1 | //===-- Format string parser for printf -------------------------*- C++ -*-===// |
2 | // |
3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
4 | // See https://llvm.org/LICENSE.txt for license information. |
5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
6 | // |
7 | //===----------------------------------------------------------------------===// |
8 | |
9 | #ifndef LLVM_LIBC_SRC_STDIO_PRINTF_CORE_PARSER_H |
10 | #define LLVM_LIBC_SRC_STDIO_PRINTF_CORE_PARSER_H |
11 | |
12 | #include "include/llvm-libc-macros/stdfix-macros.h" |
13 | #include "src/__support/CPP/algorithm.h" // max |
14 | #include "src/__support/CPP/optional.h" |
15 | #include "src/__support/CPP/type_traits.h" |
16 | #include "src/__support/str_to_integer.h" |
17 | #include "src/stdio/printf_core/core_structs.h" |
18 | #include "src/stdio/printf_core/printf_config.h" |
19 | |
20 | #include <stddef.h> |
21 | |
22 | #ifdef LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT |
23 | #include "src/__support/fixed_point/fx_rep.h" |
24 | #endif // LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT |
25 | |
26 | namespace LIBC_NAMESPACE { |
27 | namespace printf_core { |
28 | |
29 | template <typename T> struct int_type_of { |
30 | using type = T; |
31 | }; |
32 | template <> struct int_type_of<double> { |
33 | using type = fputil::FPBits<double>::StorageType; |
34 | }; |
35 | template <> struct int_type_of<long double> { |
36 | using type = fputil::FPBits<long double>::StorageType; |
37 | }; |
38 | |
39 | #ifdef LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT |
40 | template <typename T> |
41 | struct int_type_of<cpp::enable_if<cpp::is_fixed_point_v<T>, T>> { |
42 | using type = typename fixed_point::FXRep<T>::StorageType; |
43 | }; |
44 | #endif // LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT |
45 | |
46 | template <typename T> using int_type_of_v = typename int_type_of<T>::type; |
47 | |
48 | #ifndef LIBC_COPT_PRINTF_DISABLE_INDEX_MODE |
49 | #define WRITE_ARG_VAL_SIMPLEST(dst, arg_type, index) \ |
50 | { \ |
51 | auto temp = get_arg_value<arg_type>(index); \ |
52 | if (!temp.has_value()) { \ |
53 | section.has_conv = false; \ |
54 | } else { \ |
55 | dst = cpp::bit_cast<int_type_of_v<arg_type>>(temp.value()); \ |
56 | } \ |
57 | } |
58 | #else |
59 | #define WRITE_ARG_VAL_SIMPLEST(dst, arg_type, _) \ |
60 | dst = cpp::bit_cast<int_type_of_v<arg_type>>(get_next_arg_value<arg_type>()) |
61 | #endif // LIBC_COPT_PRINTF_DISABLE_INDEX_MODE |
62 | |
63 | template <typename ArgProvider> class Parser { |
64 | const char *__restrict str; |
65 | |
66 | size_t cur_pos = 0; |
67 | ArgProvider args_cur; |
68 | |
69 | #ifndef LIBC_COPT_PRINTF_DISABLE_INDEX_MODE |
70 | // args_start stores the start of the va_args, which is allows getting the |
71 | // value of arguments that have already been passed. args_index is tracked so |
72 | // that we know which argument args_cur is on. |
73 | ArgProvider args_start; |
74 | size_t args_index = 1; |
75 | |
76 | // Defined in printf_config.h |
77 | static constexpr size_t DESC_ARR_LEN = LIBC_COPT_PRINTF_INDEX_ARR_LEN; |
78 | |
79 | // desc_arr stores the sizes of the variables in the ArgProvider. This is used |
80 | // in index mode to reduce repeated string parsing. The sizes are stored as |
81 | // TypeDesc objects, which store the size as well as minimal type information. |
82 | // This is necessary because some systems separate the floating point and |
83 | // integer values in va_args. |
84 | TypeDesc desc_arr[DESC_ARR_LEN] = {type_desc_from_type<void>()}; |
85 | |
86 | // TODO: Look into object stores for optimization. |
87 | |
88 | #endif // LIBC_COPT_PRINTF_DISABLE_INDEX_MODE |
89 | |
90 | public: |
91 | #ifndef LIBC_COPT_PRINTF_DISABLE_INDEX_MODE |
92 | LIBC_INLINE Parser(const char *__restrict new_str, ArgProvider &args) |
93 | : str(new_str), args_cur(args), args_start(args) {} |
94 | #else |
95 | LIBC_INLINE Parser(const char *__restrict new_str, ArgProvider &args) |
96 | : str(new_str), args_cur(args) {} |
97 | #endif // LIBC_COPT_PRINTF_DISABLE_INDEX_MODE |
98 | |
99 | // get_next_section will parse the format string until it has a fully |
100 | // specified format section. This can either be a raw format section with no |
101 | // conversion, or a format section with a conversion that has all of its |
102 | // variables stored in the format section. |
103 | LIBC_INLINE FormatSection get_next_section() { |
104 | FormatSection section; |
105 | size_t starting_pos = cur_pos; |
106 | if (str[cur_pos] == '%') { |
107 | // format section |
108 | section.has_conv = true; |
109 | |
110 | ++cur_pos; |
111 | [[maybe_unused]] size_t conv_index = 0; |
112 | |
113 | #ifndef LIBC_COPT_PRINTF_DISABLE_INDEX_MODE |
114 | conv_index = parse_index(local_pos: &cur_pos); |
115 | #endif // LIBC_COPT_PRINTF_DISABLE_INDEX_MODE |
116 | |
117 | section.flags = parse_flags(local_pos: &cur_pos); |
118 | |
119 | // handle width |
120 | section.min_width = 0; |
121 | if (str[cur_pos] == '*') { |
122 | ++cur_pos; |
123 | |
124 | WRITE_ARG_VAL_SIMPLEST(section.min_width, int, parse_index(&cur_pos)); |
125 | } else if (internal::isdigit(ch: str[cur_pos])) { |
126 | auto result = internal::strtointeger<int>(src: str + cur_pos, base: 10); |
127 | section.min_width = result.value; |
128 | cur_pos = cur_pos + result.parsed_len; |
129 | } |
130 | if (section.min_width < 0) { |
131 | section.min_width = -section.min_width; |
132 | section.flags = static_cast<FormatFlags>(section.flags | |
133 | FormatFlags::LEFT_JUSTIFIED); |
134 | } |
135 | |
136 | // handle precision |
137 | section.precision = -1; // negative precisions are ignored. |
138 | if (str[cur_pos] == '.') { |
139 | ++cur_pos; |
140 | section.precision = 0; // if there's a . but no specified precision, the |
141 | // precision is implicitly 0. |
142 | if (str[cur_pos] == '*') { |
143 | ++cur_pos; |
144 | |
145 | WRITE_ARG_VAL_SIMPLEST(section.precision, int, parse_index(&cur_pos)); |
146 | |
147 | } else if (internal::isdigit(ch: str[cur_pos])) { |
148 | auto result = internal::strtointeger<int>(src: str + cur_pos, base: 10); |
149 | section.precision = result.value; |
150 | cur_pos = cur_pos + result.parsed_len; |
151 | } |
152 | } |
153 | |
154 | auto [lm, bw] = parse_length_modifier(local_pos: &cur_pos); |
155 | section.length_modifier = lm; |
156 | section.conv_name = str[cur_pos]; |
157 | section.bit_width = bw; |
158 | switch (str[cur_pos]) { |
159 | case ('%'): |
160 | // Regardless of options, a % conversion is always safe. The standard |
161 | // says that "The complete conversion specification shall be %%" but it |
162 | // also says that "If a conversion specification is invalid, the |
163 | // behavior is undefined." Based on that we define that any conversion |
164 | // specification ending in '%' shall display as '%' regardless of any |
165 | // valid or invalid options. |
166 | section.has_conv = true; |
167 | break; |
168 | case ('c'): |
169 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, int, conv_index); |
170 | break; |
171 | case ('d'): |
172 | case ('i'): |
173 | case ('o'): |
174 | case ('x'): |
175 | case ('X'): |
176 | case ('u'): |
177 | case ('b'): |
178 | case ('B'): |
179 | switch (lm) { |
180 | case (LengthModifier::hh): |
181 | case (LengthModifier::h): |
182 | case (LengthModifier::none): |
183 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, int, conv_index); |
184 | break; |
185 | case (LengthModifier::l): |
186 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, long, conv_index); |
187 | break; |
188 | case (LengthModifier::ll): |
189 | case (LengthModifier::L): // This isn't in the standard, but is in other |
190 | // libc implementations. |
191 | |
192 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, long long, conv_index); |
193 | break; |
194 | case (LengthModifier::j): |
195 | |
196 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, intmax_t, conv_index); |
197 | break; |
198 | case (LengthModifier::z): |
199 | |
200 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, size_t, conv_index); |
201 | break; |
202 | case (LengthModifier::t): |
203 | |
204 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, ptrdiff_t, conv_index); |
205 | break; |
206 | |
207 | case (LengthModifier::w): |
208 | case (LengthModifier::wf): |
209 | if (bw == 0) { |
210 | section.has_conv = false; |
211 | } else if (bw <= INT_WIDTH) { |
212 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, int, conv_index); |
213 | } else if (bw <= LONG_WIDTH) { |
214 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, long, conv_index); |
215 | } else if (bw <= LLONG_WIDTH) { |
216 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, long long, conv_index); |
217 | } else { |
218 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, intmax_t, conv_index); |
219 | } |
220 | break; |
221 | } |
222 | break; |
223 | #ifndef LIBC_COPT_PRINTF_DISABLE_FLOAT |
224 | case ('f'): |
225 | case ('F'): |
226 | case ('e'): |
227 | case ('E'): |
228 | case ('a'): |
229 | case ('A'): |
230 | case ('g'): |
231 | case ('G'): |
232 | if (lm != LengthModifier::L) { |
233 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, double, conv_index); |
234 | } else { |
235 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, long double, conv_index); |
236 | } |
237 | break; |
238 | #endif // LIBC_COPT_PRINTF_DISABLE_FLOAT |
239 | #ifdef LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT |
240 | // Capitalization represents sign, but we only need to get the right |
241 | // bitwidth here so we ignore that. |
242 | case ('r'): |
243 | case ('R'): |
244 | // all fract sizes we support are less than 32 bits, and currently doing |
245 | // va_args with fixed point types just doesn't work. |
246 | // TODO: Move to fixed point types once va_args supports it. |
247 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, uint32_t, conv_index); |
248 | break; |
249 | case ('k'): |
250 | case ('K'): |
251 | if (lm == LengthModifier::l) { |
252 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, uint64_t, conv_index); |
253 | } else { |
254 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_raw, uint32_t, conv_index); |
255 | } |
256 | break; |
257 | #endif // LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT |
258 | #ifndef LIBC_COPT_PRINTF_DISABLE_WRITE_INT |
259 | case ('n'): |
260 | #endif // LIBC_COPT_PRINTF_DISABLE_WRITE_INT |
261 | case ('p'): |
262 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_ptr, void *, conv_index); |
263 | break; |
264 | case ('s'): |
265 | WRITE_ARG_VAL_SIMPLEST(section.conv_val_ptr, char *, conv_index); |
266 | break; |
267 | default: |
268 | // if the conversion is undefined, change this to a raw section. |
269 | section.has_conv = false; |
270 | break; |
271 | } |
272 | // If the end of the format section is on the '\0'. This means we need to |
273 | // not advance the cur_pos. |
274 | if (str[cur_pos] != '\0') |
275 | ++cur_pos; |
276 | |
277 | } else { |
278 | // raw section |
279 | section.has_conv = false; |
280 | while (str[cur_pos] != '%' && str[cur_pos] != '\0') |
281 | ++cur_pos; |
282 | } |
283 | section.raw_string = {str + starting_pos, cur_pos - starting_pos}; |
284 | return section; |
285 | } |
286 | |
287 | private: |
288 | // parse_flags parses the flags inside a format string. It assumes that |
289 | // str[*local_pos] is inside a format specifier, and parses any flags it |
290 | // finds. It returns a FormatFlags object containing the set of found flags |
291 | // arithmetically or'd together. local_pos will be moved past any flags found. |
292 | LIBC_INLINE FormatFlags parse_flags(size_t *local_pos) { |
293 | bool found_flag = true; |
294 | FormatFlags flags = FormatFlags(0); |
295 | while (found_flag) { |
296 | switch (str[*local_pos]) { |
297 | case '-': |
298 | flags = static_cast<FormatFlags>(flags | FormatFlags::LEFT_JUSTIFIED); |
299 | break; |
300 | case '+': |
301 | flags = static_cast<FormatFlags>(flags | FormatFlags::FORCE_SIGN); |
302 | break; |
303 | case ' ': |
304 | flags = static_cast<FormatFlags>(flags | FormatFlags::SPACE_PREFIX); |
305 | break; |
306 | case '#': |
307 | flags = static_cast<FormatFlags>(flags | FormatFlags::ALTERNATE_FORM); |
308 | break; |
309 | case '0': |
310 | flags = static_cast<FormatFlags>(flags | FormatFlags::LEADING_ZEROES); |
311 | break; |
312 | default: |
313 | found_flag = false; |
314 | } |
315 | if (found_flag) |
316 | ++*local_pos; |
317 | } |
318 | return flags; |
319 | } |
320 | |
321 | // parse_length_modifier parses the length modifier inside a format string. It |
322 | // assumes that str[*local_pos] is inside a format specifier. It returns a |
323 | // LengthModifier with the length modifier it found. It will advance local_pos |
324 | // after the format specifier if one is found. |
325 | LIBC_INLINE LengthSpec parse_length_modifier(size_t *local_pos) { |
326 | switch (str[*local_pos]) { |
327 | case ('l'): |
328 | if (str[*local_pos + 1] == 'l') { |
329 | *local_pos += 2; |
330 | return {.lm: LengthModifier::ll, .bit_width: 0}; |
331 | } else { |
332 | ++*local_pos; |
333 | return {.lm: LengthModifier::l, .bit_width: 0}; |
334 | } |
335 | case ('w'): { |
336 | LengthModifier lm; |
337 | if (str[*local_pos + 1] == 'f') { |
338 | *local_pos += 2; |
339 | lm = LengthModifier::wf; |
340 | } else { |
341 | ++*local_pos; |
342 | lm = LengthModifier::w; |
343 | } |
344 | if (internal::isdigit(ch: str[*local_pos])) { |
345 | const auto result = internal::strtointeger<int>(src: str + *local_pos, base: 10); |
346 | *local_pos += result.parsed_len; |
347 | return {.lm: lm, .bit_width: static_cast<size_t>(cpp::max(a: 0, b: result.value))}; |
348 | } |
349 | return {.lm: lm, .bit_width: 0}; |
350 | } |
351 | case ('h'): |
352 | if (str[*local_pos + 1] == 'h') { |
353 | *local_pos += 2; |
354 | return {.lm: LengthModifier::hh, .bit_width: 0}; |
355 | } else { |
356 | ++*local_pos; |
357 | return {.lm: LengthModifier::h, .bit_width: 0}; |
358 | } |
359 | case ('L'): |
360 | ++*local_pos; |
361 | return {.lm: LengthModifier::L, .bit_width: 0}; |
362 | case ('j'): |
363 | ++*local_pos; |
364 | return {.lm: LengthModifier::j, .bit_width: 0}; |
365 | case ('z'): |
366 | ++*local_pos; |
367 | return {.lm: LengthModifier::z, .bit_width: 0}; |
368 | case ('t'): |
369 | ++*local_pos; |
370 | return {.lm: LengthModifier::t, .bit_width: 0}; |
371 | default: |
372 | return {.lm: LengthModifier::none, .bit_width: 0}; |
373 | } |
374 | } |
375 | |
376 | // get_next_arg_value gets the next value from the arg list as type T. |
377 | template <class T> LIBC_INLINE T get_next_arg_value() { |
378 | return args_cur.template next_var<T>(); |
379 | } |
380 | |
381 | //---------------------------------------------------- |
382 | // INDEX MODE ONLY FUNCTIONS AFTER HERE: |
383 | //---------------------------------------------------- |
384 | |
385 | #ifndef LIBC_COPT_PRINTF_DISABLE_INDEX_MODE |
386 | |
387 | // parse_index parses the index of a value inside a format string. It |
388 | // assumes that str[*local_pos] points to character after a '%' or '*', and |
389 | // returns 0 if there is no closing $, or if it finds no number. If it finds a |
390 | // number, it will move local_pos past the end of the $, else it will not move |
391 | // local_pos. |
392 | LIBC_INLINE size_t parse_index(size_t *local_pos) { |
393 | if (internal::isdigit(ch: str[*local_pos])) { |
394 | auto result = internal::strtointeger<int>(src: str + *local_pos, base: 10); |
395 | size_t index = result.value; |
396 | if (str[*local_pos + result.parsed_len] != '$') |
397 | return 0; |
398 | *local_pos = 1 + result.parsed_len + *local_pos; |
399 | return index; |
400 | } |
401 | return 0; |
402 | } |
403 | |
404 | LIBC_INLINE void set_type_desc(size_t index, TypeDesc value) { |
405 | if (index != 0 && index <= DESC_ARR_LEN) |
406 | desc_arr[index - 1] = value; |
407 | } |
408 | |
409 | // get_arg_value gets the value from the arg list at index (starting at 1). |
410 | // This may require parsing the format string. An index of 0 is interpreted as |
411 | // the next value. If the format string is not valid, it may have gaps in its |
412 | // indexes. Requesting the value for any index after a gap will fail, since |
413 | // the arg list must be read in order and with the correct types. |
414 | template <class T> LIBC_INLINE cpp::optional<T> get_arg_value(size_t index) { |
415 | if (!(index == 0 || index == args_index)) { |
416 | bool success = args_to_index(index); |
417 | if (!success) { |
418 | // If we can't get to this index, then the value of the arg can't be |
419 | // found. |
420 | return cpp::optional<T>(); |
421 | } |
422 | } |
423 | |
424 | set_type_desc(index, value: type_desc_from_type<T>()); |
425 | |
426 | ++args_index; |
427 | return get_next_arg_value<T>(); |
428 | } |
429 | |
430 | // the ArgProvider can only return the next item in the list. This function is |
431 | // used in index mode when the item that needs to be read is not the next one. |
432 | // It moves cur_args to the index requested so the appropriate value may |
433 | // be read. This may involve parsing the format string, and is in the worst |
434 | // case an O(n^2) operation. |
435 | LIBC_INLINE bool args_to_index(size_t index) { |
436 | if (args_index > index) { |
437 | args_index = 1; |
438 | args_cur = args_start; |
439 | } |
440 | |
441 | while (args_index < index) { |
442 | TypeDesc cur_type_desc = type_desc_from_type<void>(); |
443 | if (args_index <= DESC_ARR_LEN) |
444 | cur_type_desc = desc_arr[args_index - 1]; |
445 | |
446 | if (cur_type_desc == type_desc_from_type<void>()) |
447 | cur_type_desc = get_type_desc(index: args_index); |
448 | |
449 | // A type of void represents the type being unknown. If the type for the |
450 | // requested index isn't in the desc_arr and isn't found by parsing the |
451 | // string, then then advancing to the requested index is impossible. In |
452 | // that case the function returns false. |
453 | if (cur_type_desc == type_desc_from_type<void>()) |
454 | return false; |
455 | |
456 | if (cur_type_desc == type_desc_from_type<uint32_t>()) |
457 | args_cur.template next_var<uint32_t>(); |
458 | else if (cur_type_desc == type_desc_from_type<uint64_t>()) |
459 | args_cur.template next_var<uint64_t>(); |
460 | #ifndef LIBC_COPT_PRINTF_DISABLE_FLOAT |
461 | // Floating point numbers are stored separately from the other arguments. |
462 | else if (cur_type_desc == type_desc_from_type<double>()) |
463 | args_cur.template next_var<double>(); |
464 | else if (cur_type_desc == type_desc_from_type<long double>()) |
465 | args_cur.template next_var<long double>(); |
466 | #endif // LIBC_COPT_PRINTF_DISABLE_FLOAT |
467 | #ifdef LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT |
468 | // Floating point numbers may be stored separately from the other |
469 | // arguments. |
470 | else if (cur_type_desc == type_desc_from_type<short fract>()) |
471 | args_cur.template next_var<short fract>(); |
472 | else if (cur_type_desc == type_desc_from_type<fract>()) |
473 | args_cur.template next_var<fract>(); |
474 | else if (cur_type_desc == type_desc_from_type<long fract>()) |
475 | args_cur.template next_var<long fract>(); |
476 | else if (cur_type_desc == type_desc_from_type<short accum>()) |
477 | args_cur.template next_var<short accum>(); |
478 | else if (cur_type_desc == type_desc_from_type<accum>()) |
479 | args_cur.template next_var<accum>(); |
480 | else if (cur_type_desc == type_desc_from_type<long accum>()) |
481 | args_cur.template next_var<long accum>(); |
482 | #endif // LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT |
483 | // pointers may be stored separately from normal values. |
484 | else if (cur_type_desc == type_desc_from_type<void *>()) |
485 | args_cur.template next_var<void *>(); |
486 | else |
487 | args_cur.template next_var<uint32_t>(); |
488 | |
489 | ++args_index; |
490 | } |
491 | return true; |
492 | } |
493 | |
494 | // get_type_desc assumes that this format string uses index mode. It iterates |
495 | // through the format string until it finds a format specifier that defines |
496 | // the type of index, and returns a TypeDesc describing that type. It does not |
497 | // modify cur_pos. |
498 | LIBC_INLINE TypeDesc get_type_desc(size_t index) { |
499 | // index mode is assumed, and the indicies start at 1, so an index |
500 | // of 0 is invalid. |
501 | size_t local_pos = 0; |
502 | |
503 | while (str[local_pos]) { |
504 | if (str[local_pos] == '%') { |
505 | ++local_pos; |
506 | |
507 | size_t conv_index = parse_index(local_pos: &local_pos); |
508 | |
509 | // the flags aren't relevant for this situation, but I need to skip past |
510 | // them so they're parsed but the result is discarded. |
511 | parse_flags(local_pos: &local_pos); |
512 | |
513 | // handle width |
514 | if (str[local_pos] == '*') { |
515 | ++local_pos; |
516 | |
517 | size_t width_index = parse_index(local_pos: &local_pos); |
518 | set_type_desc(index: width_index, value: type_desc_from_type<int>()); |
519 | if (width_index == index) |
520 | return type_desc_from_type<int>(); |
521 | |
522 | } else if (internal::isdigit(ch: str[local_pos])) { |
523 | while (internal::isdigit(ch: str[local_pos])) |
524 | ++local_pos; |
525 | } |
526 | |
527 | // handle precision |
528 | if (str[local_pos] == '.') { |
529 | ++local_pos; |
530 | if (str[local_pos] == '*') { |
531 | ++local_pos; |
532 | |
533 | size_t precision_index = parse_index(local_pos: &local_pos); |
534 | set_type_desc(index: precision_index, value: type_desc_from_type<int>()); |
535 | if (precision_index == index) |
536 | return type_desc_from_type<int>(); |
537 | |
538 | } else if (internal::isdigit(ch: str[local_pos])) { |
539 | while (internal::isdigit(ch: str[local_pos])) |
540 | ++local_pos; |
541 | } |
542 | } |
543 | |
544 | auto [lm, bw] = parse_length_modifier(local_pos: &local_pos); |
545 | |
546 | // if we don't have an index for this conversion, then its position is |
547 | // unknown and all this information is irrelevant. The rest of this |
548 | // logic has been for skipping past this conversion properly to avoid |
549 | // weirdness with %%. |
550 | if (conv_index == 0) { |
551 | if (str[local_pos] != '\0') |
552 | ++local_pos; |
553 | continue; |
554 | } |
555 | |
556 | TypeDesc conv_size = type_desc_from_type<void>(); |
557 | switch (str[local_pos]) { |
558 | case ('%'): |
559 | conv_size = type_desc_from_type<void>(); |
560 | break; |
561 | case ('c'): |
562 | conv_size = type_desc_from_type<int>(); |
563 | break; |
564 | case ('d'): |
565 | case ('i'): |
566 | case ('o'): |
567 | case ('x'): |
568 | case ('X'): |
569 | case ('u'): |
570 | case ('b'): |
571 | case ('B'): |
572 | switch (lm) { |
573 | case (LengthModifier::hh): |
574 | case (LengthModifier::h): |
575 | case (LengthModifier::none): |
576 | conv_size = type_desc_from_type<int>(); |
577 | break; |
578 | case (LengthModifier::l): |
579 | conv_size = type_desc_from_type<long>(); |
580 | break; |
581 | case (LengthModifier::ll): |
582 | case (LengthModifier::L): // This isn't in the standard, but is in |
583 | // other libc implementations. |
584 | conv_size = type_desc_from_type<long long>(); |
585 | break; |
586 | case (LengthModifier::j): |
587 | conv_size = type_desc_from_type<intmax_t>(); |
588 | break; |
589 | case (LengthModifier::z): |
590 | conv_size = type_desc_from_type<size_t>(); |
591 | break; |
592 | case (LengthModifier::t): |
593 | conv_size = type_desc_from_type<ptrdiff_t>(); |
594 | break; |
595 | case (LengthModifier::w): |
596 | case (LengthModifier::wf): |
597 | if (bw <= INT_WIDTH) { |
598 | conv_size = type_desc_from_type<int>(); |
599 | } else if (bw <= LONG_WIDTH) { |
600 | conv_size = type_desc_from_type<long>(); |
601 | } else if (bw <= LLONG_WIDTH) { |
602 | conv_size = type_desc_from_type<long long>(); |
603 | } else { |
604 | conv_size = type_desc_from_type<intmax_t>(); |
605 | } |
606 | break; |
607 | } |
608 | break; |
609 | #ifndef LIBC_COPT_PRINTF_DISABLE_FLOAT |
610 | case ('f'): |
611 | case ('F'): |
612 | case ('e'): |
613 | case ('E'): |
614 | case ('a'): |
615 | case ('A'): |
616 | case ('g'): |
617 | case ('G'): |
618 | if (lm != LengthModifier::L) |
619 | conv_size = type_desc_from_type<double>(); |
620 | else |
621 | conv_size = type_desc_from_type<long double>(); |
622 | break; |
623 | #endif // LIBC_COPT_PRINTF_DISABLE_FLOAT |
624 | #ifdef LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT |
625 | // Capitalization represents sign, but we only need to get the right |
626 | // bitwidth here so we ignore that. |
627 | case ('r'): |
628 | case ('R'): |
629 | conv_size = type_desc_from_type<uint32_t>(); |
630 | break; |
631 | case ('k'): |
632 | case ('K'): |
633 | if (lm == LengthModifier::l) { |
634 | conv_size = type_desc_from_type<uint64_t>(); |
635 | } else { |
636 | conv_size = type_desc_from_type<uint32_t>(); |
637 | } |
638 | break; |
639 | #endif // LIBC_INTERNAL_PRINTF_HAS_FIXED_POINT |
640 | #ifndef LIBC_COPT_PRINTF_DISABLE_WRITE_INT |
641 | case ('n'): |
642 | #endif // LIBC_COPT_PRINTF_DISABLE_WRITE_INT |
643 | case ('p'): |
644 | case ('s'): |
645 | conv_size = type_desc_from_type<void *>(); |
646 | break; |
647 | default: |
648 | conv_size = type_desc_from_type<int>(); |
649 | break; |
650 | } |
651 | |
652 | set_type_desc(index: conv_index, value: conv_size); |
653 | if (conv_index == index) |
654 | return conv_size; |
655 | } |
656 | // If the end of the format section is on the '\0'. This means we need to |
657 | // not advance the local_pos. |
658 | if (str[local_pos] != '\0') |
659 | ++local_pos; |
660 | } |
661 | |
662 | // If there is no size for the requested index, then it's unknown. Return |
663 | // void. |
664 | return type_desc_from_type<void>(); |
665 | } |
666 | |
667 | #endif // LIBC_COPT_PRINTF_DISABLE_INDEX_MODE |
668 | }; |
669 | |
670 | } // namespace printf_core |
671 | } // namespace LIBC_NAMESPACE |
672 | |
673 | #endif // LLVM_LIBC_SRC_STDIO_PRINTF_CORE_PARSER_H |
674 | |