2 //===----------------------------------------------------------------------===//
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
8 //===----------------------------------------------------------------------===//
10 #ifndef _LIBCPP___FORMAT_FORMATTER_INTEGRAL_H
11 #define _LIBCPP___FORMAT_FORMATTER_INTEGRAL_H
13 #include <__algorithm/copy.h>
14 #include <__algorithm/copy_n.h>
15 #include <__algorithm/fill_n.h>
16 #include <__algorithm/transform.h>
18 #include <__format/format_error.h>
19 #include <__format/format_fwd.h>
20 #include <__format/formatter.h>
21 #include <__format/parser_std_format_spec.h>
28 #ifndef _LIBCPP_HAS_NO_LOCALIZATION
32 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
33 #pragma GCC system_header
37 #include <__undef_macros>
39 _LIBCPP_BEGIN_NAMESPACE_STD
41 #if _LIBCPP_STD_VER > 17
43 // TODO FMT Remove this once we require compilers with proper C++20 support.
44 // If the compiler has no concepts support, the format header will be disabled.
45 // Without concepts support enable_if needs to be used and that too much effort
46 // to support compilers with partial C++20 support.
47 #if !defined(_LIBCPP_HAS_NO_CONCEPTS)
50 * Integral formatting classes.
52 * There are two types used here:
53 * * C++-type, the type as used in C++.
54 * * format-type, the output type specified in the std-format-spec.
56 * Design of the integral formatters consists of several layers.
57 * * @ref __parser_integral The basic std-format-spec parser for all integral
58 * classes. This parser does the basic sanity checks. It also contains some
59 * helper functions that are nice to have available for all parsers.
60 * * A C++-type specific parser. These parsers must derive from
61 * @ref __parser_integral. Their task is to validate whether the parsed
62 * std-format-spec is valid for the C++-type and selected format-type. After
63 * validation they need to make sure all members are properly set. For
64 * example, when the alignment hasn't changed it needs to set the proper
65 * default alignment for the format-type. The following parsers are available:
66 * - @ref __parser_integer
67 * - @ref __parser_char
68 * - @ref __parser_bool
69 * * A general formatter for all integral types @ref __formatter_integral. This
70 * formatter can handle all formatting of integers and characters. The class
71 * derives from the proper formatter.
72 * Note the boolean string format-type isn't supported in this class.
73 * * A typedef C++-type group combining the @ref __formatter_integral with a
75 * * @ref __formatter_integer
76 * * @ref __formatter_char
77 * * @ref __formatter_bool
78 * * Then every C++-type has its own formatter specializations. They inherit
79 * from the C++-type group typedef. Most specializations need nothing else.
80 * Others need some additional specializations in this class.
82 namespace __format_spec {
84 /** Wrapper around @ref to_chars, returning the output pointer. */
85 template <integral _Tp>
86 _LIBCPP_HIDE_FROM_ABI char* __to_buffer(char* __first, char* __last,
87 _Tp __value, int __base) {
88 // TODO FMT Evaluate code overhead due to not calling the internal function
89 // directly. (Should be zero overhead.)
90 to_chars_result __r = _VSTD::to_chars(__first, __last, __value, __base);
91 _LIBCPP_ASSERT(__r.ec == errc(0), "Internal buffer too small");
96 * Helper to determine the buffer size to output a integer in Base @em x.
98 * There are several overloads for the supported bases. The function uses the
99 * base as template argument so it can be used in a constant expression.
101 template <unsigned_integral _Tp, size_t _Base>
102 _LIBCPP_HIDE_FROM_ABI constexpr size_t __buffer_size() noexcept
103 requires(_Base == 2) {
104 return numeric_limits<_Tp>::digits // The number of binary digits.
105 + 2 // Reserve space for the '0[Bb]' prefix.
106 + 1; // Reserve space for the sign.
109 template <unsigned_integral _Tp, size_t _Base>
110 _LIBCPP_HIDE_FROM_ABI constexpr size_t __buffer_size() noexcept
111 requires(_Base == 8) {
112 return numeric_limits<_Tp>::digits // The number of binary digits.
113 / 3 // Adjust to octal.
114 + 1 // Turn floor to ceil.
115 + 1 // Reserve space for the '0' prefix.
116 + 1; // Reserve space for the sign.
119 template <unsigned_integral _Tp, size_t _Base>
120 _LIBCPP_HIDE_FROM_ABI constexpr size_t __buffer_size() noexcept
121 requires(_Base == 10) {
122 return numeric_limits<_Tp>::digits10 // The floored value.
123 + 1 // Turn floor to ceil.
124 + 1; // Reserve space for the sign.
127 template <unsigned_integral _Tp, size_t _Base>
128 _LIBCPP_HIDE_FROM_ABI constexpr size_t __buffer_size() noexcept
129 requires(_Base == 16) {
130 return numeric_limits<_Tp>::digits // The number of binary digits.
131 / 4 // Adjust to hexadecimal.
132 + 2 // Reserve space for the '0[Xx]' prefix.
133 + 1; // Reserve space for the sign.
137 * Determines the required grouping based on the size of the input.
139 * The grouping's last element will be repeated. For simplicity this repeating
140 * is unwrapped based on the length of the input. (When the input is short some
141 * groups are not processed.)
143 * @returns The size of the groups to write. This means the number of
144 * separator characters written is size() - 1.
146 * @note Since zero-sized groups cause issues they are silently ignored.
148 * @note The grouping field of the locale is always a @c std::string,
149 * regardless whether the @c std::numpunct's type is @c char or @c wchar_t.
151 _LIBCPP_HIDE_FROM_ABI inline string
152 __determine_grouping(ptrdiff_t __size, const string& __grouping) {
153 _LIBCPP_ASSERT(!__grouping.empty() && __size > __grouping[0],
154 "The slow grouping formatting is used while there will be no "
155 "separators written");
157 auto __end = __grouping.end() - 1;
158 auto __ptr = __grouping.begin();
163 __r.push_back(*__ptr);
165 // __size <= 0 so the value pushed will be <= *__ptr.
166 __r.push_back(*__ptr + __size);
170 // Proceed to the next group.
171 if (__ptr != __end) {
174 // Skip grouping with a width of 0.
175 } while (*__ptr == 0 && __ptr != __end);
179 _LIBCPP_UNREACHABLE();
182 template <class _Parser>
183 requires __formatter::__char_type<typename _Parser::char_type>
184 class _LIBCPP_TEMPLATE_VIS __formatter_integral : public _Parser {
186 using _CharT = typename _Parser::char_type;
188 template <integral _Tp>
189 _LIBCPP_HIDE_FROM_ABI auto format(_Tp __value, auto& __ctx)
190 -> decltype(__ctx.out()) {
191 if (this->__width_needs_substitution())
192 this->__substitute_width_arg_id(__ctx.arg(this->__width));
194 if (this->__type == _Flags::_Type::__char)
195 return __format_as_char(__value, __ctx);
197 if constexpr (unsigned_integral<_Tp>)
198 return __format_unsigned_integral(__value, false, __ctx);
200 // Depending on the std-format-spec string the sign and the value
201 // might not be outputted together:
202 // - alternate form may insert a prefix string.
203 // - zero-padding may insert additional '0' characters.
204 // Therefore the value is processed as a positive unsigned value.
205 // The function @ref __insert_sign will a '-' when the value was negative.
206 auto __r = __to_unsigned_like(__value);
207 bool __negative = __value < 0;
209 __r = __complement(__r);
211 return __format_unsigned_integral(__r, __negative, __ctx);
216 /** Generic formatting for format-type c. */
217 _LIBCPP_HIDE_FROM_ABI auto __format_as_char(integral auto __value,
219 -> decltype(__ctx.out()) {
220 if (this->__alignment == _Flags::_Alignment::__default)
221 this->__alignment = _Flags::_Alignment::__right;
223 using _Tp = decltype(__value);
224 if constexpr (!same_as<_CharT, _Tp>) {
225 // cmp_less and cmp_greater can't be used for character types.
226 if constexpr (signed_integral<_CharT> == signed_integral<_Tp>) {
227 if (__value < numeric_limits<_CharT>::min() ||
228 __value > numeric_limits<_CharT>::max())
229 __throw_format_error(
230 "Integral value outside the range of the char type");
231 } else if constexpr (signed_integral<_CharT>) {
232 // _CharT is signed _Tp is unsigned
234 static_cast<make_unsigned_t<_CharT>>(numeric_limits<_CharT>::max()))
235 __throw_format_error(
236 "Integral value outside the range of the char type");
238 // _CharT is unsigned _Tp is signed
239 if (__value < 0 || static_cast<make_unsigned_t<_Tp>>(__value) >
240 numeric_limits<_CharT>::max())
241 __throw_format_error(
242 "Integral value outside the range of the char type");
246 const auto __c = static_cast<_CharT>(__value);
247 return __write(_VSTD::addressof(__c), _VSTD::addressof(__c) + 1,
252 * Generic formatting for format-type bBdoxX.
254 * This small wrapper allocates a buffer with the required size. Then calls
255 * the real formatter with the buffer and the prefix for the base.
257 _LIBCPP_HIDE_FROM_ABI auto
258 __format_unsigned_integral(unsigned_integral auto __value, bool __negative,
259 auto& __ctx) -> decltype(__ctx.out()) {
260 switch (this->__type) {
261 case _Flags::_Type::__binary_lower_case: {
262 array<char, __buffer_size<decltype(__value), 2>()> __array;
263 return __format_unsigned_integral(__array.begin(), __array.end(), __value,
264 __negative, 2, __ctx, "0b");
266 case _Flags::_Type::__binary_upper_case: {
267 array<char, __buffer_size<decltype(__value), 2>()> __array;
268 return __format_unsigned_integral(__array.begin(), __array.end(), __value,
269 __negative, 2, __ctx, "0B");
271 case _Flags::_Type::__octal: {
272 // Octal is special; if __value == 0 there's no prefix.
273 array<char, __buffer_size<decltype(__value), 8>()> __array;
274 return __format_unsigned_integral(__array.begin(), __array.end(), __value,
275 __negative, 8, __ctx,
276 __value != 0 ? "0" : nullptr);
278 case _Flags::_Type::__decimal: {
279 array<char, __buffer_size<decltype(__value), 10>()> __array;
280 return __format_unsigned_integral(__array.begin(), __array.end(), __value,
281 __negative, 10, __ctx, nullptr);
283 case _Flags::_Type::__hexadecimal_lower_case: {
284 array<char, __buffer_size<decltype(__value), 16>()> __array;
285 return __format_unsigned_integral(__array.begin(), __array.end(), __value,
286 __negative, 16, __ctx, "0x");
288 case _Flags::_Type::__hexadecimal_upper_case: {
289 array<char, __buffer_size<decltype(__value), 16>()> __array;
290 return __format_unsigned_integral(__array.begin(), __array.end(), __value,
291 __negative, 16, __ctx, "0X");
294 _LIBCPP_ASSERT(false, "The parser should have validated the type");
295 _LIBCPP_UNREACHABLE();
300 requires(same_as<char, _Tp> || same_as<wchar_t, _Tp>) _LIBCPP_HIDE_FROM_ABI
301 auto __write(const _Tp* __first, const _Tp* __last, auto __out_it)
302 -> decltype(__out_it) {
304 unsigned __size = __last - __first;
305 if (this->__type != _Flags::_Type::__hexadecimal_upper_case) [[likely]] {
306 if (__size >= this->__width)
307 return _VSTD::copy(__first, __last, _VSTD::move(__out_it));
309 return __formatter::__write(_VSTD::move(__out_it), __first, __last,
310 __size, this->__width, this->__fill,
314 // this->__type == _Flags::_Type::__hexadecimal_upper_case
315 // This means all characters in the range [a-f] need to be changed to their
316 // uppercase representation. The transformation is done as transformation
317 // in the output routine instead of before. This avoids another pass over
319 // TODO FMT See whether it's possible to do this transformation during the
320 // conversion. (This probably requires changing std::to_chars' alphabet.)
321 if (__size >= this->__width)
322 return _VSTD::transform(__first, __last, _VSTD::move(__out_it),
325 return __formatter::__write(_VSTD::move(__out_it), __first, __last, __size,
326 __hex_to_upper, this->__width, this->__fill,
330 _LIBCPP_HIDE_FROM_ABI auto
331 __format_unsigned_integral(char* __begin, char* __end,
332 unsigned_integral auto __value, bool __negative,
333 int __base, auto& __ctx, const char* __prefix)
334 -> decltype(__ctx.out()) {
335 char* __first = __insert_sign(__begin, __negative, this->__sign);
336 if (this->__alternate_form && __prefix)
338 *__first++ = *__prefix++;
340 char* __last = __to_buffer(__first, __end, __value, __base);
341 #ifndef _LIBCPP_HAS_NO_LOCALIZATION
342 if (this->__locale_specific_form) {
343 const auto& __np = use_facet<numpunct<_CharT>>(__ctx.locale());
344 string __grouping = __np.grouping();
345 ptrdiff_t __size = __last - __first;
346 // Writing the grouped form has more overhead than the normal output
347 // routines. If there will be no separators written the locale-specific
348 // form is identical to the normal routine. Test whether to grouped form
350 if (!__grouping.empty() && __size > __grouping[0])
351 return __format_grouping(__ctx.out(), __begin, __first, __last,
352 __determine_grouping(__size, __grouping),
353 __np.thousands_sep());
356 auto __out_it = __ctx.out();
357 if (this->__alignment != _Flags::_Alignment::__default)
360 // __buf contains [sign][prefix]data
361 // ^ location of __first
362 // The zero padding is done like:
363 // - Write [sign][prefix]
364 // - Write data right aligned with '0' as fill character.
365 __out_it = _VSTD::copy(__begin, __first, _VSTD::move(__out_it));
366 this->__alignment = _Flags::_Alignment::__right;
367 this->__fill = _CharT('0');
368 uint32_t __size = __first - __begin;
369 this->__width -= _VSTD::min(__size, this->__width);
372 return __write(__first, __last, _VSTD::move(__out_it));
375 #ifndef _LIBCPP_HAS_NO_LOCALIZATION
376 /** Format's the locale-specific form's groupings. */
377 template <class _OutIt, class _CharT>
378 _LIBCPP_HIDE_FROM_ABI _OutIt
379 __format_grouping(_OutIt __out_it, const char* __begin, const char* __first,
380 const char* __last, string&& __grouping, _CharT __sep) {
382 // TODO FMT This function duplicates some functionality of the normal
383 // output routines. Evaluate whether these parts can be efficiently
384 // combined with the existing routines.
386 unsigned __size = (__first - __begin) + // [sign][prefix]
387 (__last - __first) + // data
388 (__grouping.size() - 1); // number of separator characters
390 __formatter::__padding_size_result __padding = {0, 0};
391 if (this->__alignment == _Flags::_Alignment::__default) {
392 // Write [sign][prefix].
393 __out_it = _VSTD::copy(__begin, __first, _VSTD::move(__out_it));
395 if (this->__width > __size) {
396 // Write zero padding.
397 __padding.__before = this->__width - __size;
398 __out_it = _VSTD::fill_n(_VSTD::move(__out_it), this->__width - __size,
402 if (this->__width > __size) {
403 // Determine padding and write padding.
404 __padding = __formatter::__padding_size(__size, this->__width,
407 __out_it = _VSTD::fill_n(_VSTD::move(__out_it), __padding.__before,
410 // Write [sign][prefix].
411 __out_it = _VSTD::copy(__begin, __first, _VSTD::move(__out_it));
414 auto __r = __grouping.rbegin();
415 auto __e = __grouping.rend() - 1;
416 _LIBCPP_ASSERT(__r != __e, "The slow grouping formatting is used while "
417 "there will be no separators written.");
418 // The output is divided in small groups of numbers to write:
419 // - A group before the first separator.
420 // - A separator and a group, repeated for the number of separators.
421 // - A group after the last separator.
422 // This loop achieves that process by testing the termination condition
423 // midway in the loop.
425 // TODO FMT This loop evaluates the loop invariant `this->__type !=
426 // _Flags::_Type::__hexadecimal_upper_case` for every iteration. (This test
427 // happens in the __write call.) Benchmark whether making two loops and
428 // hoisting the invariant is worth the effort.
430 if (this->__type == _Flags::_Type::__hexadecimal_upper_case) {
431 __last = __first + *__r;
432 __out_it = _VSTD::transform(__first, __last, _VSTD::move(__out_it),
436 __out_it = _VSTD::copy_n(__first, *__r, _VSTD::move(__out_it));
447 return _VSTD::fill_n(_VSTD::move(__out_it), __padding.__after,
450 #endif // _LIBCPP_HAS_NO_LOCALIZATION
453 } // namespace __format_spec
455 #endif // !defined(_LIBCPP_HAS_NO_CONCEPTS)
457 #endif //_LIBCPP_STD_VER > 17
459 _LIBCPP_END_NAMESPACE_STD
463 #endif // _LIBCPP___FORMAT_FORMATTER_INTEGRAL_H