuration where we loop on the ' '
359 // separately from oddspaces was empirically fastest.
360 auto oddspace = [] (char c) {
361 return c == '\n' || c == '\t' || c == '\r';
365 for (; !range_.empty() && range_.front() == ' '; range_.pop_front()) {
367 if (!range_.empty() && oddspace(range_.front())) {
374 void expect(char c) {
376 throw ParseError(lineNum_, context(),
377 to<std::string>("expected '", c, '\''));
382 std::size_t size() const {
383 return range_.size();
386 int operator*() const {
398 return to<T>(&range_);
399 } catch (std::exception const& e) {
404 bool consume(StringPiece str) {
405 if (boost::starts_with(range_, str)) {
406 range_.advance(str.size());
413 std::string context() const {
414 return range_.subpiece(0, 16 /* arbitrary */).toString();
417 dynamic error(char const* what) const {
418 throw ParseError(lineNum_, context(), what);
422 void storeCurrent() {
423 current_ = range_.empty() ? EOF : range_.front();
432 dynamic parseValue(Input& in);
433 fbstring parseString(Input& in);
435 dynamic parseObject(Input& in) {
439 dynamic ret = dynamic::object;
449 in.error("expected string for object key name");
451 auto key = parseString(in);
455 ret.insert(std::move(key), parseValue(in));
468 dynamic parseArray(Input& in) {
481 ret.push_back(parseValue(in));
494 dynamic parseNumber(Input& in) {
495 bool const negative = (*in == '-');
498 if (in.consume("Infinity")) {
499 return -std::numeric_limits<double>::infinity();
503 auto integral = in.skipDigits();
504 if (integral.empty()) {
505 in.error("expected digits after `-'");
507 auto const wasE = *in == 'e' || *in == 'E';
508 if (*in != '.' && !wasE) {
509 auto val = to<int64_t>(integral);
517 auto end = !wasE ? (++in, in.skipDigits().end()) : in.begin();
518 if (*in == 'e' || *in == 'E') {
520 if (*in == '+' || *in == '-') {
523 auto expPart = in.skipDigits();
526 auto fullNum = makeRange(integral.begin(), end);
528 auto val = to<double>(fullNum);
535 fbstring decodeUnicodeEscape(Input& in) {
536 auto hexVal = [&] (char c) -> unsigned {
537 return c >= '0' && c <= '9' ? c - '0' :
538 c >= 'a' && c <= 'f' ? c - 'a' + 10 :
539 c >= 'A' && c <= 'F' ? c - 'A' + 10 :
540 (in.error("invalid hex digit"), 0);
545 in.error("expected 4 hex digits");
548 uint16_t ret = hexVal(*in) * 4096;
550 ret += hexVal(*in) * 256;
552 ret += hexVal(*in) * 16;
560 * If the value encoded is in the surrogate pair range, we need to
561 * make sure there is another escape that we can use also.
563 uint32_t codePoint = readHex();
564 if (codePoint >= 0xd800 && codePoint <= 0xdbff) {
565 if (!in.consume("\\u")) {
566 in.error("expected another unicode escape for second half of "
569 uint16_t second = readHex();
570 if (second >= 0xdc00 && second <= 0xdfff) {
571 codePoint = 0x10000 + ((codePoint & 0x3ff) << 10) +
574 in.error("second character in surrogate pair is invalid");
576 } else if (codePoint >= 0xdc00 && codePoint <= 0xdfff) {
577 in.error("invalid unicode code point (in range [0xdc00,0xdfff])");
580 return codePointToUtf8(codePoint);
583 fbstring parseString(Input& in) {
589 auto range = in.skipWhile(
590 [] (char c) { return c != '\"' && c != '\\'; }
592 ret.append(range.begin(), range.end());
601 case '\"': ret.push_back('\"'); ++in; break;
602 case '\\': ret.push_back('\\'); ++in; break;
603 case '/': ret.push_back('/'); ++in; break;
604 case 'b': ret.push_back('\b'); ++in; break;
605 case 'f': ret.push_back('\f'); ++in; break;
606 case 'n': ret.push_back('\n'); ++in; break;
607 case 'r': ret.push_back('\r'); ++in; break;
608 case 't': ret.push_back('\t'); ++in; break;
609 case 'u': ++in; ret += decodeUnicodeEscape(in); break;
610 default: in.error(to<fbstring>("unknown escape ", *in,
611 " in string").c_str());
616 in.error("unterminated string");
620 * Apparently we're actually supposed to ban all control
621 * characters from strings. This seems unnecessarily
622 * restrictive, so we're only banning zero bytes. (Since the
623 * string is presumed to be UTF-8 encoded it's fine to just
626 in.error("null byte in string");
636 dynamic parseValue(Input& in) {
638 return *in == '[' ? parseArray(in) :
639 *in == '{' ? parseObject(in) :
640 *in == '\"' ? parseString(in) :
641 (*in == '-' || (*in >= '0' && *in <= '9')) ? parseNumber(in) :
642 in.consume("true") ? true :
643 in.consume("false") ? false :
644 in.consume("null") ? nullptr :
645 in.consume("Infinity") ? std::numeric_limits<double>::infinity() :
646 in.consume("NaN") ? std::numeric_limits<double>::quiet_NaN() :
647 in.error("expected json value");
652 //////////////////////////////////////////////////////////////////////
654 fbstring serialize(dynamic const& dyn, serialization_opts const& opts) {
656 unsigned indentLevel = 0;
657 Printer p(ret, opts.pretty_formatting ? &indentLevel : nullptr, &opts);
664 //////////////////////////////////////////////////////////////////////
666 dynamic parseJson(StringPiece range) {
667 json::Input in(range);
669 auto ret = parseValue(in);
671 if (*in != '\0' && in.size()) {
672 in.error("parsing didn't consume all input");
677 fbstring toJson(dynamic const& dyn) {
678 return json::serialize(dyn, json::serialization_opts());
681 fbstring toPrettyJson(dynamic const& dyn) {
682 json::serialization_opts opts;
683 opts.pretty_formatting = true;
684 return json::serialize(dyn, opts);
687 //////////////////////////////////////////////////////////////////////
688 // dynamic::print_as_pseudo_json() is implemented here for header
689 // ordering reasons (most of the dynamic implementation is in
690 // dynamic-inl.h, which we don't want to include json.h).
692 void dynamic::print_as_pseudo_json(std::ostream& out) const {
693 json::serialization_opts opts;
694 opts.allow_non_string_keys = true;
695 out << json::serialize(*this, opts);
698 //////////////////////////////////////////////////////////////////////