From d9dd84487bda8303590d5b30987f1d76b93867ba Mon Sep 17 00:00:00 2001 From: Boris Kolpackov Date: Fri, 30 Sep 2022 13:31:25 +0200 Subject: Add JSON parser (copy of libstud-json) --- libbutl/buildfile | 4 +- libbutl/json/parser.cxx | 502 +++++++++++++++++++++++ libbutl/json/parser.hxx | 408 +++++++++++++++++++ libbutl/json/parser.ixx | 216 ++++++++++ libbutl/json/pdjson.c | 1020 +++++++++++++++++++++++++++++++++++++++++++++++ libbutl/json/pdjson.h | 147 +++++++ 6 files changed, 2295 insertions(+), 2 deletions(-) create mode 100644 libbutl/json/parser.cxx create mode 100644 libbutl/json/parser.hxx create mode 100644 libbutl/json/parser.ixx create mode 100644 libbutl/json/pdjson.c create mode 100644 libbutl/json/pdjson.h (limited to 'libbutl') diff --git a/libbutl/buildfile b/libbutl/buildfile index 6c490af..ba4ad96 100644 --- a/libbutl/buildfile +++ b/libbutl/buildfile @@ -29,9 +29,9 @@ lib{butl}: {hxx ixx cxx}{win32-utility}: include = $windows lib{butl}: hxx{mingw-*}: include = $mingw_stdthread # Our C-files are always included into C++-files that wrap the corresponding -# API so treat them as files exclude from the compilation. +# API so treat them as files to exclude from the compilation. # -lib{butl}: file{*.c *.h} +lib{butl}: file{**.c **.h} # Platform-specific UUID implementations. # diff --git a/libbutl/json/parser.cxx b/libbutl/json/parser.cxx new file mode 100644 index 0000000..fa8916b --- /dev/null +++ b/libbutl/json/parser.cxx @@ -0,0 +1,502 @@ +#define PDJSON_SYMEXPORT static // See below. + +#include + +#include + +// There is an issue (segfault) with using std::current_exception() and +// std::rethrow_exception() with older versions of libc++ on Linux. While the +// exact root cause hasn't been determined, the suspicion is that something +// gets messed up if we "smuggle" std::exception_ptr through extern "C" call +// frames (we cannot even destroy such an exception without a segfault). We +// also could not determine in which version exactly this has been fixed but +// we know that libc++ 6.0.0 doesn't appear to have this issue (though we are +// not entirely sure the issue is (only) in libc++; libgcc_s could also be +// involved). +// +// The workaround is to just catch (and note) the exception and then throw a +// new instance of generic std::istream::failure. In order not to drag the +// below test into the header, we wrap exception_ptr with optional<> and use +// NULL to indicate the presence of the exception when the workaround is +// required. +// +// Note that if/when we drop this workaround, we should also get rid of +// optional<> in stream::exception member. +// +#undef LIBBUTL_JSON_NO_EXCEPTION_PTR + +#if defined (__linux__) && defined(__clang__) +# if __has_include(<__config>) +# include <__config> // _LIBCPP_VERSION +# if _LIBCPP_VERSION < 6000 +# define LIBBUTL_JSON_NO_EXCEPTION_PTR 1 +# endif +# endif +#endif + +namespace butl +{ + namespace json + { + using namespace std; + + parser:: + ~parser () + { + json_close (impl_); + } + + static int + stream_get (void* x) + { + auto& s (*static_cast (x)); + + // In the multi-value mode reading of whitespaces/separators is split + // between our code and pdjson's. As a result, these functions may end + // up being called more than once after EOF is reached. Which is + // something iostream does not handle gracefully. + // + if (!s.is->eof ()) + { + try + { + // We first peek not to trip failbit on EOF. + // + if (s.is->peek () != istream::traits_type::eof ()) + return static_cast (s.is->get ()); + } + catch (...) + { +#ifndef LIBBUTL_JSON_NO_EXCEPTION_PTR + s.exception = current_exception (); +#else + s.exception = nullptr; +#endif + } + } + + return EOF; + } + + static int + stream_peek (void* x) + { + auto& s (*static_cast (x)); + + if (!s.is->eof ()) + { + try + { + auto c (s.is->peek ()); + if (c != istream::traits_type::eof ()) + return static_cast (c); + } + catch (...) + { +#ifndef LIBBUTL_JSON_NO_EXCEPTION_PTR + s.exception = current_exception (); +#else + s.exception = nullptr; +#endif + } + } + + return EOF; + } + + // NOTE: watch out for exception safety (specifically, doing anything that + // might throw after opening the stream). + // + parser:: + parser (istream& is, const char* n, bool mv, const char* sep) noexcept + : input_name (n), + stream_ {&is, nullopt}, + multi_value_ (mv), + separators_ (sep), + raw_s_ (nullptr), + raw_n_ (0) + { + json_open_user (impl_, &stream_get, &stream_peek, &stream_); + json_set_streaming (impl_, multi_value_); + } + + parser:: + parser (const void* t, + size_t s, + const char* n, + bool mv, + const char* sep) noexcept + : input_name (n), + stream_ {nullptr, nullopt}, + multi_value_ (mv), + separators_ (sep), + raw_s_ (nullptr), + raw_n_ (0) + { + json_open_buffer (impl_, t, s); + json_set_streaming (impl_, multi_value_); + } + + optional parser:: + next () + { + name_p_ = value_p_ = location_p_ = false; + + // Note that for now we don't worry about the state of the parser if + // next_impl() throws assuming it is not going to be reused. + // + if (peeked_) + { + parsed_ = peeked_; + peeked_ = nullopt; + } + else + parsed_ = next_impl (); + + return translate (*parsed_); + } + + optional parser:: + peek () + { + if (!peeked_) + { + if (parsed_) + { + cache_parsed_data (); + cache_parsed_location (); + } + peeked_ = next_impl (); + } + return translate (*peeked_); + } + + std::uint64_t parser:: + line () const noexcept + { + if (!location_p_) + { + if (!parsed_) + return 0; + + assert (!peeked_); + + return static_cast ( + json_get_lineno (const_cast (impl_))); + } + + return line_; + } + + std::uint64_t parser:: + column () const noexcept + { + if (!location_p_) + { + if (!parsed_) + return 0; + + assert (!peeked_); + + return static_cast ( + json_get_column (const_cast (impl_))); + } + + return column_; + } + + std::uint64_t parser:: + position () const noexcept + { + if (!location_p_) + { + if (!parsed_) + return 0; + + assert (!peeked_); + + return static_cast ( + json_get_position (const_cast (impl_))); + } + + return position_; + } + + json_type parser:: + next_impl () + { + raw_s_ = nullptr; + raw_n_ = 0; + json_type e; + + // Read characters between values skipping required separators and JSON + // whitespaces. Return whether a required separator was encountered as + // well as the first non-separator/whitespace character (which, if EOF, + // should trigger a check for input/output errors). + // + // Note that the returned non-separator will not have been extracted + // from the input (so position, column, etc. will still refer to its + // predecessor). + // + auto skip_separators = [this] () -> pair + { + bool r (separators_ == nullptr); + + int c; + for (; (c = json_source_peek (impl_)) != EOF; json_source_get (impl_)) + { + // User separator. + // + if (separators_ != nullptr && *separators_ != '\0') + { + if (strchr (separators_, c) != nullptr) + { + r = true; + continue; + } + } + + // JSON separator. + // + if (json_isspace (c)) + { + if (separators_ != nullptr && *separators_ == '\0') + r = true; + + continue; + } + + break; + } + + return make_pair (r, c); + }; + + // In the multi-value mode skip any instances of required separators + // (and any other JSON whitespace) preceding the first JSON value. + // + if (multi_value_ && !parsed_ && !peeked_) + { + if (skip_separators ().second == EOF && stream_.is != nullptr) + { + if (stream_.exception) goto fail_rethrow; + if (stream_.is->fail ()) goto fail_stream; + } + } + + e = json_next (impl_); + + // First check for a pending input/output error. + // + if (stream_.is != nullptr) + { + if (stream_.exception) goto fail_rethrow; + if (stream_.is->fail ()) goto fail_stream; + } + + // There are two ways to view separation between two values: as following + // the first value or as preceding the second value. And one aspect that + // is determined by this is whether a separation violation is a problem + // with the first value or with the second, which becomes important if + // the user bails out before parsing the second value. + // + // Consider these two unseparated value (yes, in JSON they are two + // values, leading zeros are not allowed in JSON numbers): + // + // 01 + // + // If the user bails out after parsing 0 in a stream that should have + // been newline-delimited, they most likely would want to get an error + // since this is most definitely an invalid value rather than two + // values that are not properly separated. So in this light we handle + // separators at the end of the first value. + // + switch (e) + { + case JSON_DONE: + { + // Deal with the following value separators. + // + // Note that we must not do this for the second JSON_DONE (or the + // first one in case there are no values) that signals the end of + // input. + // + if (multi_value_ && + (parsed_ || peeked_) && + (peeked_ ? *peeked_ : *parsed_) != JSON_DONE) + { + auto p (skip_separators ()); + + if (p.second == EOF && stream_.is != nullptr) + { + if (stream_.exception) goto fail_rethrow; + if (stream_.is->fail ()) goto fail_stream; + } + + // Note that we don't require separators after the last value. + // + if (!p.first && p.second != EOF) + { + json_source_get (impl_); // Consume to update column number. + goto fail_separation; + } + + json_reset (impl_); + } + break; + } + case JSON_ERROR: goto fail_json; + case JSON_STRING: + case JSON_NUMBER: + raw_s_ = json_get_string (impl_, &raw_n_); + raw_n_--; // Includes terminating `\0`. + break; + case JSON_TRUE: raw_s_ = "true"; raw_n_ = 4; break; + case JSON_FALSE: raw_s_ = "false"; raw_n_ = 5; break; + case JSON_NULL: raw_s_ = "null"; raw_n_ = 4; break; + default: break; + } + + return e; + + fail_json: + throw invalid_json_input ( + input_name != nullptr ? input_name : "", + static_cast (json_get_lineno (impl_)), + static_cast (json_get_column (impl_)), + static_cast (json_get_position (impl_)), + json_get_error (impl_)); + + fail_separation: + throw invalid_json_input ( + input_name != nullptr ? input_name : "", + static_cast (json_get_lineno (impl_)), + static_cast (json_get_column (impl_)), + static_cast (json_get_position (impl_)), + "missing separator between JSON values"); + + fail_stream: + throw invalid_json_input ( + input_name != nullptr ? input_name : "", + static_cast (json_get_lineno (impl_)), + static_cast (json_get_column (impl_)), + static_cast (json_get_position (impl_)), + "unable to read JSON input text"); + + fail_rethrow: +#ifndef LIBBUTL_JSON_NO_EXCEPTION_PTR + rethrow_exception (move (*stream_.exception)); +#else + throw istream::failure ("unable to read"); +#endif + } + + optional parser:: + translate (json_type e) const noexcept + { + switch (e) + { + case JSON_DONE: return nullopt; + case JSON_OBJECT: return event::begin_object; + case JSON_OBJECT_END: return event::end_object; + case JSON_ARRAY: return event::begin_array; + case JSON_ARRAY_END: return event::end_array; + case JSON_STRING: + { + // This can be a value or, inside an object, a name from the + // name/value pair. + // + size_t n; + return json_get_context (const_cast (impl_), &n) == + JSON_OBJECT && + n % 2 == 1 + ? event::name + : event::string; + } + case JSON_NUMBER: return event::number; + case JSON_TRUE: return event::boolean; + case JSON_FALSE: return event::boolean; + case JSON_NULL: return event::null; + case JSON_ERROR: assert (false); // Should've been handled by caller. + } + + return nullopt; // Should never reach. + } + + void parser:: + cache_parsed_data () + { + name_p_ = value_p_ = false; + if (const optional e = translate (*parsed_)) + { + if (e == event::name) + { + name_.assign (raw_s_, raw_n_); + name_p_ = true; + } + else if (value_event (e)) + { + value_.assign (raw_s_, raw_n_); + value_p_ = true; + } + } + } + + void parser:: + cache_parsed_location () noexcept + { + line_ = static_cast (json_get_lineno (impl_)); + column_ = static_cast (json_get_column (impl_)); + position_ = static_cast (json_get_position (impl_)); + location_p_ = true; + } + + bool parser:: + value_event (optional e) noexcept + { + if (!e) + return false; + + switch (*e) + { + case event::string: + case event::number: + case event::boolean: + case event::null: + return true; + default: + return false; + } + } + + [[noreturn]] void parser:: + throw_invalid_value (const char* type, const char* v, size_t n) const + { + string d (string ("invalid ") + type + " value: '"); + d.append (v, n); + d += '\''; + + throw invalid_json_input (input_name != nullptr ? input_name : "", + line (), + column (), + position (), + move (d)); + } + } // namespace json +} // namespace butl + +// Include the implementation into our translation unit (instead of compiling +// it separately) to (hopefully) get function inlining without LTO. +// +// Let's keep it last since the implementation defines a couple of macros. +// +#if defined(__clang__) || defined(__GNUC__) +# pragma GCC diagnostic ignored "-Wunused-function" +#endif + +extern "C" +{ +#define PDJSON_STACK_INC 16 +#define PDJSON_STACK_MAX 2048 +#include "pdjson.c" +} diff --git a/libbutl/json/parser.hxx b/libbutl/json/parser.hxx new file mode 100644 index 0000000..241ca46 --- /dev/null +++ b/libbutl/json/parser.hxx @@ -0,0 +1,408 @@ +#pragma once + +#ifdef BUILD2_BOOTSTRAP +# error JSON parser not available during bootstrap +#endif + +#include +#include +#include // size_t +#include // uint64_t +#include // pair +#include // exception_ptr +#include // invalid_argument + +#include // butl::optional is std::optional or similar. + +#include + +#include // Implementation details. + +#include + +namespace butl +{ + // Using the RFC8259 terminology: JSON (input) text, JSON value, object + // member. + // + namespace json + { + class invalid_json_input: public std::invalid_argument + { + public: + std::string name; + std::uint64_t line; + std::uint64_t column; + std::uint64_t position; + + invalid_json_input (std::string name, + std::uint64_t line, + std::uint64_t column, + std::uint64_t position, + const std::string& description); + + invalid_json_input (std::string name, + std::uint64_t line, + std::uint64_t column, + std::uint64_t position, + const char* description); + }; + + class LIBBUTL_SYMEXPORT parser + { + public: + const char* input_name; + + // Construction. + // + + // Parse JSON input text from std::istream. + // + // The name argument is used to identify the input being parsed. Note + // that the stream, name, and separators are kept as references so they + // must outlive the parser instance. + // + // If stream exceptions are enabled then the std::ios_base::failure + // exception is used to report input/output errors (badbit and failbit). + // Otherwise, those are reported as the invalid_json_input exception. + // + // If multi_value is true, enable the multi-value mode in which case the + // input stream may contain multiple JSON values (more precisely, zero + // or more). If false (the default), parsing will fail unless there is + // exactly one JSON value in the input stream. + // + // If multi_value is true, the separators argument specifies the + // required separator characters between JSON values. At least one of + // them must be present between every pair of JSON values (in addition + // to any number of JSON whitespaces). No separators are required after + // the last JSON value (but any found will be skipped). + // + // Specifically, if it is NULL, then no separation is required (that is, + // both `{...}{...}` and `{...} {...}` would be valid). If it is empty, + // then at least one JSON whitespace is required. And if it is non- + // empty, then at least one of its characters must be present (for + // example, "\n\t" would require at least one newline or TAB character + // between JSON values). + // + // Note that a separator need not be valid JSON whitespace: any + // character is acceptable (though it probably shouldn't be an object, + // array, or string delimiter and should not occur within a non-self- + // delimited top-level value, such as `true`, `false`, `null`, or a + // number). All instances of required separators before and after a + // value are skipped. Therefore JSON Text Sequences (RFC 7464; AKA + // Record Separator-delimited JSON), which requires the RS (0x1E) + // character before each value, can be handled as well. + // + parser (std::istream&, + const std::string& name, + bool multi_value = false, + const char* separators = nullptr) noexcept; + + parser (std::istream&, + const char* name, + bool multi_value = false, + const char* separators = nullptr) noexcept; + + parser (std::istream&, + std::string&&, + bool = false, + const char* = nullptr) = delete; + + // Parse a memory buffer that contains the entire JSON input text. + // + // The name argument is used to identify the input being parsed. Note + // that the buffer, name, and separators are kept as references so they + // must outlive the parser instance. + // + parser (const void* text, + std::size_t size, + const std::string& name, + bool multi_value = false, + const char* separators = nullptr) noexcept; + + parser (const void* text, + std::size_t size, + const char* name, + bool multi_value = false, + const char* separators = nullptr) noexcept; + + parser (const void*, + std::size_t, + std::string&&, + bool = false, + const char* = nullptr) = delete; + + // Similar to the above but parse a string. + // + parser (const std::string& text, + const std::string& name, + bool multi_value = false, + const char* separators = nullptr) noexcept; + + parser (const std::string& text, + const char* name, + bool multi_value = false, + const char* separators = nullptr) noexcept; + + parser (const std::string&, + std::string&&, + bool = false, + const char* = nullptr) = delete; + + // Similar to the above but parse a C-string. + // + parser (const char* text, + const std::string& name, + bool multi_value = false, + const char* separators = nullptr) noexcept; + + parser (const char* text, + const char* name, + bool multi_value = false, + const char* separators = nullptr) noexcept; + + parser (const char*, + std::string&&, + bool = false, + const char* = nullptr) = delete; + + parser (parser&&) = delete; + parser (const parser&) = delete; + + parser& operator= (parser&&) = delete; + parser& operator= (const parser&) = delete; + + // Return the next event or nullopt if end of input is reached. + // + // In the single-value parsing mode (default) the parsing code could + // look like this: + // + // while (optional e = p.next ()) + // { + // switch (*e) + // { + // // ... + // } + // } + // + // In the multi-value mode the parser additionally returns nullopt after + // every JSON value parsed (so there will be two nullopt's after the + // last JSON value, the second indicating the end of input). + // + // One way to perform multi-value parsing is with the help of the peek() + // function (see below): + // + // while (p.peek ()) + // { + // while (optional e = p.next ()) + // { + // switch (*e) + // { + // //... + // } + // } + // } + // + // Note that while the single-value mode will always parse exactly one + // value, the multi-value mode will accept zero values in which case a + // single nullopt is returned. + // + optional + next (); + + // The range-based for loop support. + // + // In the single-value parsing mode (default) the parsing code could + // look like this: + // + // for (event e: p) + // { + // switch (e) + // { + // //... + // } + // } + // + // And in the multi-value mode (see next() for more information) like + // this: + // + // while (p.peek ()) + // { + // for (event e: p) + // { + // switch (e) + // { + // //... + // } + // } + // } + // + // Note that generally, the iterator interface doesn't make much sense + // for the parser so for now we have an implementation that is just + // enough for the range-based for. + // + struct iterator; + + iterator begin () {return iterator (this, next ());} + iterator end () {return iterator (nullptr, nullopt);} + + // Return the next event without considering it parsed. In other words, + // after this call, any subsequent calls to peek() and the next call to + // next() (if any) will all return the same event. + // + // Note that the name, value, and line corresponding to the peeked event + // are not accessible with name(), value() and line(); these functions + // will still return values corresponding to the most recent call to + // next(). The peeked values, however, can be accessed in the raw form + // using data(). + // + optional + peek (); + + // Event data. + // + + // Return the object member name. + // + const std::string& + name (); + + // Any value (string, number, boolean, and null) can be retrieved as a + // string. Calling this function after any non-value events is illegal. + // + // Note that the value is returned as a non-const string reference and + // you are allowed to move the value out of it. However, this should not + // be done unnecessarily or in cases where the small string optimization + // is likely since the string's buffer is reused to store subsequent + // values. + // + std::string& + value (); + + // Convert the value to an integer, floating point, or bool. Throw + // invalid_json_input if the conversion is impossible without a loss. + // + template + T + value () const; + + // Return the value or object member name in the raw form. + // + // Calling this function on non-value/name events is legal in which case + // NULL is returned. Note also that the returned data corresponds to the + // most recent event, whether peeked or parsed. + // + std::pair + data () const {return std::make_pair (raw_s_, raw_n_);} + + // Return the line number (1-based) corresponding to the most recently + // parsed event or 0 if nothing has been parsed yet. + // + std::uint64_t + line () const noexcept; + + // Return the column number (1-based) corresponding to the beginning of + // the most recently parsed event or 0 if nothing has been parsed yet. + // + std::uint64_t + column () const noexcept; + + // Return the position (byte offset) pointing immediately after the most + // recently parsed event or 0 if nothing has been parsed yet. + // + std::uint64_t + position () const noexcept; + + // Implementation details. + // + public: + struct iterator + { + using value_type = event; + + explicit + iterator (parser* p = nullptr, optional e = nullopt) + : p_ (p), e_ (e) {} + + event operator* () const {return *e_;} + iterator& operator++ () {e_ = p_->next (); return *this;} + + // Comparison only makes sense when comparing to end (eof). + // + bool operator== (iterator y) const {return !e_ && !y.e_;} + bool operator!= (iterator y) const {return !(*this == y);} + + private: + parser* p_; + optional e_; + }; + + struct stream + { + std::istream* is; + optional exception; + }; + + [[noreturn]] void + throw_invalid_value (const char* type, const char*, std::size_t) const; + + ~parser (); + + private: + // Functionality shared by next() and peek(). + // + json_type + next_impl (); + + // Translate the event produced by the most recent call to next_impl(). + // + // Note that the underlying parser state determines whether name or + // value is returned when translating JSON_STRING. + // + optional + translate (json_type) const noexcept; + + // Cache state (name/value) produced by the most recent call to + // next_impl(). + // + void + cache_parsed_data (); + + // Cache the location numbers as determined by the most recent call to + // next_impl(). + // + void + cache_parsed_location () noexcept; + + // Return true if this is a value event (string, number, boolean, or + // null). + // + static bool + value_event (optional) noexcept; + + stream stream_; + + bool multi_value_; + const char* separators_; + + // The *_p_ members indicate whether the value is present (cached). + // Note: not using optional not to reallocate the string's buffer. + // + std::string name_; bool name_p_ = false; + std::string value_; bool value_p_ = false; + std::uint64_t line_, column_, position_; bool location_p_ = false; + + optional parsed_; // Current parsed event if any. + optional peeked_; // Current peeked event if any. + + ::json_stream impl_[1]; + + // Cached raw value. + // + const char* raw_s_; + std::size_t raw_n_; + }; + } +} + +#include diff --git a/libbutl/json/parser.ixx b/libbutl/json/parser.ixx new file mode 100644 index 0000000..3f02a1e --- /dev/null +++ b/libbutl/json/parser.ixx @@ -0,0 +1,216 @@ +#include +#include // numeric_limits +#include // move() +#include +#include // strto*() +#include // enable_if, is_* +#include // strlen() + +namespace butl +{ + namespace json + { + inline invalid_json_input:: + invalid_json_input (std::string n, + std::uint64_t l, + std::uint64_t c, + std::uint64_t p, + const std::string& d) + : invalid_json_input (move (n), l, c, p, d.c_str ()) + { + } + + inline invalid_json_input:: + invalid_json_input (std::string n, + std::uint64_t l, + std::uint64_t c, + std::uint64_t p, + const char* d) + : invalid_argument (d), + name (std::move (n)), + line (l), column (c), position (p) + { + } + + inline parser:: + parser (std::istream& is, + const std::string& n, + bool mv, + const char* sep) noexcept + : parser (is, n.c_str (), mv, sep) + { + } + + inline parser:: + parser (const void* t, + std::size_t s, + const std::string& n, + bool mv, + const char* sep) noexcept + : parser (t, s, n.c_str (), mv, sep) + { + } + + inline parser:: + parser (const std::string& t, + const std::string& n, + bool mv, + const char* sep) noexcept + : parser (t.data (), t.size (), n.c_str (), mv, sep) + { + } + + inline parser:: + parser (const std::string& t, + const char* n, + bool mv, + const char* sep) noexcept + : parser (t.data (), t.size (), n, mv, sep) + { + } + + inline parser:: + parser (const char* t, + const std::string& n, + bool mv, + const char* sep) noexcept + : parser (t, std::strlen (t), n.c_str (), mv, sep) + { + } + + inline parser:: + parser (const char* t, + const char* n, + bool mv, + const char* sep) noexcept + : parser (t, std::strlen (t), n, mv, sep) + { + } + + inline const std::string& parser:: + name () + { + if (!name_p_) + { + assert (parsed_ && !peeked_ && !value_p_); + cache_parsed_data (); + assert (name_p_); + } + return name_; + } + + inline std::string& parser:: + value () + { + if (!value_p_) + { + assert (parsed_ && !peeked_ && !name_p_); + cache_parsed_data (); + assert (value_p_); + } + return value_; + } + + // Note: one day we will be able to use C++17 from_chars() which was made + // exactly for this. + // + template + inline typename std::enable_if::value, T>::type + parse_value (const char* b, size_t, const parser&) + { + return *b == 't'; + } + + template + inline typename std::enable_if< + std::is_integral::value && + std::is_signed::value && + !std::is_same::value, T>::type + parse_value (const char* b, size_t n, const parser& p) + { + char* e (nullptr); + errno = 0; // We must clear it according to POSIX. + std::int64_t v (strtoll (b, &e, 10)); // Can't throw. + + if (e == b || e != b + n || errno == ERANGE || + v < std::numeric_limits::min () || + v > std::numeric_limits::max ()) + p.throw_invalid_value ("signed integer", b, n); + + return static_cast (v); + } + + template + inline typename std::enable_if< + std::is_integral::value && + std::is_unsigned::value && + !std::is_same::value, T>::type + parse_value (const char* b, size_t n, const parser& p) + { + char* e (nullptr); + errno = 0; // We must clear it according to POSIX. + std::uint64_t v (strtoull (b, &e, 10)); // Can't throw. + + if (e == b || e != b + n || errno == ERANGE || + v > std::numeric_limits::max ()) + p.throw_invalid_value ("unsigned integer", b, n); + + return static_cast (v); + } + + template + inline typename std::enable_if::value, T>::type + parse_value (const char* b, size_t n, const parser& p) + { + char* e (nullptr); + errno = 0; // We must clear it according to POSIX. + T r (std::strtof (b, &e)); + + if (e == b || e != b + n || errno == ERANGE) + p.throw_invalid_value ("float", b, n); + + return r; + } + + template + inline typename std::enable_if::value, T>::type + parse_value (const char* b, size_t n, const parser& p) + { + char* e (nullptr); + errno = 0; // We must clear it according to POSIX. + T r (std::strtod (b, &e)); + + if (e == b || e != b + n || errno == ERANGE) + p.throw_invalid_value ("double", b, n); + + return r; + } + + template + inline typename std::enable_if::value, T>::type + parse_value (const char* b, size_t n, const parser& p) + { + char* e (nullptr); + errno = 0; // We must clear it according to POSIX. + T r (std::strtold (b, &e)); + + if (e == b || e != b + n || errno == ERANGE) + p.throw_invalid_value ("long double", b, n); + + return r; + } + + template + inline T parser:: + value () const + { + if (!value_p_) + { + assert (parsed_ && !peeked_ && value_event (translate (*parsed_))); + return parse_value (raw_s_, raw_n_, *this); + } + + return parse_value (value_.data (), value_.size (), *this); + } + } +} diff --git a/libbutl/json/pdjson.c b/libbutl/json/pdjson.c new file mode 100644 index 0000000..279d169 --- /dev/null +++ b/libbutl/json/pdjson.c @@ -0,0 +1,1020 @@ +#ifndef _POSIX_C_SOURCE +# define _POSIX_C_SOURCE 200112L +#elif _POSIX_C_SOURCE < 200112L +# error incompatible _POSIX_C_SOURCE level +#endif + +#include +#include +#include + +#ifndef PDJSON_H +# include "pdjson.h" +#endif + +#define JSON_FLAG_ERROR (1u << 0) +#define JSON_FLAG_STREAMING (1u << 1) + +#if defined(_MSC_VER) && (_MSC_VER < 1900) + +#define json_error(json, format, ...) \ + if (!(json->flags & JSON_FLAG_ERROR)) { \ + json->flags |= JSON_FLAG_ERROR; \ + _snprintf_s(json->errmsg, sizeof(json->errmsg), \ + _TRUNCATE, \ + format, \ + __VA_ARGS__); \ + } \ + +#else + +#define json_error(json, format, ...) \ + if (!(json->flags & JSON_FLAG_ERROR)) { \ + json->flags |= JSON_FLAG_ERROR; \ + snprintf(json->errmsg, sizeof(json->errmsg), \ + format, \ + __VA_ARGS__); \ + } \ + +#endif /* _MSC_VER */ + +/* See also PDJSON_STACK_MAX below. */ +#ifndef PDJSON_STACK_INC +# define PDJSON_STACK_INC 4 +#endif + +struct json_stack { + enum json_type type; + long count; +}; + +static enum json_type +push(json_stream *json, enum json_type type) +{ + json->stack_top++; + +#ifdef PDJSON_STACK_MAX + if (json->stack_top > PDJSON_STACK_MAX) { + json_error(json, "%s", "maximum depth of nesting reached"); + return JSON_ERROR; + } +#endif + + if (json->stack_top >= json->stack_size) { + struct json_stack *stack; + size_t size = (json->stack_size + PDJSON_STACK_INC) * sizeof(*json->stack); + stack = (struct json_stack *)json->alloc.realloc(json->stack, size); + if (stack == NULL) { + json_error(json, "%s", "out of memory"); + return JSON_ERROR; + } + + json->stack_size += PDJSON_STACK_INC; + json->stack = stack; + } + + json->stack[json->stack_top].type = type; + json->stack[json->stack_top].count = 0; + + return type; +} + +static enum json_type +pop(json_stream *json, int c, enum json_type expected) +{ + if (json->stack == NULL || json->stack[json->stack_top].type != expected) { + json_error(json, "unexpected byte '%c'", c); + return JSON_ERROR; + } + json->stack_top--; + return expected == JSON_ARRAY ? JSON_ARRAY_END : JSON_OBJECT_END; +} + +static int buffer_peek(struct json_source *source) +{ + if (source->position < source->source.buffer.length) + return source->source.buffer.buffer[source->position]; + else + return EOF; +} + +static int buffer_get(struct json_source *source) +{ + int c = source->peek(source); + if (c != EOF) + source->position++; + return c; +} + +static int stream_get(struct json_source *source) +{ + int c = fgetc(source->source.stream.stream); + if (c != EOF) + source->position++; + return c; +} + +static int stream_peek(struct json_source *source) +{ + int c = fgetc(source->source.stream.stream); + ungetc(c, source->source.stream.stream); + return c; +} + +static void init(json_stream *json) +{ + json->lineno = 1; + json->linepos = 0; + json->lineadj = 0; + json->linecon = 0; + json->colno = 0; + json->flags = JSON_FLAG_STREAMING; + json->errmsg[0] = '\0'; + json->ntokens = 0; + json->next = (enum json_type)0; + + json->stack = NULL; + json->stack_top = -1; + json->stack_size = 0; + + json->data.string = NULL; + json->data.string_size = 0; + json->data.string_fill = 0; + json->source.position = 0; + + json->alloc.malloc = malloc; + json->alloc.realloc = realloc; + json->alloc.free = free; +} + +static enum json_type +is_match(json_stream *json, const char *pattern, enum json_type type) +{ + int c; + for (const char *p = pattern; *p; p++) { + if (*p != (c = json->source.get(&json->source))) { + json_error(json, "expected '%c' instead of byte '%c'", *p, c); + return JSON_ERROR; + } + } + return type; +} + +static int pushchar(json_stream *json, int c) +{ + if (json->data.string_fill == json->data.string_size) { + size_t size = json->data.string_size * 2; + char *buffer = (char *)json->alloc.realloc(json->data.string, size); + if (buffer == NULL) { + json_error(json, "%s", "out of memory"); + return -1; + } else { + json->data.string_size = size; + json->data.string = buffer; + } + } + json->data.string[json->data.string_fill++] = c; + return 0; +} + +static int init_string(json_stream *json) +{ + json->data.string_fill = 0; + if (json->data.string == NULL) { + json->data.string_size = 1024; + json->data.string = (char *)json->alloc.malloc(json->data.string_size); + if (json->data.string == NULL) { + json_error(json, "%s", "out of memory"); + return -1; + } + } + json->data.string[0] = '\0'; + return 0; +} + +static int encode_utf8(json_stream *json, unsigned long c) +{ + if (c < 0x80UL) { + return pushchar(json, c); + } else if (c < 0x0800UL) { + return !((pushchar(json, (c >> 6 & 0x1F) | 0xC0) == 0) && + (pushchar(json, (c >> 0 & 0x3F) | 0x80) == 0)); + } else if (c < 0x010000UL) { + if (c >= 0xd800 && c <= 0xdfff) { + json_error(json, "invalid codepoint %06lx", c); + return -1; + } + return !((pushchar(json, (c >> 12 & 0x0F) | 0xE0) == 0) && + (pushchar(json, (c >> 6 & 0x3F) | 0x80) == 0) && + (pushchar(json, (c >> 0 & 0x3F) | 0x80) == 0)); + } else if (c < 0x110000UL) { + return !((pushchar(json, (c >> 18 & 0x07) | 0xF0) == 0) && + (pushchar(json, (c >> 12 & 0x3F) | 0x80) == 0) && + (pushchar(json, (c >> 6 & 0x3F) | 0x80) == 0) && + (pushchar(json, (c >> 0 & 0x3F) | 0x80) == 0)); + } else { + json_error(json, "unable to encode %06lx as UTF-8", c); + return -1; + } +} + +static int hexchar(int c) +{ + switch (c) { + case '0': return 0; + case '1': return 1; + case '2': return 2; + case '3': return 3; + case '4': return 4; + case '5': return 5; + case '6': return 6; + case '7': return 7; + case '8': return 8; + case '9': return 9; + case 'a': + case 'A': return 10; + case 'b': + case 'B': return 11; + case 'c': + case 'C': return 12; + case 'd': + case 'D': return 13; + case 'e': + case 'E': return 14; + case 'f': + case 'F': return 15; + default: + return -1; + } +} + +static long +read_unicode_cp(json_stream *json) +{ + long cp = 0; + int shift = 12; + + for (size_t i = 0; i < 4; i++) { + int c = json->source.get(&json->source); + int hc; + + if (c == EOF) { + json_error(json, "%s", "unterminated string literal in Unicode"); + return -1; + } else if ((hc = hexchar(c)) == -1) { + json_error(json, "invalid escape Unicode byte '%c'", c); + return -1; + } + + cp += hc * (1 << shift); + shift -= 4; + } + + + return cp; +} + +static int read_unicode(json_stream *json) +{ + long cp, h, l; + + if ((cp = read_unicode_cp(json)) == -1) { + return -1; + } + + if (cp >= 0xd800 && cp <= 0xdbff) { + /* This is the high portion of a surrogate pair; we need to read the + * lower portion to get the codepoint + */ + h = cp; + + int c = json->source.get(&json->source); + if (c == EOF) { + json_error(json, "%s", "unterminated string literal in Unicode"); + return -1; + } else if (c != '\\') { + json_error(json, "invalid continuation for surrogate pair '%c', " + "expected '\\'", c); + return -1; + } + + c = json->source.get(&json->source); + if (c == EOF) { + json_error(json, "%s", "unterminated string literal in Unicode"); + return -1; + } else if (c != 'u') { + json_error(json, "invalid continuation for surrogate pair '%c', " + "expected 'u'", c); + return -1; + } + + if ((l = read_unicode_cp(json)) == -1) { + return -1; + } + + if (l < 0xdc00 || l > 0xdfff) { + json_error(json, "surrogate pair continuation \\u%04lx out " + "of range (dc00-dfff)", l); + return -1; + } + + cp = ((h - 0xd800) * 0x400) + ((l - 0xdc00) + 0x10000); + } else if (cp >= 0xdc00 && cp <= 0xdfff) { + json_error(json, "dangling surrogate \\u%04lx", cp); + return -1; + } + + return encode_utf8(json, cp); +} + +static int +read_escaped(json_stream *json) +{ + int c = json->source.get(&json->source); + if (c == EOF) { + json_error(json, "%s", "unterminated string literal in escape"); + return -1; + } else if (c == 'u') { + if (read_unicode(json) != 0) + return -1; + } else { + switch (c) { + case '\\': + case 'b': + case 'f': + case 'n': + case 'r': + case 't': + case '/': + case '"': + { + const char *codes = "\\bfnrt/\""; + const char *p = strchr(codes, c); + if (pushchar(json, "\\\b\f\n\r\t/\""[p - codes]) != 0) + return -1; + } + break; + default: + json_error(json, "invalid escaped byte '%c'", c); + return -1; + } + } + return 0; +} + +static int +char_needs_escaping(int c) +{ + if ((c >= 0) && (c < 0x20 || c == 0x22 || c == 0x5c)) { + return 1; + } + + return 0; +} + +static int +utf8_seq_length(char byte) +{ + unsigned char u = (unsigned char) byte; + if (u < 0x80) return 1; + + if (0x80 <= u && u <= 0xBF) + { + // second, third or fourth byte of a multi-byte + // sequence, i.e. a "continuation byte" + return 0; + } + else if (u == 0xC0 || u == 0xC1) + { + // overlong encoding of an ASCII byte + return 0; + } + else if (0xC2 <= u && u <= 0xDF) + { + // 2-byte sequence + return 2; + } + else if (0xE0 <= u && u <= 0xEF) + { + // 3-byte sequence + return 3; + } + else if (0xF0 <= u && u <= 0xF4) + { + // 4-byte sequence + return 4; + } + else + { + // u >= 0xF5 + // Restricted (start of 4-, 5- or 6-byte sequence) or invalid UTF-8 + return 0; + } +} + +static int +is_legal_utf8(const unsigned char *bytes, int length) +{ + if (0 == bytes || 0 == length) return 0; + + unsigned char a; + const unsigned char* srcptr = bytes + length; + switch (length) + { + default: + return 0; + // Everything else falls through when true. + case 4: + if ((a = (*--srcptr)) < 0x80 || a > 0xBF) return 0; + /* FALLTHRU */ + case 3: + if ((a = (*--srcptr)) < 0x80 || a > 0xBF) return 0; + /* FALLTHRU */ + case 2: + a = (*--srcptr); + switch (*bytes) + { + case 0xE0: + if (a < 0xA0 || a > 0xBF) return 0; + break; + case 0xED: + if (a < 0x80 || a > 0x9F) return 0; + break; + case 0xF0: + if (a < 0x90 || a > 0xBF) return 0; + break; + case 0xF4: + if (a < 0x80 || a > 0x8F) return 0; + break; + default: + if (a < 0x80 || a > 0xBF) return 0; + break; + } + /* FALLTHRU */ + case 1: + if (*bytes >= 0x80 && *bytes < 0xC2) return 0; + } + return *bytes <= 0xF4; +} + +static int +read_utf8(json_stream* json, int next_char) +{ + int count = utf8_seq_length(next_char); + if (!count) + { + json_error(json, "%s", "invalid UTF-8 character"); + return -1; + } + + char buffer[4]; + buffer[0] = next_char; + int i; + for (i = 1; i < count; ++i) + { + if ((buffer[i] = json->source.get(&json->source)) != EOF) + json->lineadj++; + } + + if (!is_legal_utf8((unsigned char*) buffer, count)) + { + json_error(json, "%s", "invalid UTF-8 text"); + return -1; + } + + for (i = 0; i < count; ++i) + { + if (pushchar(json, buffer[i]) != 0) + return -1; + } + return 0; +} + +static enum json_type +read_string(json_stream *json) +{ + if (init_string(json) != 0) + return JSON_ERROR; + while (1) { + int c = json->source.get(&json->source); + if (c == EOF) { + json_error(json, "%s", "unterminated string literal"); + return JSON_ERROR; + } else if (c == '"') { + if (pushchar(json, '\0') == 0) + return JSON_STRING; + else + return JSON_ERROR; + } else if (c == '\\') { + if (read_escaped(json) != 0) + return JSON_ERROR; + } else if ((unsigned) c >= 0x80) { + if (read_utf8(json, c) != 0) + return JSON_ERROR; + } else { + if (char_needs_escaping(c)) { + json_error(json, "%s", "unescaped control character in string"); + return JSON_ERROR; + } + + if (pushchar(json, c) != 0) + return JSON_ERROR; + } + } + return JSON_ERROR; +} + +static int +is_digit(int c) +{ + return c >= 48 /*0*/ && c <= 57 /*9*/; +} + +static int +read_digits(json_stream *json) +{ + int c; + unsigned nread = 0; + while (is_digit(c = json->source.peek(&json->source))) { + if (pushchar(json, json->source.get(&json->source)) != 0) + return -1; + + nread++; + } + + if (nread == 0) { + json_error(json, "expected digit instead of byte '%c'", c); + return -1; + } + + return 0; +} + +static enum json_type +read_number(json_stream *json, int c) +{ + if (pushchar(json, c) != 0) + return JSON_ERROR; + if (c == '-') { + c = json->source.get(&json->source); + if (is_digit(c)) { + return read_number(json, c); + } else { + json_error(json, "unexpected byte '%c' in number", c); + return JSON_ERROR; + } + } else if (strchr("123456789", c) != NULL) { + c = json->source.peek(&json->source); + if (is_digit(c)) { + if (read_digits(json) != 0) + return JSON_ERROR; + } + } + /* Up to decimal or exponent has been read. */ + c = json->source.peek(&json->source); + if (strchr(".eE", c) == NULL) { + if (pushchar(json, '\0') != 0) + return JSON_ERROR; + else + return JSON_NUMBER; + } + if (c == '.') { + json->source.get(&json->source); // consume . + if (pushchar(json, c) != 0) + return JSON_ERROR; + if (read_digits(json) != 0) + return JSON_ERROR; + } + /* Check for exponent. */ + c = json->source.peek(&json->source); + if (c == 'e' || c == 'E') { + json->source.get(&json->source); // consume e/E + if (pushchar(json, c) != 0) + return JSON_ERROR; + c = json->source.peek(&json->source); + if (c == '+' || c == '-') { + json->source.get(&json->source); // consume + if (pushchar(json, c) != 0) + return JSON_ERROR; + if (read_digits(json) != 0) + return JSON_ERROR; + } else if (is_digit(c)) { + if (read_digits(json) != 0) + return JSON_ERROR; + } else { + json->source.get(&json->source); // consume (for column) + json_error(json, "unexpected byte '%c' in number", c); + return JSON_ERROR; + } + } + if (pushchar(json, '\0') != 0) + return JSON_ERROR; + else + return JSON_NUMBER; +} + +bool +json_isspace(int c) +{ + switch (c) { + case 0x09: + case 0x0a: + case 0x0d: + case 0x20: + return true; + } + + return false; +} + +static void newline(json_stream *json) +{ + json->lineno++; + json->linepos = json->source.position; + json->lineadj = 0; + json->linecon = 0; +} + +/* Returns the next non-whitespace character in the stream. + * + * Note that this is the only function (besides user-facing json_source_get()) + * that needs to worry about newline housekeeping. + */ +static int next(json_stream *json) +{ + int c; + while (json_isspace(c = json->source.get(&json->source))) + if (c == '\n') + newline(json); + return c; +} + +static enum json_type +read_value(json_stream *json, int c) +{ + enum json_type type; + size_t colno = json_get_column(json); + + json->ntokens++; + + switch (c) { + case EOF: + json_error(json, "%s", "unexpected end of text"); + type = JSON_ERROR; + break; + case '{': + type = push(json, JSON_OBJECT); + break; + case '[': + type = push(json, JSON_ARRAY); + break; + case '"': + type = read_string(json); + break; + case 'n': + type = is_match(json, "ull", JSON_NULL); + break; + case 'f': + type = is_match(json, "alse", JSON_FALSE); + break; + case 't': + type = is_match(json, "rue", JSON_TRUE); + break; + case '0': + case '1': + case '2': + case '3': + case '4': + case '5': + case '6': + case '7': + case '8': + case '9': + case '-': + type = init_string(json) == 0 ? read_number(json, c) : JSON_ERROR; + break; + default: + type = JSON_ERROR; + json_error(json, "unexpected byte '%c' in value", c); + break; + } + + if (type != JSON_ERROR) + json->colno = colno; + + return type; +} + +enum json_type json_peek(json_stream *json) +{ + enum json_type next; + if (json->next) + next = json->next; + else + next = json->next = json_next(json); + return next; +} + +enum json_type json_next(json_stream *json) +{ + if (json->flags & JSON_FLAG_ERROR) + return JSON_ERROR; + if (json->next != 0) { + enum json_type next = json->next; + json->next = (enum json_type)0; + return next; + } + + json->colno = 0; + + if (json->ntokens > 0 && json->stack_top == (size_t)-1) { + + /* In the streaming mode leave any trailing whitespaces in the stream. + * This allows the user to validate any desired separation between + * values (such as newlines) using json_source_get/peek() with any + * remaining whitespaces ignored as leading when we parse the next + * value. */ + if (!(json->flags & JSON_FLAG_STREAMING)) { + int c = next(json); + if (c != EOF) { + json_error(json, "expected end of text instead of byte '%c'", c); + return JSON_ERROR; + } + } + + return JSON_DONE; + } + int c = next(json); + if (json->stack_top == (size_t)-1) { + if (c == EOF && (json->flags & JSON_FLAG_STREAMING)) + return JSON_DONE; + + return read_value(json, c); + } + if (json->stack[json->stack_top].type == JSON_ARRAY) { + if (json->stack[json->stack_top].count == 0) { + if (c == ']') { + return pop(json, c, JSON_ARRAY); + } + json->stack[json->stack_top].count++; + return read_value(json, c); + } else if (c == ',') { + json->stack[json->stack_top].count++; + return read_value(json, next(json)); + } else if (c == ']') { + return pop(json, c, JSON_ARRAY); + } else { + json_error(json, "unexpected byte '%c'", c); + return JSON_ERROR; + } + } else if (json->stack[json->stack_top].type == JSON_OBJECT) { + if (json->stack[json->stack_top].count == 0) { + if (c == '}') { + return pop(json, c, JSON_OBJECT); + } + + /* No member name/value pairs yet. */ + enum json_type value = read_value(json, c); + if (value != JSON_STRING) { + if (value != JSON_ERROR) + json_error(json, "%s", "expected member name or '}'"); + return JSON_ERROR; + } else { + json->stack[json->stack_top].count++; + return value; + } + } else if ((json->stack[json->stack_top].count % 2) == 0) { + /* Expecting comma followed by member name. */ + if (c != ',' && c != '}') { + json_error(json, "%s", "expected ',' or '}' after member value"); + return JSON_ERROR; + } else if (c == '}') { + return pop(json, c, JSON_OBJECT); + } else { + enum json_type value = read_value(json, next(json)); + if (value != JSON_STRING) { + if (value != JSON_ERROR) + json_error(json, "%s", "expected member name"); + return JSON_ERROR; + } else { + json->stack[json->stack_top].count++; + return value; + } + } + } else if ((json->stack[json->stack_top].count % 2) == 1) { + /* Expecting colon followed by value. */ + if (c != ':') { + json_error(json, "%s", "expected ':' after member name"); + return JSON_ERROR; + } else { + json->stack[json->stack_top].count++; + return read_value(json, next(json)); + } + } + } + json_error(json, "%s", "invalid parser state"); + return JSON_ERROR; +} + +void json_reset(json_stream *json) +{ + json->stack_top = -1; + json->ntokens = 0; + json->flags &= ~JSON_FLAG_ERROR; + json->errmsg[0] = '\0'; +} + +enum json_type json_skip(json_stream *json) +{ + enum json_type type = json_next(json); + size_t cnt_arr = 0; + size_t cnt_obj = 0; + + for (enum json_type skip = type; ; skip = json_next(json)) { + if (skip == JSON_ERROR || skip == JSON_DONE) + return skip; + + if (skip == JSON_ARRAY) { + ++cnt_arr; + } else if (skip == JSON_ARRAY_END && cnt_arr > 0) { + --cnt_arr; + } else if (skip == JSON_OBJECT) { + ++cnt_obj; + } else if (skip == JSON_OBJECT_END && cnt_obj > 0) { + --cnt_obj; + } + + if (!cnt_arr && !cnt_obj) + break; + } + + return type; +} + +enum json_type json_skip_until(json_stream *json, enum json_type type) +{ + while (1) { + enum json_type skip = json_skip(json); + + if (skip == JSON_ERROR || skip == JSON_DONE) + return skip; + + if (skip == type) + break; + } + + return type; +} + +const char *json_get_string(json_stream *json, size_t *length) +{ + if (length != NULL) + *length = json->data.string_fill; + if (json->data.string == NULL) + return ""; + else + return json->data.string; +} + +double json_get_number(json_stream *json) +{ + char *p = json->data.string; + return p == NULL ? 0 : strtod(p, NULL); +} + +const char *json_get_error(json_stream *json) +{ + return json->flags & JSON_FLAG_ERROR ? json->errmsg : NULL; +} + +size_t json_get_lineno(json_stream *json) +{ + return json->lineno; +} + +size_t json_get_position(json_stream *json) +{ + return json->source.position; +} + +size_t json_get_column(json_stream *json) +{ + return json->colno == 0 + ? json->source.position == 0 ? 1 : json->source.position - json->linepos - json->lineadj + : json->colno; +} + +size_t json_get_depth(json_stream *json) +{ + return json->stack_top + 1; +} + +/* Return the current parsing context, that is, JSON_OBJECT if we are inside + an object, JSON_ARRAY if we are inside an array, and JSON_DONE if we are + not yet/anymore in either. + + Additionally, for the first two cases, also return the number of parsing + events that have already been observed at this level with json_next/peek(). + In particular, inside an object, an odd number would indicate that the just + observed JSON_STRING event is a member name. +*/ +enum json_type json_get_context(json_stream *json, size_t *count) +{ + if (json->stack_top == (size_t)-1) + return JSON_DONE; + + if (count != NULL) + *count = json->stack[json->stack_top].count; + + return json->stack[json->stack_top].type; +} + +int json_source_get(json_stream *json) +{ + /* If the caller reads a multi-byte UTF-8 sequence, we expect them to read + * it in its entirety. We also assume that any invalid bytes within such a + * sequence belong to the same column (as opposed to starting a new column + * or some such). */ + + int c = json->source.get(&json->source); + if (json->linecon > 0) { + /* Expecting a continuation byte within a multi-byte UTF-8 sequence. */ + json->linecon--; + if (c != EOF) + json->lineadj++; + } else if (c == '\n') + newline(json); + else if (c >= 0xC2 && c <= 0xF4) /* First in multi-byte UTF-8 sequence. */ + json->linecon = utf8_seq_length(c) - 1; + + return c; +} + +int json_source_peek(json_stream *json) +{ + return json->source.peek(&json->source); +} + +void json_open_buffer(json_stream *json, const void *buffer, size_t size) +{ + init(json); + json->source.get = buffer_get; + json->source.peek = buffer_peek; + json->source.source.buffer.buffer = (const char *)buffer; + json->source.source.buffer.length = size; +} + +void json_open_string(json_stream *json, const char *string) +{ + json_open_buffer(json, string, strlen(string)); +} + +void json_open_stream(json_stream *json, FILE * stream) +{ + init(json); + json->source.get = stream_get; + json->source.peek = stream_peek; + json->source.source.stream.stream = stream; +} + +static int user_get(struct json_source *json) +{ + int c = json->source.user.get(json->source.user.ptr); + if (c != EOF) + json->position++; + return c; +} + +static int user_peek(struct json_source *json) +{ + return json->source.user.peek(json->source.user.ptr); +} + +void json_open_user(json_stream *json, json_user_io get, json_user_io peek, void *user) +{ + init(json); + json->source.get = user_get; + json->source.peek = user_peek; + json->source.source.user.ptr = user; + json->source.source.user.get = get; + json->source.source.user.peek = peek; +} + +void json_set_allocator(json_stream *json, json_allocator *a) +{ + json->alloc = *a; +} + +void json_set_streaming(json_stream *json, bool streaming) +{ + if (streaming) + json->flags |= JSON_FLAG_STREAMING; + else + json->flags &= ~JSON_FLAG_STREAMING; +} + +void json_close(json_stream *json) +{ + json->alloc.free(json->stack); + json->alloc.free(json->data.string); +} diff --git a/libbutl/json/pdjson.h b/libbutl/json/pdjson.h new file mode 100644 index 0000000..ac698e4 --- /dev/null +++ b/libbutl/json/pdjson.h @@ -0,0 +1,147 @@ +#ifndef PDJSON_H +#define PDJSON_H + +#ifndef PDJSON_SYMEXPORT +# define PDJSON_SYMEXPORT +#endif + +#ifdef __cplusplus +extern "C" { +#else +#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) + #include +#else + #ifndef bool + #define bool int + #define true 1 + #define false 0 + #endif /* bool */ +#endif /* __STDC_VERSION__ */ +#endif /* __cplusplus */ + +#include + +enum json_type { + JSON_ERROR = 1, JSON_DONE, + JSON_OBJECT, JSON_OBJECT_END, JSON_ARRAY, JSON_ARRAY_END, + JSON_STRING, JSON_NUMBER, JSON_TRUE, JSON_FALSE, JSON_NULL +}; + +struct json_allocator { + void *(*malloc)(size_t); + void *(*realloc)(void *, size_t); + void (*free)(void *); +}; + +typedef int (*json_user_io)(void *user); + +typedef struct json_stream json_stream; +typedef struct json_allocator json_allocator; + +PDJSON_SYMEXPORT void json_open_buffer(json_stream *json, const void *buffer, size_t size); +PDJSON_SYMEXPORT void json_open_string(json_stream *json, const char *string); +PDJSON_SYMEXPORT void json_open_stream(json_stream *json, FILE *stream); +PDJSON_SYMEXPORT void json_open_user(json_stream *json, json_user_io get, json_user_io peek, void *user); +PDJSON_SYMEXPORT void json_close(json_stream *json); + +PDJSON_SYMEXPORT void json_set_allocator(json_stream *json, json_allocator *a); +PDJSON_SYMEXPORT void json_set_streaming(json_stream *json, bool mode); + +PDJSON_SYMEXPORT enum json_type json_next(json_stream *json); +PDJSON_SYMEXPORT enum json_type json_peek(json_stream *json); +PDJSON_SYMEXPORT void json_reset(json_stream *json); +PDJSON_SYMEXPORT const char *json_get_string(json_stream *json, size_t *length); +PDJSON_SYMEXPORT double json_get_number(json_stream *json); + +PDJSON_SYMEXPORT enum json_type json_skip(json_stream *json); +PDJSON_SYMEXPORT enum json_type json_skip_until(json_stream *json, enum json_type type); + +PDJSON_SYMEXPORT size_t json_get_lineno(json_stream *json); +PDJSON_SYMEXPORT size_t json_get_position(json_stream *json); +PDJSON_SYMEXPORT size_t json_get_column(json_stream *json); +PDJSON_SYMEXPORT size_t json_get_depth(json_stream *json); +PDJSON_SYMEXPORT enum json_type json_get_context(json_stream *json, size_t *count); +PDJSON_SYMEXPORT const char *json_get_error(json_stream *json); + +PDJSON_SYMEXPORT int json_source_get(json_stream *json); +PDJSON_SYMEXPORT int json_source_peek(json_stream *json); +PDJSON_SYMEXPORT bool json_isspace(int c); + +/* internal */ + +struct json_source { + int (*get)(struct json_source *); + int (*peek)(struct json_source *); + size_t position; + union { + struct { + FILE *stream; + } stream; + struct { + const char *buffer; + size_t length; + } buffer; + struct { + void *ptr; + json_user_io get; + json_user_io peek; + } user; + } source; +}; + +struct json_stream { + size_t lineno; + + /* While counting lines is straightforward, columns are tricky because we + * have to count codepoints, not bytes. We could have peppered the code + * with increments in all the relevant places but that seems inelegant. + * So instead we calculate the column dynamically, based on the current + * position. + * + * Specifically, we will remember the position at the beginning of each + * line (linepos) and, assuming only the ASCII characters on the line, the + * column will be the difference between the current position and linepos. + * Of course there could also be multi-byte UTF-8 sequences which we will + * handle by keeping an adjustment (lineadj) -- the number of continuation + * bytes encountered on this line so far. Finally, for json_source_get() + * we also have to keep the number of remaining continuation bytes in the + * current multi-byte UTF-8 sequence (linecon). + * + * This is not the end of the story, however: with only the just described + * approach we will always end up with the column of the latest character + * read which is not what we want when returning potentially multi- + * character value events (string, number, etc); in these cases we want to + * return the column of the first character (note that if the value itself + * is invalid and we are returning JSON_ERROR, we still want the current + * column). So to handle this we will cache the start column (colno) for + * such events. + */ + size_t linepos; /* Position at the beginning of the current line. */ + size_t lineadj; /* Adjustment for multi-byte UTF-8 sequences. */ + size_t linecon; /* Number of remaining UTF-8 continuation bytes. */ + size_t colno; /* Start column for value events or 0. */ + + struct json_stack *stack; + size_t stack_top; + size_t stack_size; + enum json_type next; + unsigned flags; + + struct { + char *string; + size_t string_fill; + size_t string_size; + } data; + + size_t ntokens; + + struct json_source source; + struct json_allocator alloc; + char errmsg[128]; +}; + +#ifdef __cplusplus +} /* extern "C" */ +#endif /* __cplusplus */ + +#endif -- cgit v1.1