| 1 | //===-- lib/Parser/basic-parsers.h ------------------------------*- 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 FORTRAN_PARSER_BASIC_PARSERS_H_ |
| 10 | #define FORTRAN_PARSER_BASIC_PARSERS_H_ |
| 11 | |
| 12 | // Let a "parser" be an instance of any class that supports this |
| 13 | // type definition and member (or static) function: |
| 14 | // |
| 15 | // using resultType = ...; |
| 16 | // std::optional<resultType> Parse(ParseState &) const; |
| 17 | // |
| 18 | // which either returns a value to signify a successful recognition or else |
| 19 | // returns {} to signify failure. On failure, the state cannot be assumed |
| 20 | // to still be valid, in general -- see below for exceptions. |
| 21 | // |
| 22 | // This header defines the fundamental parser class templates and helper |
| 23 | // template functions. See parser-combinators.txt for documentation. |
| 24 | |
| 25 | #include "flang/Common/idioms.h" |
| 26 | #include "flang/Common/indirection.h" |
| 27 | #include "flang/Parser/char-block.h" |
| 28 | #include "flang/Parser/message.h" |
| 29 | #include "flang/Parser/parse-state.h" |
| 30 | #include "flang/Parser/provenance.h" |
| 31 | #include "flang/Parser/user-state.h" |
| 32 | #include "flang/Support/Fortran-features.h" |
| 33 | #include <cstring> |
| 34 | #include <functional> |
| 35 | #include <list> |
| 36 | #include <memory> |
| 37 | #include <optional> |
| 38 | #include <string> |
| 39 | #include <tuple> |
| 40 | #include <type_traits> |
| 41 | #include <utility> |
| 42 | |
| 43 | namespace Fortran::parser { |
| 44 | |
| 45 | // fail<A>("..."_err_en_US) returns a parser that never succeeds. It reports an |
| 46 | // error message at the current position. The result type is unused, |
| 47 | // but might have to be specified at the point of call to satisfy |
| 48 | // the type checker. The state remains valid. |
| 49 | template <typename A> class FailParser { |
| 50 | public: |
| 51 | using resultType = A; |
| 52 | constexpr FailParser(const FailParser &) = default; |
| 53 | constexpr explicit FailParser(MessageFixedText t) : text_{t} {} |
| 54 | std::optional<A> Parse(ParseState &state) const { |
| 55 | state.Say(text_); |
| 56 | return std::nullopt; |
| 57 | } |
| 58 | |
| 59 | private: |
| 60 | const MessageFixedText text_; |
| 61 | }; |
| 62 | |
| 63 | template <typename A = Success> inline constexpr auto fail(MessageFixedText t) { |
| 64 | return FailParser<A>{t}; |
| 65 | } |
| 66 | |
| 67 | // pure(x) returns a parser that always succeeds, does not advance the |
| 68 | // parse, and returns a captured value x whose type must be copy-constructible. |
| 69 | // |
| 70 | // pure<A>() is essentially pure(A{}); it returns a default-constructed A{}, |
| 71 | // and works even when A is not copy-constructible. |
| 72 | template <typename A> class PureParser { |
| 73 | public: |
| 74 | using resultType = A; |
| 75 | constexpr PureParser(const PureParser &) = default; |
| 76 | constexpr explicit PureParser(A &&x) : value_(std::move(x)) {} |
| 77 | std::optional<A> Parse(ParseState &) const { return value_; } |
| 78 | |
| 79 | private: |
| 80 | const A value_; |
| 81 | }; |
| 82 | |
| 83 | template <typename A> inline constexpr auto pure(A x) { |
| 84 | return PureParser<A>(std::move(x)); |
| 85 | } |
| 86 | |
| 87 | template <typename A> class PureDefaultParser { |
| 88 | public: |
| 89 | using resultType = A; |
| 90 | constexpr PureDefaultParser(const PureDefaultParser &) = default; |
| 91 | constexpr PureDefaultParser() {} |
| 92 | std::optional<A> Parse(ParseState &) const { return std::make_optional<A>(); } |
| 93 | }; |
| 94 | |
| 95 | template <typename A> inline constexpr auto pure() { |
| 96 | return PureDefaultParser<A>(); |
| 97 | } |
| 98 | |
| 99 | // If a is a parser, attempt(a) is the same parser, but on failure |
| 100 | // the ParseState is guaranteed to have been restored to its initial value. |
| 101 | template <typename A> class BacktrackingParser { |
| 102 | public: |
| 103 | using resultType = typename A::resultType; |
| 104 | constexpr BacktrackingParser(const BacktrackingParser &) = default; |
| 105 | constexpr BacktrackingParser(const A &parser) : parser_{parser} {} |
| 106 | std::optional<resultType> Parse(ParseState &state) const { |
| 107 | Messages messages{std::move(state.messages())}; |
| 108 | ParseState backtrack{state}; |
| 109 | std::optional<resultType> result{parser_.Parse(state)}; |
| 110 | if (result) { |
| 111 | state.messages().Annex(std::move(messages)); |
| 112 | } else { |
| 113 | state = std::move(backtrack); |
| 114 | state.messages() = std::move(messages); |
| 115 | } |
| 116 | return result; |
| 117 | } |
| 118 | |
| 119 | private: |
| 120 | const A parser_; |
| 121 | }; |
| 122 | |
| 123 | template <typename A> inline constexpr auto attempt(const A &parser) { |
| 124 | return BacktrackingParser<A>{parser}; |
| 125 | } |
| 126 | |
| 127 | // For any parser x, the parser returned by !x is one that succeeds when |
| 128 | // x fails, returning a useless (but present) result. !x fails when x succeeds. |
| 129 | template <typename PA> class NegatedParser { |
| 130 | public: |
| 131 | using resultType = Success; |
| 132 | constexpr NegatedParser(const NegatedParser &) = default; |
| 133 | constexpr NegatedParser(PA p) : parser_{p} {} |
| 134 | std::optional<Success> Parse(ParseState &state) const { |
| 135 | ParseState forked{state}; |
| 136 | forked.set_deferMessages(true); |
| 137 | if (parser_.Parse(forked)) { |
| 138 | return std::nullopt; |
| 139 | } |
| 140 | return Success{}; |
| 141 | } |
| 142 | |
| 143 | private: |
| 144 | const PA parser_; |
| 145 | }; |
| 146 | |
| 147 | template <typename PA, typename = typename PA::resultType> |
| 148 | constexpr auto operator!(PA p) { |
| 149 | return NegatedParser<PA>(p); |
| 150 | } |
| 151 | |
| 152 | // For any parser x, the parser returned by lookAhead(x) is one that succeeds |
| 153 | // or fails if x does, but the state is not modified. |
| 154 | template <typename PA> class LookAheadParser { |
| 155 | public: |
| 156 | using resultType = Success; |
| 157 | constexpr LookAheadParser(const LookAheadParser &) = default; |
| 158 | constexpr LookAheadParser(PA p) : parser_{p} {} |
| 159 | std::optional<Success> Parse(ParseState &state) const { |
| 160 | ParseState forked{state}; |
| 161 | forked.set_deferMessages(true); |
| 162 | if (parser_.Parse(forked)) { |
| 163 | return Success{}; |
| 164 | } |
| 165 | return std::nullopt; |
| 166 | } |
| 167 | |
| 168 | private: |
| 169 | const PA parser_; |
| 170 | }; |
| 171 | |
| 172 | template <typename PA> inline constexpr auto lookAhead(PA p) { |
| 173 | return LookAheadParser<PA>{p}; |
| 174 | } |
| 175 | |
| 176 | // If a is a parser, inContext("..."_en_US, a) runs it in a nested message |
| 177 | // context. |
| 178 | template <typename PA> class MessageContextParser { |
| 179 | public: |
| 180 | using resultType = typename PA::resultType; |
| 181 | constexpr MessageContextParser(const MessageContextParser &) = default; |
| 182 | constexpr MessageContextParser(MessageFixedText t, PA p) |
| 183 | : text_{t}, parser_{p} {} |
| 184 | std::optional<resultType> Parse(ParseState &state) const { |
| 185 | state.PushContext(text_); |
| 186 | std::optional<resultType> result{parser_.Parse(state)}; |
| 187 | state.PopContext(); |
| 188 | return result; |
| 189 | } |
| 190 | |
| 191 | private: |
| 192 | const MessageFixedText text_; |
| 193 | const PA parser_; |
| 194 | }; |
| 195 | |
| 196 | template <typename PA> |
| 197 | inline constexpr auto inContext(MessageFixedText context, PA parser) { |
| 198 | return MessageContextParser{context, parser}; |
| 199 | } |
| 200 | |
| 201 | // If a is a parser, withMessage("..."_en_US, a) runs it unchanged if it |
| 202 | // succeeds, and overrides its messages with a specific one if it fails and |
| 203 | // has matched no tokens. |
| 204 | template <typename PA> class WithMessageParser { |
| 205 | public: |
| 206 | using resultType = typename PA::resultType; |
| 207 | constexpr WithMessageParser(const WithMessageParser &) = default; |
| 208 | constexpr WithMessageParser(MessageFixedText t, PA p) |
| 209 | : text_{t}, parser_{p} {} |
| 210 | std::optional<resultType> Parse(ParseState &state) const { |
| 211 | if (state.deferMessages()) { // fast path |
| 212 | std::optional<resultType> result{parser_.Parse(state)}; |
| 213 | if (!result) { |
| 214 | state.set_anyDeferredMessages(); |
| 215 | } |
| 216 | return result; |
| 217 | } |
| 218 | Messages messages{std::move(state.messages())}; |
| 219 | bool hadAnyTokenMatched{state.anyTokenMatched()}; |
| 220 | state.set_anyTokenMatched(false); |
| 221 | std::optional<resultType> result{parser_.Parse(state)}; |
| 222 | bool emitMessage{false}; |
| 223 | if (result) { |
| 224 | messages.Annex(std::move(state.messages())); |
| 225 | if (hadAnyTokenMatched) { |
| 226 | state.set_anyTokenMatched(); |
| 227 | } |
| 228 | } else if (state.anyTokenMatched()) { |
| 229 | emitMessage = state.messages().empty(); |
| 230 | messages.Annex(std::move(state.messages())); |
| 231 | } else { |
| 232 | emitMessage = true; |
| 233 | if (hadAnyTokenMatched) { |
| 234 | state.set_anyTokenMatched(); |
| 235 | } |
| 236 | } |
| 237 | state.messages() = std::move(messages); |
| 238 | if (emitMessage) { |
| 239 | state.Say(text_); |
| 240 | } |
| 241 | return result; |
| 242 | } |
| 243 | |
| 244 | private: |
| 245 | const MessageFixedText text_; |
| 246 | const PA parser_; |
| 247 | }; |
| 248 | |
| 249 | template <typename PA> |
| 250 | inline constexpr auto withMessage(MessageFixedText msg, PA parser) { |
| 251 | return WithMessageParser{msg, parser}; |
| 252 | } |
| 253 | |
| 254 | // If a and b are parsers, then a >> b returns a parser that succeeds when |
| 255 | // b succeeds after a does so, but fails when either a or b does. The |
| 256 | // result is taken from b. Similarly, a / b also succeeds if both a and b |
| 257 | // do so, but the result is that returned by a. |
| 258 | template <typename PA, typename PB> class SequenceParser { |
| 259 | public: |
| 260 | using resultType = typename PB::resultType; |
| 261 | constexpr SequenceParser(const SequenceParser &) = default; |
| 262 | constexpr SequenceParser(PA pa, PB pb) : pa_{pa}, pb2_{pb} {} |
| 263 | std::optional<resultType> Parse(ParseState &state) const { |
| 264 | if (pa_.Parse(state)) { |
| 265 | return pb2_.Parse(state); |
| 266 | } else { |
| 267 | return std::nullopt; |
| 268 | } |
| 269 | } |
| 270 | |
| 271 | private: |
| 272 | const PA pa_; |
| 273 | const PB pb2_; |
| 274 | }; |
| 275 | |
| 276 | template <typename PA, typename PB> |
| 277 | inline constexpr auto operator>>(PA pa, PB pb) { |
| 278 | return SequenceParser<PA, PB>{pa, pb}; |
| 279 | } |
| 280 | |
| 281 | template <typename PA, typename PB> class FollowParser { |
| 282 | public: |
| 283 | using resultType = typename PA::resultType; |
| 284 | constexpr FollowParser(const FollowParser &) = default; |
| 285 | constexpr FollowParser(PA pa, PB pb) : pa_{pa}, pb_{pb} {} |
| 286 | std::optional<resultType> Parse(ParseState &state) const { |
| 287 | if (std::optional<resultType> ax{pa_.Parse(state)}) { |
| 288 | if (pb_.Parse(state)) { |
| 289 | return ax; |
| 290 | } |
| 291 | } |
| 292 | return std::nullopt; |
| 293 | } |
| 294 | |
| 295 | private: |
| 296 | const PA pa_; |
| 297 | const PB pb_; |
| 298 | }; |
| 299 | |
| 300 | template <typename PA, typename PB> |
| 301 | inline constexpr auto operator/(PA pa, PB pb) { |
| 302 | return FollowParser<PA, PB>{pa, pb}; |
| 303 | } |
| 304 | |
| 305 | template <typename PA, typename... Ps> class AlternativesParser { |
| 306 | public: |
| 307 | using resultType = typename PA::resultType; |
| 308 | constexpr AlternativesParser(PA pa, Ps... ps) : ps_{pa, ps...} {} |
| 309 | constexpr AlternativesParser(const AlternativesParser &) = default; |
| 310 | std::optional<resultType> Parse(ParseState &state) const { |
| 311 | Messages messages{std::move(state.messages())}; |
| 312 | ParseState backtrack{state}; |
| 313 | std::optional<resultType> result{std::get<0>(ps_).Parse(state)}; |
| 314 | if constexpr (sizeof...(Ps) > 0) { |
| 315 | if (!result) { |
| 316 | ParseRest<1>(result, state, backtrack); |
| 317 | } |
| 318 | } |
| 319 | state.messages().Annex(std::move(messages)); |
| 320 | return result; |
| 321 | } |
| 322 | |
| 323 | private: |
| 324 | template <int J> |
| 325 | void ParseRest(std::optional<resultType> &result, ParseState &state, |
| 326 | ParseState &backtrack) const { |
| 327 | ParseState prevState{std::move(state)}; |
| 328 | state = backtrack; |
| 329 | result = std::get<J>(ps_).Parse(state); |
| 330 | if (!result) { |
| 331 | state.CombineFailedParses(std::move(prevState)); |
| 332 | if constexpr (J < sizeof...(Ps)) { |
| 333 | ParseRest<J + 1>(result, state, backtrack); |
| 334 | } |
| 335 | } |
| 336 | } |
| 337 | |
| 338 | const std::tuple<PA, Ps...> ps_; |
| 339 | }; |
| 340 | |
| 341 | template <typename... Ps> inline constexpr auto first(Ps... ps) { |
| 342 | return AlternativesParser<Ps...>{ps...}; |
| 343 | } |
| 344 | |
| 345 | template <typename PA, typename PB> |
| 346 | inline constexpr auto operator||(PA pa, PB pb) { |
| 347 | return AlternativesParser<PA, PB>{pa, pb}; |
| 348 | } |
| 349 | |
| 350 | // If a and b are parsers, then recovery(a,b) returns a parser that succeeds if |
| 351 | // a does so, or if a fails and b succeeds. If a succeeds, b is not attempted. |
| 352 | // All messages from the first parse are retained. |
| 353 | // The two parsers must return values of the same type. |
| 354 | template <typename PA, typename PB> class RecoveryParser { |
| 355 | public: |
| 356 | using resultType = typename PA::resultType; |
| 357 | static_assert(std::is_same_v<resultType, typename PB::resultType>); |
| 358 | constexpr RecoveryParser(const RecoveryParser &) = default; |
| 359 | constexpr RecoveryParser(PA pa, PB pb) : pa_{pa}, pb_{pb} {} |
| 360 | std::optional<resultType> Parse(ParseState &state) const { |
| 361 | bool originallyDeferred{state.deferMessages()}; |
| 362 | ParseState backtrack{state}; |
| 363 | if (!originallyDeferred && state.messages().empty() && |
| 364 | !state.anyErrorRecovery()) { |
| 365 | // Fast path. There are no messages or recovered errors in the incoming |
| 366 | // state. Attempt to parse with messages deferred, expecting that the |
| 367 | // parse will succeed silently. |
| 368 | state.set_deferMessages(true); |
| 369 | if (std::optional<resultType> ax{pa_.Parse(state)}) { |
| 370 | if (!state.anyDeferredMessages() && !state.anyErrorRecovery()) { |
| 371 | state.set_deferMessages(false); |
| 372 | return ax; |
| 373 | } |
| 374 | } |
| 375 | state = backtrack; |
| 376 | } |
| 377 | Messages messages{std::move(state.messages())}; |
| 378 | if (std::optional<resultType> ax{pa_.Parse(state)}) { |
| 379 | state.messages().Annex(std::move(messages)); |
| 380 | return ax; |
| 381 | } |
| 382 | messages.Annex(std::move(state.messages())); |
| 383 | bool hadDeferredMessages{state.anyDeferredMessages()}; |
| 384 | bool anyTokenMatched{state.anyTokenMatched()}; |
| 385 | state = std::move(backtrack); |
| 386 | state.set_deferMessages(true); |
| 387 | std::optional<resultType> bx{pb_.Parse(state)}; |
| 388 | state.messages() = std::move(messages); |
| 389 | state.set_deferMessages(originallyDeferred); |
| 390 | if (anyTokenMatched) { |
| 391 | state.set_anyTokenMatched(); |
| 392 | } |
| 393 | if (hadDeferredMessages) { |
| 394 | state.set_anyDeferredMessages(); |
| 395 | } |
| 396 | if (bx) { |
| 397 | // Error recovery situations must also produce messages. |
| 398 | CHECK(state.anyDeferredMessages() || state.messages().AnyFatalError()); |
| 399 | state.set_anyErrorRecovery(); |
| 400 | } |
| 401 | return bx; |
| 402 | } |
| 403 | |
| 404 | private: |
| 405 | const PA pa_; |
| 406 | const PB pb_; |
| 407 | }; |
| 408 | |
| 409 | template <typename PA, typename PB> |
| 410 | inline constexpr auto recovery(PA pa, PB pb) { |
| 411 | return RecoveryParser<PA, PB>{pa, pb}; |
| 412 | } |
| 413 | |
| 414 | // If x is a parser, then many(x) returns a parser that always succeeds |
| 415 | // and whose value is a list, possibly empty, of the values returned from |
| 416 | // repeated application of x until it fails or does not advance the parse. |
| 417 | template <typename PA> class ManyParser { |
| 418 | using paType = typename PA::resultType; |
| 419 | |
| 420 | public: |
| 421 | using resultType = std::list<paType>; |
| 422 | constexpr ManyParser(const ManyParser &) = default; |
| 423 | constexpr ManyParser(PA parser) : parser_{parser} {} |
| 424 | std::optional<resultType> Parse(ParseState &state) const { |
| 425 | resultType result; |
| 426 | auto at{state.GetLocation()}; |
| 427 | while (std::optional<paType> x{parser_.Parse(state)}) { |
| 428 | result.emplace_back(std::move(*x)); |
| 429 | if (state.GetLocation() <= at) { |
| 430 | break; // no forward progress, don't loop |
| 431 | } |
| 432 | at = state.GetLocation(); |
| 433 | } |
| 434 | return {std::move(result)}; |
| 435 | } |
| 436 | |
| 437 | private: |
| 438 | const BacktrackingParser<PA> parser_; |
| 439 | }; |
| 440 | |
| 441 | template <typename PA> inline constexpr auto many(PA parser) { |
| 442 | return ManyParser<PA>{parser}; |
| 443 | } |
| 444 | |
| 445 | // If x is a parser, then some(x) returns a parser that succeeds if x does |
| 446 | // and whose value is a nonempty list of the values returned from repeated |
| 447 | // application of x until it fails or does not advance the parse. In other |
| 448 | // words, some(x) is a variant of many(x) that has to succeed at least once. |
| 449 | template <typename PA> class SomeParser { |
| 450 | using paType = typename PA::resultType; |
| 451 | |
| 452 | public: |
| 453 | using resultType = std::list<paType>; |
| 454 | constexpr SomeParser(const SomeParser &) = default; |
| 455 | constexpr SomeParser(PA parser) : parser_{parser} {} |
| 456 | std::optional<resultType> Parse(ParseState &state) const { |
| 457 | auto start{state.GetLocation()}; |
| 458 | if (std::optional<paType> first{parser_.Parse(state)}) { |
| 459 | resultType result; |
| 460 | result.emplace_back(std::move(*first)); |
| 461 | if (state.GetLocation() > start) { |
| 462 | result.splice(result.end(), many(parser_).Parse(state).value()); |
| 463 | } |
| 464 | return {std::move(result)}; |
| 465 | } |
| 466 | return std::nullopt; |
| 467 | } |
| 468 | |
| 469 | private: |
| 470 | const PA parser_; |
| 471 | }; |
| 472 | |
| 473 | template <typename PA> inline constexpr auto some(PA parser) { |
| 474 | return SomeParser<PA>{parser}; |
| 475 | } |
| 476 | |
| 477 | // If x is a parser, skipMany(x) is equivalent to many(x) but with no result. |
| 478 | template <typename PA> class SkipManyParser { |
| 479 | public: |
| 480 | using resultType = Success; |
| 481 | constexpr SkipManyParser(const SkipManyParser &) = default; |
| 482 | constexpr SkipManyParser(PA parser) : parser_{parser} {} |
| 483 | std::optional<Success> Parse(ParseState &state) const { |
| 484 | for (auto at{state.GetLocation()}; |
| 485 | parser_.Parse(state) && state.GetLocation() > at; |
| 486 | at = state.GetLocation()) { |
| 487 | } |
| 488 | return Success{}; |
| 489 | } |
| 490 | |
| 491 | private: |
| 492 | const BacktrackingParser<PA> parser_; |
| 493 | }; |
| 494 | |
| 495 | template <typename PA> inline constexpr auto skipMany(PA parser) { |
| 496 | return SkipManyParser<PA>{parser}; |
| 497 | } |
| 498 | |
| 499 | // If x is a parser, skipManyFast(x) is equivalent to skipMany(x). |
| 500 | // The parser x must always advance on success and never invalidate the |
| 501 | // state on failure. |
| 502 | template <typename PA> class SkipManyFastParser { |
| 503 | public: |
| 504 | using resultType = Success; |
| 505 | constexpr SkipManyFastParser(const SkipManyFastParser &) = default; |
| 506 | constexpr SkipManyFastParser(PA parser) : parser_{parser} {} |
| 507 | std::optional<Success> Parse(ParseState &state) const { |
| 508 | while (parser_.Parse(state)) { |
| 509 | } |
| 510 | return Success{}; |
| 511 | } |
| 512 | |
| 513 | private: |
| 514 | const PA parser_; |
| 515 | }; |
| 516 | |
| 517 | template <typename PA> inline constexpr auto skipManyFast(PA parser) { |
| 518 | return SkipManyFastParser<PA>{parser}; |
| 519 | } |
| 520 | |
| 521 | // If x is a parser returning some type A, then maybe(x) returns a |
| 522 | // parser that returns std::optional<A>, always succeeding. |
| 523 | template <typename PA> class MaybeParser { |
| 524 | using paType = typename PA::resultType; |
| 525 | |
| 526 | public: |
| 527 | using resultType = std::optional<paType>; |
| 528 | constexpr MaybeParser(const MaybeParser &) = default; |
| 529 | constexpr MaybeParser(PA parser) : parser_{parser} {} |
| 530 | std::optional<resultType> Parse(ParseState &state) const { |
| 531 | if (resultType result{parser_.Parse(state)}) { |
| 532 | // permit optional<optional<...>> |
| 533 | return {std::move(result)}; |
| 534 | } |
| 535 | return resultType{}; |
| 536 | } |
| 537 | |
| 538 | private: |
| 539 | const BacktrackingParser<PA> parser_; |
| 540 | }; |
| 541 | |
| 542 | template <typename PA> inline constexpr auto maybe(PA parser) { |
| 543 | return MaybeParser<PA>{parser}; |
| 544 | } |
| 545 | |
| 546 | // If x is a parser, then defaulted(x) returns a parser that always |
| 547 | // succeeds. When x succeeds, its result is that of x; otherwise, its |
| 548 | // result is a default-constructed value of x's result type. |
| 549 | template <typename PA> class DefaultedParser { |
| 550 | public: |
| 551 | using resultType = typename PA::resultType; |
| 552 | constexpr DefaultedParser(const DefaultedParser &) = default; |
| 553 | constexpr DefaultedParser(PA p) : parser_{p} {} |
| 554 | std::optional<resultType> Parse(ParseState &state) const { |
| 555 | std::optional<std::optional<resultType>> ax{maybe(parser_).Parse(state)}; |
| 556 | if (ax.value()) { // maybe() always succeeds |
| 557 | return std::move(*ax); |
| 558 | } |
| 559 | return resultType{}; |
| 560 | } |
| 561 | |
| 562 | private: |
| 563 | const BacktrackingParser<PA> parser_; |
| 564 | }; |
| 565 | |
| 566 | template <typename PA> inline constexpr auto defaulted(PA p) { |
| 567 | return DefaultedParser<PA>(p); |
| 568 | } |
| 569 | |
| 570 | // If a is a parser, and f is a function mapping an rvalue of a's result type |
| 571 | // to some other type T, then applyFunction(f, a) returns a parser that succeeds |
| 572 | // iff a does, and whose result value ax has been passed through the function; |
| 573 | // the final result is that returned by the call f(std::move(ax)). |
| 574 | // |
| 575 | // Function application is generalized to functions with more than one |
| 576 | // argument with applyFunction(f, a, b, ...) succeeding if all of the parsers |
| 577 | // a, b, &c. do so, and the result is the value of applying f to their |
| 578 | // results. |
| 579 | // |
| 580 | // applyLambda(f, ...) is the same concept extended to std::function<> functors. |
| 581 | // It is not constexpr. |
| 582 | // |
| 583 | // Member function application is supported by applyMem(&C::f, a). If the |
| 584 | // parser a succeeds and returns some value ax of type C, the result is that |
| 585 | // returned by ax.f(). Additional parser arguments can be specified to supply |
| 586 | // their results to the member function call, so applyMem(&C::f, a, b) succeeds |
| 587 | // if both a and b do so and returns the result of calling ax.f(std::move(bx)). |
| 588 | |
| 589 | // Runs a sequence of parsers until one fails or all have succeeded. |
| 590 | // Collects their results in a std::tuple<std::optional<>...>. |
| 591 | template <typename... PARSER> |
| 592 | using ApplyArgs = std::tuple<std::optional<typename PARSER::resultType>...>; |
| 593 | |
| 594 | template <typename... PARSER, std::size_t... J> |
| 595 | inline bool ApplyHelperArgs(const std::tuple<PARSER...> &parsers, |
| 596 | ApplyArgs<PARSER...> &args, ParseState &state, std::index_sequence<J...>) { |
| 597 | return (... && |
| 598 | (std::get<J>(args) = std::get<J>(parsers).Parse(state), |
| 599 | std::get<J>(args).has_value())); |
| 600 | } |
| 601 | |
| 602 | // Applies a function to the arguments collected by ApplyHelperArgs. |
| 603 | template <typename RESULT, typename... PARSER> |
| 604 | using ApplicableFunctionPointer = RESULT (*)(typename PARSER::resultType &&...); |
| 605 | template <typename RESULT, typename... PARSER> |
| 606 | using ApplicableFunctionObject = |
| 607 | const std::function<RESULT(typename PARSER::resultType &&...)> &; |
| 608 | |
| 609 | template <template <typename...> class FUNCTION, typename RESULT, |
| 610 | typename... PARSER, std::size_t... J> |
| 611 | inline RESULT ApplyHelperFunction(FUNCTION<RESULT, PARSER...> f, |
| 612 | ApplyArgs<PARSER...> &&args, std::index_sequence<J...>) { |
| 613 | return f(std::move(*std::get<J>(args))...); |
| 614 | } |
| 615 | |
| 616 | template <template <typename...> class FUNCTION, typename RESULT, |
| 617 | typename... PARSER> |
| 618 | class ApplyFunction { |
| 619 | using funcType = FUNCTION<RESULT, PARSER...>; |
| 620 | |
| 621 | public: |
| 622 | using resultType = RESULT; |
| 623 | constexpr ApplyFunction(const ApplyFunction &) = default; |
| 624 | constexpr ApplyFunction(funcType f, PARSER... p) |
| 625 | : function_{f}, parsers_{p...} {} |
| 626 | std::optional<resultType> Parse(ParseState &state) const { |
| 627 | ApplyArgs<PARSER...> results; |
| 628 | using Sequence = std::index_sequence_for<PARSER...>; |
| 629 | if (ApplyHelperArgs(parsers_, results, state, Sequence{})) { |
| 630 | return ApplyHelperFunction<FUNCTION, RESULT, PARSER...>( |
| 631 | function_, std::move(results), Sequence{}); |
| 632 | } else { |
| 633 | return std::nullopt; |
| 634 | } |
| 635 | } |
| 636 | |
| 637 | private: |
| 638 | const funcType function_; |
| 639 | const std::tuple<PARSER...> parsers_; |
| 640 | }; |
| 641 | |
| 642 | template <typename RESULT, typename... PARSER> |
| 643 | inline constexpr auto applyFunction( |
| 644 | ApplicableFunctionPointer<RESULT, PARSER...> f, const PARSER &...parser) { |
| 645 | return ApplyFunction<ApplicableFunctionPointer, RESULT, PARSER...>{ |
| 646 | f, parser...}; |
| 647 | } |
| 648 | |
| 649 | template <typename RESULT, typename... PARSER> |
| 650 | inline /* not constexpr */ auto applyLambda( |
| 651 | ApplicableFunctionObject<RESULT, PARSER...> f, const PARSER &...parser) { |
| 652 | return ApplyFunction<ApplicableFunctionObject, RESULT, PARSER...>{ |
| 653 | f, parser...}; |
| 654 | } |
| 655 | |
| 656 | // Member function application |
| 657 | template <typename MEMFUNC, typename OBJPARSER, typename... PARSER, |
| 658 | std::size_t... J> |
| 659 | inline auto ApplyHelperMember(MEMFUNC mfp, |
| 660 | ApplyArgs<OBJPARSER, PARSER...> &&args, std::index_sequence<J...>) { |
| 661 | return ((*std::get<0>(args)).*mfp)(std::move(*std::get<J + 1>(args))...); |
| 662 | } |
| 663 | |
| 664 | template <typename MEMFUNC, typename OBJPARSER, typename... PARSER> |
| 665 | class ApplyMemberFunction { |
| 666 | static_assert(std::is_member_function_pointer_v<MEMFUNC>); |
| 667 | using funcType = MEMFUNC; |
| 668 | |
| 669 | public: |
| 670 | using resultType = |
| 671 | std::invoke_result_t<MEMFUNC, typename OBJPARSER::resultType, PARSER...>; |
| 672 | |
| 673 | constexpr ApplyMemberFunction(const ApplyMemberFunction &) = default; |
| 674 | constexpr ApplyMemberFunction(MEMFUNC f, OBJPARSER o, PARSER... p) |
| 675 | : function_{f}, parsers_{o, p...} {} |
| 676 | std::optional<resultType> Parse(ParseState &state) const { |
| 677 | ApplyArgs<OBJPARSER, PARSER...> results; |
| 678 | using Sequence1 = std::index_sequence_for<OBJPARSER, PARSER...>; |
| 679 | using Sequence2 = std::index_sequence_for<PARSER...>; |
| 680 | if (ApplyHelperArgs(parsers_, results, state, Sequence1{})) { |
| 681 | return ApplyHelperMember<MEMFUNC, OBJPARSER, PARSER...>( |
| 682 | function_, std::move(results), Sequence2{}); |
| 683 | } else { |
| 684 | return std::nullopt; |
| 685 | } |
| 686 | } |
| 687 | |
| 688 | private: |
| 689 | const funcType function_; |
| 690 | const std::tuple<OBJPARSER, PARSER...> parsers_; |
| 691 | }; |
| 692 | |
| 693 | template <typename MEMFUNC, typename OBJPARSER, typename... PARSER> |
| 694 | inline constexpr auto applyMem( |
| 695 | MEMFUNC memfn, const OBJPARSER &objParser, PARSER... parser) { |
| 696 | return ApplyMemberFunction<MEMFUNC, OBJPARSER, PARSER...>{ |
| 697 | memfn, objParser, parser...}; |
| 698 | } |
| 699 | |
| 700 | // As is done with function application via applyFunction() above, class |
| 701 | // instance construction can also be based upon the results of successful |
| 702 | // parses. For some type T and zero or more parsers a, b, &c., the call |
| 703 | // construct<T>(a, b, ...) returns a parser that succeeds if all of |
| 704 | // its argument parsers do so in succession, and whose result is an |
| 705 | // instance of T constructed upon the values they returned. |
| 706 | // With a single argument that is a parser with no usable value, |
| 707 | // construct<T>(p) invokes T's default nullary constructor (T(){}). |
| 708 | // (This means that "construct<T>(Foo >> Bar >> ok)" is functionally |
| 709 | // equivalent to "Foo >> Bar >> construct<T>()", but I'd like to hold open |
| 710 | // the opportunity to make construct<> capture source provenance all of the |
| 711 | // time, and the first form will then lead to better error positioning.) |
| 712 | |
| 713 | template <typename RESULT, typename... PARSER, std::size_t... J> |
| 714 | inline RESULT ApplyHelperConstructor( |
| 715 | ApplyArgs<PARSER...> &&args, std::index_sequence<J...>) { |
| 716 | return RESULT{std::move(*std::get<J>(args))...}; |
| 717 | } |
| 718 | |
| 719 | template <typename RESULT, typename... PARSER> class ApplyConstructor { |
| 720 | public: |
| 721 | using resultType = RESULT; |
| 722 | constexpr ApplyConstructor(const ApplyConstructor &) = default; |
| 723 | constexpr explicit ApplyConstructor(PARSER... p) : parsers_{p...} {} |
| 724 | std::optional<resultType> Parse(ParseState &state) const { |
| 725 | if constexpr (sizeof...(PARSER) == 0) { |
| 726 | return RESULT{}; |
| 727 | } else { |
| 728 | if constexpr (sizeof...(PARSER) == 1) { |
| 729 | return ParseOne(state); |
| 730 | } else { |
| 731 | ApplyArgs<PARSER...> results; |
| 732 | using Sequence = std::index_sequence_for<PARSER...>; |
| 733 | if (ApplyHelperArgs(parsers_, results, state, Sequence{})) { |
| 734 | return ApplyHelperConstructor<RESULT, PARSER...>( |
| 735 | std::move(results), Sequence{}); |
| 736 | } |
| 737 | } |
| 738 | return std::nullopt; |
| 739 | } |
| 740 | } |
| 741 | |
| 742 | private: |
| 743 | std::optional<resultType> ParseOne(ParseState &state) const { |
| 744 | if constexpr (std::is_same_v<Success, typename PARSER::resultType...>) { |
| 745 | if (std::get<0>(parsers_).Parse(state)) { |
| 746 | return RESULT{}; |
| 747 | } |
| 748 | } else if (auto arg{std::get<0>(parsers_).Parse(state)}) { |
| 749 | return RESULT{std::move(*arg)}; |
| 750 | } |
| 751 | return std::nullopt; |
| 752 | } |
| 753 | |
| 754 | const std::tuple<PARSER...> parsers_; |
| 755 | }; |
| 756 | |
| 757 | template <typename RESULT, typename... PARSER> |
| 758 | inline constexpr auto construct(PARSER... p) { |
| 759 | return ApplyConstructor<RESULT, PARSER...>{p...}; |
| 760 | } |
| 761 | |
| 762 | // For a parser p, indirect(p) returns a parser that builds an indirect |
| 763 | // reference to p's return type. |
| 764 | template <typename PA> inline constexpr auto indirect(PA p) { |
| 765 | return construct<common::Indirection<typename PA::resultType>>(p); |
| 766 | } |
| 767 | |
| 768 | // If a and b are parsers, then nonemptySeparated(a, b) returns a parser |
| 769 | // that succeeds if a does. If a succeeds, it then applies many(b >> a). |
| 770 | // The result is the list of the values returned from all of the applications |
| 771 | // of a. |
| 772 | template <typename T> |
| 773 | common::IfNoLvalue<std::list<T>, T> prepend(T &&head, std::list<T> &&rest) { |
| 774 | rest.push_front(std::move(head)); |
| 775 | return std::move(rest); |
| 776 | } |
| 777 | |
| 778 | template <typename PA, typename PB> class NonemptySeparated { |
| 779 | private: |
| 780 | using paType = typename PA::resultType; |
| 781 | |
| 782 | public: |
| 783 | using resultType = std::list<paType>; |
| 784 | constexpr NonemptySeparated(const NonemptySeparated &) = default; |
| 785 | constexpr NonemptySeparated(PA p, PB sep) : parser_{p}, separator_{sep} {} |
| 786 | std::optional<resultType> Parse(ParseState &state) const { |
| 787 | return applyFunction<std::list<paType>>( |
| 788 | prepend<paType>, parser_, many(separator_ >> parser_)) |
| 789 | .Parse(state); |
| 790 | } |
| 791 | |
| 792 | private: |
| 793 | const PA parser_; |
| 794 | const PB separator_; |
| 795 | }; |
| 796 | |
| 797 | template <typename PA, typename PB> |
| 798 | inline constexpr auto nonemptySeparated(PA p, PB sep) { |
| 799 | return NonemptySeparated<PA, PB>{p, sep}; |
| 800 | } |
| 801 | |
| 802 | // ok is a parser that always succeeds. It is useful when a parser |
| 803 | // must discard its result in order to be compatible in type with other |
| 804 | // parsers in an alternative, e.g. "x >> ok || y >> ok" is type-safe even |
| 805 | // when x and y have distinct result types. |
| 806 | struct OkParser { |
| 807 | using resultType = Success; |
| 808 | constexpr OkParser() {} |
| 809 | static constexpr std::optional<Success> Parse(ParseState &) { |
| 810 | return Success{}; |
| 811 | } |
| 812 | }; |
| 813 | constexpr OkParser ok; |
| 814 | |
| 815 | // A variant of recovery() above for convenience. |
| 816 | template <typename PA, typename PB> |
| 817 | inline constexpr auto localRecovery(MessageFixedText msg, PA pa, PB pb) { |
| 818 | return recovery(withMessage(msg, pa), pb >> pure<typename PA::resultType>()); |
| 819 | } |
| 820 | |
| 821 | // nextCh is a parser that succeeds if the parsing state is not |
| 822 | // at the end of its input, returning the next character location and |
| 823 | // advancing the parse when it does so. |
| 824 | struct NextCh { |
| 825 | using resultType = const char *; |
| 826 | constexpr NextCh() {} |
| 827 | std::optional<const char *> Parse(ParseState &state) const { |
| 828 | if (std::optional<const char *> result{state.GetNextChar()}) { |
| 829 | return result; |
| 830 | } |
| 831 | state.Say("end of file"_err_en_US ); |
| 832 | return std::nullopt; |
| 833 | } |
| 834 | }; |
| 835 | |
| 836 | constexpr NextCh nextCh; |
| 837 | |
| 838 | // If a is a parser for some nonstandard language feature LF, extension<LF>(a) |
| 839 | // is a parser that optionally enabled, sets a strict conformance violation |
| 840 | // flag, and may emit a warning message, if those are enabled. |
| 841 | template <LanguageFeature LF, typename PA> class NonstandardParser { |
| 842 | public: |
| 843 | using resultType = typename PA::resultType; |
| 844 | constexpr NonstandardParser(const NonstandardParser &) = default; |
| 845 | constexpr NonstandardParser(PA parser, MessageFixedText msg) |
| 846 | : parser_{parser}, message_{msg} {} |
| 847 | constexpr NonstandardParser(PA parser) : parser_{parser} {} |
| 848 | std::optional<resultType> Parse(ParseState &state) const { |
| 849 | if (UserState * ustate{state.userState()}) { |
| 850 | if (!ustate->features().IsEnabled(LF)) { |
| 851 | return std::nullopt; |
| 852 | } |
| 853 | } |
| 854 | auto at{state.GetLocation()}; |
| 855 | auto result{parser_.Parse(state)}; |
| 856 | if (result && !message_.empty()) { |
| 857 | state.Nonstandard( |
| 858 | CharBlock{at, std::max(state.GetLocation(), at + 1)}, LF, message_); |
| 859 | } |
| 860 | return result; |
| 861 | } |
| 862 | |
| 863 | private: |
| 864 | const PA parser_; |
| 865 | const MessageFixedText message_; |
| 866 | }; |
| 867 | |
| 868 | template <LanguageFeature LF, typename PA> |
| 869 | inline constexpr auto extension(MessageFixedText feature, PA parser) { |
| 870 | return NonstandardParser<LF, PA>(parser, feature); |
| 871 | } |
| 872 | |
| 873 | template <LanguageFeature LF, typename PA> |
| 874 | inline constexpr auto extension(PA parser) { |
| 875 | return NonstandardParser<LF, PA>(parser); |
| 876 | } |
| 877 | |
| 878 | // If a is a parser for some deprecated or deleted language feature LF, |
| 879 | // deprecated<LF>(a) is a parser that is optionally enabled, sets a strict |
| 880 | // conformance violation flag, and may emit a warning message, if enabled. |
| 881 | template <LanguageFeature LF, typename PA> class DeprecatedParser { |
| 882 | public: |
| 883 | using resultType = typename PA::resultType; |
| 884 | constexpr DeprecatedParser(const DeprecatedParser &) = default; |
| 885 | constexpr DeprecatedParser(PA parser) : parser_{parser} {} |
| 886 | std::optional<resultType> Parse(ParseState &state) const { |
| 887 | if (UserState * ustate{state.userState()}) { |
| 888 | if (!ustate->features().IsEnabled(LF)) { |
| 889 | return std::nullopt; |
| 890 | } |
| 891 | } |
| 892 | auto at{state.GetLocation()}; |
| 893 | auto result{parser_.Parse(state)}; |
| 894 | if (result) { |
| 895 | state.Nonstandard(CharBlock{at, state.GetLocation()}, LF, |
| 896 | "deprecated usage"_port_en_US ); |
| 897 | } |
| 898 | return result; |
| 899 | } |
| 900 | |
| 901 | private: |
| 902 | const PA parser_; |
| 903 | }; |
| 904 | |
| 905 | template <LanguageFeature LF, typename PA> |
| 906 | inline constexpr auto deprecated(PA parser) { |
| 907 | return DeprecatedParser<LF, PA>(parser); |
| 908 | } |
| 909 | |
| 910 | // Parsing objects with "source" members. |
| 911 | template <typename PA> class SourcedParser { |
| 912 | public: |
| 913 | using resultType = typename PA::resultType; |
| 914 | constexpr SourcedParser(const SourcedParser &) = default; |
| 915 | constexpr SourcedParser(PA parser) : parser_{parser} {} |
| 916 | std::optional<resultType> Parse(ParseState &state) const { |
| 917 | const char *start{state.GetLocation()}; |
| 918 | auto result{parser_.Parse(state)}; |
| 919 | if (result) { |
| 920 | const char *end{state.GetLocation()}; |
| 921 | for (; start < end && start[0] == ' '; ++start) { |
| 922 | } |
| 923 | for (; start < end && end[-1] == ' '; --end) { |
| 924 | } |
| 925 | result->source = CharBlock{start, end}; |
| 926 | } |
| 927 | return result; |
| 928 | } |
| 929 | |
| 930 | private: |
| 931 | const PA parser_; |
| 932 | }; |
| 933 | |
| 934 | template <typename PA> inline constexpr auto sourced(PA parser) { |
| 935 | return SourcedParser<PA>{parser}; |
| 936 | } |
| 937 | } // namespace Fortran::parser |
| 938 | #endif // FORTRAN_PARSER_BASIC_PARSERS_H_ |
| 939 | |