master
1// -*- C++ -*-
2//===----------------------------------------------------------------------===//
3//
4// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
5// See https://llvm.org/LICENSE.txt for license information.
6// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7//
8//===----------------------------------------------------------------------===//
9
10#ifndef _LIBCPP___FORMAT_FORMAT_STRING_H
11#define _LIBCPP___FORMAT_FORMAT_STRING_H
12
13#include <__assert>
14#include <__config>
15#include <__cstddef/size_t.h>
16#include <__format/format_error.h>
17#include <__iterator/concepts.h>
18#include <__iterator/iterator_traits.h> // iter_value_t
19#include <cstdint>
20
21#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
22# pragma GCC system_header
23#endif
24
25_LIBCPP_BEGIN_NAMESPACE_STD
26
27#if _LIBCPP_STD_VER >= 20
28
29namespace __format {
30
31template <contiguous_iterator _Iterator>
32struct __parse_number_result {
33 _Iterator __last;
34 uint32_t __value;
35};
36
37template <contiguous_iterator _Iterator>
38__parse_number_result(_Iterator, uint32_t) -> __parse_number_result<_Iterator>;
39
40template <contiguous_iterator _Iterator>
41_LIBCPP_HIDE_FROM_ABI constexpr __parse_number_result<_Iterator> __parse_number(_Iterator __begin, _Iterator __end);
42
43/**
44 * The maximum value of a numeric argument.
45 *
46 * This is used for:
47 * * arg-id
48 * * width as value or arg-id.
49 * * precision as value or arg-id.
50 *
51 * The value is compatible with the maximum formatting width and precision
52 * using the `%*` syntax on a 32-bit system.
53 */
54inline constexpr uint32_t __number_max = INT32_MAX;
55
56namespace __detail {
57template <contiguous_iterator _Iterator>
58_LIBCPP_HIDE_FROM_ABI constexpr __parse_number_result<_Iterator>
59__parse_zero(_Iterator __begin, _Iterator, auto& __parse_ctx) {
60 __parse_ctx.check_arg_id(0);
61 return {++__begin, 0}; // can never be larger than the maximum.
62}
63
64template <contiguous_iterator _Iterator>
65_LIBCPP_HIDE_FROM_ABI constexpr __parse_number_result<_Iterator>
66__parse_automatic(_Iterator __begin, _Iterator, auto& __parse_ctx) {
67 size_t __value = __parse_ctx.next_arg_id();
68 _LIBCPP_ASSERT_UNCATEGORIZED(__value <= __number_max, "Compilers don't support this number of arguments");
69
70 return {__begin, uint32_t(__value)};
71}
72
73template <contiguous_iterator _Iterator>
74_LIBCPP_HIDE_FROM_ABI constexpr __parse_number_result<_Iterator>
75__parse_manual(_Iterator __begin, _Iterator __end, auto& __parse_ctx) {
76 __parse_number_result<_Iterator> __r = __format::__parse_number(__begin, __end);
77 __parse_ctx.check_arg_id(__r.__value);
78 return __r;
79}
80
81} // namespace __detail
82
83/**
84 * Parses a number.
85 *
86 * The number is used for the 31-bit values @em width and @em precision. This
87 * allows a maximum value of 2147483647.
88 */
89template <contiguous_iterator _Iterator>
90_LIBCPP_HIDE_FROM_ABI constexpr __parse_number_result<_Iterator>
91__parse_number(_Iterator __begin, _Iterator __end_input) {
92 using _CharT = iter_value_t<_Iterator>;
93 static_assert(__format::__number_max == INT32_MAX, "The algorithm is implemented based on this value.");
94 /*
95 * Limit the input to 9 digits, otherwise we need two checks during every
96 * iteration:
97 * - Are we at the end of the input?
98 * - Does the value exceed width of an uint32_t? (Switching to uint64_t would
99 * have the same issue, but with a higher maximum.)
100 */
101 _Iterator __end = __end_input - __begin > 9 ? __begin + 9 : __end_input;
102 uint32_t __value = *__begin - _CharT('0');
103 while (++__begin != __end) {
104 if (*__begin < _CharT('0') || *__begin > _CharT('9'))
105 return {__begin, __value};
106
107 __value = __value * 10 + *__begin - _CharT('0');
108 }
109
110 if (__begin != __end_input && *__begin >= _CharT('0') && *__begin <= _CharT('9')) {
111 /*
112 * There are more than 9 digits, do additional validations:
113 * - Does the 10th digit exceed the maximum allowed value?
114 * - Are there more than 10 digits?
115 * (More than 10 digits always overflows the maximum.)
116 */
117 uint64_t __v = uint64_t(__value) * 10 + *__begin++ - _CharT('0');
118 if (__v > __number_max || (__begin != __end_input && *__begin >= _CharT('0') && *__begin <= _CharT('9')))
119 std::__throw_format_error("The numeric value of the format specifier is too large");
120
121 __value = __v;
122 }
123
124 return {__begin, __value};
125}
126
127/**
128 * Multiplexer for all parse functions.
129 *
130 * The parser will return a pointer beyond the last consumed character. This
131 * should be the closing '}' of the arg-id.
132 */
133template <contiguous_iterator _Iterator>
134_LIBCPP_HIDE_FROM_ABI constexpr __parse_number_result<_Iterator>
135__parse_arg_id(_Iterator __begin, _Iterator __end, auto& __parse_ctx) {
136 using _CharT = iter_value_t<_Iterator>;
137 switch (*__begin) {
138 case _CharT('0'):
139 return __detail::__parse_zero(__begin, __end, __parse_ctx);
140
141 case _CharT(':'):
142 // This case is conditionally valid. It's allowed in an arg-id in the
143 // replacement-field, but not in the std-format-spec. The caller can
144 // provide a better diagnostic, so accept it here unconditionally.
145 case _CharT('}'):
146 return __detail::__parse_automatic(__begin, __end, __parse_ctx);
147 }
148 if (*__begin < _CharT('0') || *__begin > _CharT('9'))
149 std::__throw_format_error("The argument index starts with an invalid character");
150
151 return __detail::__parse_manual(__begin, __end, __parse_ctx);
152}
153
154} // namespace __format
155
156#endif // _LIBCPP_STD_VER >= 20
157
158_LIBCPP_END_NAMESPACE_STD
159
160#endif // _LIBCPP___FORMAT_FORMAT_STRING_H