mirror of
https://github.com/ToruNiina/toml11.git
synced 2024-11-22 12:30:04 +00:00
249 lines
12 KiB
C++
249 lines
12 KiB
C++
// Copyright Toru Niina 2017.
|
|
// Distributed under the MIT License.
|
|
#ifndef TOML11_LEXER_HPP
|
|
#define TOML11_LEXER_HPP
|
|
#include "combinator.hpp"
|
|
#include <stdexcept>
|
|
#include <istream>
|
|
#include <sstream>
|
|
#include <fstream>
|
|
|
|
namespace toml
|
|
{
|
|
namespace detail
|
|
{
|
|
|
|
// these scans contents from current location in a container of char
|
|
// and extract a region that matches their own pattern.
|
|
// to see the implementation of each component, see combinator.hpp.
|
|
|
|
using lex_wschar = either<character<' '>, character<'\t'>>;
|
|
using lex_ws = repeat<lex_wschar, at_least<1>>;
|
|
using lex_newline = either<character<'\n'>,
|
|
sequence<character<'\r'>, character<'\n'>>>;
|
|
using lex_lower = in_range<'a', 'z'>;
|
|
using lex_upper = in_range<'A', 'Z'>;
|
|
using lex_alpha = either<lex_lower, lex_upper>;
|
|
using lex_digit = in_range<'0', '9'>;
|
|
using lex_nonzero = in_range<'1', '9'>;
|
|
using lex_oct_dig = in_range<'0', '7'>;
|
|
using lex_bin_dig = in_range<'0', '1'>;
|
|
using lex_hex_dig = either<lex_digit, in_range<'A', 'F'>, in_range<'a', 'f'>>;
|
|
|
|
using lex_hex_prefix = sequence<character<'0'>, character<'x'>>;
|
|
using lex_oct_prefix = sequence<character<'0'>, character<'o'>>;
|
|
using lex_bin_prefix = sequence<character<'0'>, character<'b'>>;
|
|
using lex_underscore = character<'_'>;
|
|
using lex_plus = character<'+'>;
|
|
using lex_minus = character<'-'>;
|
|
using lex_sign = either<lex_plus, lex_minus>;
|
|
|
|
// digit | nonzero 1*(digit | _ digit)
|
|
using lex_unsigned_dec_int = either<sequence<lex_nonzero, repeat<
|
|
either<lex_digit, sequence<lex_underscore, lex_digit>>, at_least<1>>>,
|
|
lex_digit>;
|
|
// (+|-)? unsigned_dec_int
|
|
using lex_dec_int = sequence<maybe<lex_sign>, lex_unsigned_dec_int>;
|
|
|
|
// hex_prefix hex_dig *(hex_dig | _ hex_dig)
|
|
using lex_hex_int = sequence<lex_hex_prefix, sequence<lex_hex_dig, repeat<
|
|
either<lex_hex_dig, sequence<lex_underscore, lex_hex_dig>>, unlimited>>>;
|
|
// oct_prefix oct_dig *(oct_dig | _ oct_dig)
|
|
using lex_oct_int = sequence<lex_oct_prefix, sequence<lex_oct_dig, repeat<
|
|
either<lex_oct_dig, sequence<lex_underscore, lex_oct_dig>>, unlimited>>>;
|
|
// bin_prefix bin_dig *(bin_dig | _ bin_dig)
|
|
using lex_bin_int = sequence<lex_bin_prefix, sequence<lex_bin_dig, repeat<
|
|
either<lex_bin_dig, sequence<lex_underscore, lex_bin_dig>>, unlimited>>>;
|
|
|
|
// (dec_int | hex_int | oct_int | bin_int)
|
|
using lex_integer = either<lex_bin_int, lex_oct_int, lex_hex_int, lex_dec_int>;
|
|
|
|
// ===========================================================================
|
|
|
|
using lex_inf = sequence<character<'i'>, character<'n'>, character<'f'>>;
|
|
using lex_nan = sequence<character<'n'>, character<'a'>, character<'n'>>;
|
|
using lex_special_float = sequence<maybe<lex_sign>, either<lex_inf, lex_nan>>;
|
|
|
|
using lex_zero_prefixable_int = sequence<lex_digit, repeat<either<lex_digit,
|
|
sequence<lex_underscore, lex_digit>>, unlimited>>;
|
|
|
|
using lex_fractional_part = sequence<character<'.'>, lex_zero_prefixable_int>;
|
|
|
|
#ifdef TOML11_USE_UNRELEASED_TOML_FEATURES
|
|
// use toml-lang/toml HEAD
|
|
using lex_exponent_part = sequence<either<character<'e'>, character<'E'>>,
|
|
maybe<lex_sign>, lex_zero_prefixable_int>;
|
|
#else
|
|
// strictly TOML v0.5.0
|
|
using lex_exponent_part = sequence<either<character<'e'>, character<'E'>>,
|
|
lex_dec_int>;
|
|
#endif
|
|
|
|
using lex_float = either<lex_special_float,
|
|
sequence<lex_dec_int, either<lex_exponent_part,
|
|
sequence<lex_fractional_part, maybe<lex_exponent_part>>>>>;
|
|
|
|
// ===========================================================================
|
|
|
|
using lex_true = sequence<character<'t'>, character<'r'>,
|
|
character<'u'>, character<'e'>>;
|
|
using lex_false = sequence<character<'f'>, character<'a'>, character<'l'>,
|
|
character<'s'>, character<'e'>>;
|
|
using lex_boolean = either<lex_true, lex_false>;
|
|
|
|
// ===========================================================================
|
|
|
|
using lex_date_fullyear = repeat<lex_digit, exactly<4>>;
|
|
using lex_date_month = repeat<lex_digit, exactly<2>>;
|
|
using lex_date_mday = repeat<lex_digit, exactly<2>>;
|
|
using lex_time_delim = either<character<'T'>, character<'t'>, character<' '>>;
|
|
using lex_time_hour = repeat<lex_digit, exactly<2>>;
|
|
using lex_time_minute = repeat<lex_digit, exactly<2>>;
|
|
using lex_time_second = repeat<lex_digit, exactly<2>>;
|
|
using lex_time_secfrac = sequence<character<'.'>,
|
|
repeat<lex_digit, at_least<1>>>;
|
|
|
|
using lex_time_numoffset = sequence<either<character<'+'>, character<'-'>>,
|
|
sequence<lex_time_hour, character<':'>,
|
|
lex_time_minute>>;
|
|
using lex_time_offset = either<character<'Z'>, character<'z'>,
|
|
lex_time_numoffset>;
|
|
|
|
using lex_partial_time = sequence<lex_time_hour, character<':'>,
|
|
lex_time_minute, character<':'>,
|
|
lex_time_second, maybe<lex_time_secfrac>>;
|
|
using lex_full_date = sequence<lex_date_fullyear, character<'-'>,
|
|
lex_date_month, character<'-'>,
|
|
lex_date_mday>;
|
|
using lex_full_time = sequence<lex_partial_time, lex_time_offset>;
|
|
|
|
using lex_offset_date_time = sequence<lex_full_date, lex_time_delim, lex_full_time>;
|
|
using lex_local_date_time = sequence<lex_full_date, lex_time_delim, lex_partial_time>;
|
|
using lex_local_date = lex_full_date;
|
|
using lex_local_time = lex_partial_time;
|
|
|
|
// ===========================================================================
|
|
|
|
using lex_quotation_mark = character<'"'>;
|
|
#ifdef TOML11_USE_UNRELEASED_TOML_FEATURES
|
|
using lex_basic_unescaped = exclude<either<in_range<0x00, 0x08>, // 0x09 (tab)
|
|
in_range<0x0a, 0x1F>, // is allowed
|
|
character<0x22>, character<0x5C>,
|
|
character<0x7F>>>;
|
|
#else
|
|
using lex_basic_unescaped = exclude<either<in_range<0x00, 0x1F>,
|
|
character<0x22>, character<0x5C>,
|
|
character<0x7F>>>;
|
|
|
|
#endif
|
|
using lex_escape = character<'\\'>;
|
|
using lex_escape_unicode_short = sequence<character<'u'>,
|
|
repeat<lex_hex_dig, exactly<4>>>;
|
|
using lex_escape_unicode_long = sequence<character<'U'>,
|
|
repeat<lex_hex_dig, exactly<8>>>;
|
|
using lex_escape_seq_char = either<character<'"'>, character<'\\'>,
|
|
character<'b'>, character<'f'>,
|
|
character<'n'>, character<'r'>,
|
|
character<'t'>,
|
|
lex_escape_unicode_short,
|
|
lex_escape_unicode_long
|
|
>;
|
|
using lex_escaped = sequence<lex_escape, lex_escape_seq_char>;
|
|
using lex_basic_char = either<lex_basic_unescaped, lex_escaped>;
|
|
using lex_basic_string = sequence<lex_quotation_mark,
|
|
repeat<lex_basic_char, unlimited>,
|
|
lex_quotation_mark>;
|
|
|
|
using lex_ml_basic_string_delim = repeat<lex_quotation_mark, exactly<3>>;
|
|
#ifdef TOML11_USE_UNRELEASED_TOML_FEATURES
|
|
using lex_ml_basic_unescaped = exclude<either<in_range<0x00, 0x08>, // 0x09
|
|
in_range<0x0a, 0x1F>, // is tab
|
|
character<0x5C>,
|
|
character<0x7F>,
|
|
lex_ml_basic_string_delim>>;
|
|
#else // TOML v0.5.0
|
|
using lex_ml_basic_unescaped = exclude<either<in_range<0x00,0x1F>,
|
|
character<0x5C>,
|
|
character<0x7F>,
|
|
lex_ml_basic_string_delim>>;
|
|
#endif
|
|
|
|
using lex_ml_basic_escaped_newline = sequence<
|
|
lex_escape, maybe<lex_ws>, lex_newline,
|
|
repeat<either<lex_ws, lex_newline>, unlimited>>;
|
|
|
|
using lex_ml_basic_char = either<lex_ml_basic_unescaped, lex_escaped>;
|
|
using lex_ml_basic_body = repeat<either<lex_ml_basic_char, lex_newline,
|
|
lex_ml_basic_escaped_newline>,
|
|
unlimited>;
|
|
using lex_ml_basic_string = sequence<lex_ml_basic_string_delim,
|
|
lex_ml_basic_body,
|
|
lex_ml_basic_string_delim>;
|
|
|
|
using lex_literal_char = exclude<either<in_range<0x00, 0x08>,
|
|
in_range<0x10, 0x19>, character<0x27>>>;
|
|
using lex_apostrophe = character<'\''>;
|
|
using lex_literal_string = sequence<lex_apostrophe,
|
|
repeat<lex_literal_char, unlimited>,
|
|
lex_apostrophe>;
|
|
|
|
using lex_ml_literal_string_delim = repeat<lex_apostrophe, exactly<3>>;
|
|
|
|
using lex_ml_literal_char = exclude<either<in_range<0x00, 0x08>,
|
|
in_range<0x10, 0x1F>,
|
|
character<0x7F>,
|
|
lex_ml_literal_string_delim>>;
|
|
using lex_ml_literal_body = repeat<either<lex_ml_literal_char, lex_newline>,
|
|
unlimited>;
|
|
using lex_ml_literal_string = sequence<lex_ml_literal_string_delim,
|
|
lex_ml_literal_body,
|
|
lex_ml_literal_string_delim>;
|
|
|
|
using lex_string = either<lex_ml_basic_string, lex_basic_string,
|
|
lex_ml_literal_string, lex_literal_string>;
|
|
|
|
// ===========================================================================
|
|
|
|
using lex_comment_start_symbol = character<'#'>;
|
|
using lex_non_eol = either<character<'\t'>, exclude<in_range<0x00, 0x19>>>;
|
|
using lex_comment = sequence<lex_comment_start_symbol,
|
|
repeat<lex_non_eol, unlimited>>;
|
|
|
|
using lex_dot_sep = sequence<maybe<lex_ws>, character<'.'>, maybe<lex_ws>>;
|
|
|
|
using lex_unquoted_key = repeat<either<lex_alpha, lex_digit,
|
|
character<'-'>, character<'_'>>,
|
|
at_least<1>>;
|
|
using lex_quoted_key = either<lex_basic_string, lex_literal_string>;
|
|
using lex_simple_key = either<lex_unquoted_key, lex_quoted_key>;
|
|
using lex_dotted_key = sequence<lex_simple_key,
|
|
repeat<sequence<lex_dot_sep, lex_simple_key>,
|
|
at_least<1>
|
|
>
|
|
>;
|
|
using lex_key = either<lex_dotted_key, lex_simple_key>;
|
|
|
|
using lex_keyval_sep = sequence<maybe<lex_ws>,
|
|
character<'='>,
|
|
maybe<lex_ws>>;
|
|
|
|
using lex_std_table_open = character<'['>;
|
|
using lex_std_table_close = character<']'>;
|
|
using lex_std_table = sequence<lex_std_table_open,
|
|
maybe<lex_ws>,
|
|
lex_key,
|
|
maybe<lex_ws>,
|
|
lex_std_table_close>;
|
|
|
|
using lex_array_table_open = sequence<lex_std_table_open, lex_std_table_open>;
|
|
using lex_array_table_close = sequence<lex_std_table_close, lex_std_table_close>;
|
|
using lex_array_table = sequence<lex_array_table_open,
|
|
maybe<lex_ws>,
|
|
lex_key,
|
|
maybe<lex_ws>,
|
|
lex_array_table_close>;
|
|
|
|
} // detail
|
|
} // toml
|
|
#endif // TOML_LEXER_HPP
|