1 //===- MicrosoftDemangle.cpp ----------------------------------------------===// 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 // This file defines a demangler for MSVC-style mangled symbols. 10 // 11 // This file has no dependencies on the rest of LLVM so that it can be 12 // easily reused in other programs such as libcxxabi. 13 // 14 //===----------------------------------------------------------------------===// 15 16 #include "llvm/Demangle/MicrosoftDemangle.h" 17 #include "llvm/Demangle/Demangle.h" 18 #include "llvm/Demangle/MicrosoftDemangleNodes.h" 19 20 #include "llvm/Demangle/DemangleConfig.h" 21 #include "llvm/Demangle/StringView.h" 22 #include "llvm/Demangle/Utility.h" 23 24 #include <array> 25 #include <cctype> 26 #include <cstdio> 27 #include <tuple> 28 29 using namespace llvm; 30 using namespace ms_demangle; 31 32 static bool startsWithDigit(StringView S) { 33 return !S.empty() && std::isdigit(S.front()); 34 } 35 36 37 struct NodeList { 38 Node *N = nullptr; 39 NodeList *Next = nullptr; 40 }; 41 42 static bool isMemberPointer(StringView MangledName, bool &Error) { 43 Error = false; 44 switch (MangledName.popFront()) { 45 case '$': 46 // This is probably an rvalue reference (e.g. $$Q), and you cannot have an 47 // rvalue reference to a member. 48 return false; 49 case 'A': 50 // 'A' indicates a reference, and you cannot have a reference to a member 51 // function or member. 52 return false; 53 case 'P': 54 case 'Q': 55 case 'R': 56 case 'S': 57 // These 4 values indicate some kind of pointer, but we still don't know 58 // what. 59 break; 60 default: 61 // isMemberPointer() is called only if isPointerType() returns true, 62 // and it rejects other prefixes. 63 DEMANGLE_UNREACHABLE; 64 } 65 66 // If it starts with a number, then 6 indicates a non-member function 67 // pointer, and 8 indicates a member function pointer. 68 if (startsWithDigit(MangledName)) { 69 if (MangledName[0] != '6' && MangledName[0] != '8') { 70 Error = true; 71 return false; 72 } 73 return (MangledName[0] == '8'); 74 } 75 76 // Remove ext qualifiers since those can appear on either type and are 77 // therefore not indicative. 78 MangledName.consumeFront('E'); // 64-bit 79 MangledName.consumeFront('I'); // restrict 80 MangledName.consumeFront('F'); // unaligned 81 82 if (MangledName.empty()) { 83 Error = true; 84 return false; 85 } 86 87 // The next value should be either ABCD (non-member) or QRST (member). 88 switch (MangledName.front()) { 89 case 'A': 90 case 'B': 91 case 'C': 92 case 'D': 93 return false; 94 case 'Q': 95 case 'R': 96 case 'S': 97 case 'T': 98 return true; 99 default: 100 Error = true; 101 return false; 102 } 103 } 104 105 static SpecialIntrinsicKind 106 consumeSpecialIntrinsicKind(StringView &MangledName) { 107 if (MangledName.consumeFront("?_7")) 108 return SpecialIntrinsicKind::Vftable; 109 if (MangledName.consumeFront("?_8")) 110 return SpecialIntrinsicKind::Vbtable; 111 if (MangledName.consumeFront("?_9")) 112 return SpecialIntrinsicKind::VcallThunk; 113 if (MangledName.consumeFront("?_A")) 114 return SpecialIntrinsicKind::Typeof; 115 if (MangledName.consumeFront("?_B")) 116 return SpecialIntrinsicKind::LocalStaticGuard; 117 if (MangledName.consumeFront("?_C")) 118 return SpecialIntrinsicKind::StringLiteralSymbol; 119 if (MangledName.consumeFront("?_P")) 120 return SpecialIntrinsicKind::UdtReturning; 121 if (MangledName.consumeFront("?_R0")) 122 return SpecialIntrinsicKind::RttiTypeDescriptor; 123 if (MangledName.consumeFront("?_R1")) 124 return SpecialIntrinsicKind::RttiBaseClassDescriptor; 125 if (MangledName.consumeFront("?_R2")) 126 return SpecialIntrinsicKind::RttiBaseClassArray; 127 if (MangledName.consumeFront("?_R3")) 128 return SpecialIntrinsicKind::RttiClassHierarchyDescriptor; 129 if (MangledName.consumeFront("?_R4")) 130 return SpecialIntrinsicKind::RttiCompleteObjLocator; 131 if (MangledName.consumeFront("?_S")) 132 return SpecialIntrinsicKind::LocalVftable; 133 if (MangledName.consumeFront("?__E")) 134 return SpecialIntrinsicKind::DynamicInitializer; 135 if (MangledName.consumeFront("?__F")) 136 return SpecialIntrinsicKind::DynamicAtexitDestructor; 137 if (MangledName.consumeFront("?__J")) 138 return SpecialIntrinsicKind::LocalStaticThreadGuard; 139 return SpecialIntrinsicKind::None; 140 } 141 142 static bool startsWithLocalScopePattern(StringView S) { 143 if (!S.consumeFront('?')) 144 return false; 145 146 size_t End = S.find('?'); 147 if (End == StringView::npos) 148 return false; 149 StringView Candidate = S.substr(0, End); 150 if (Candidate.empty()) 151 return false; 152 153 // \?[0-9]\? 154 // ?@? is the discriminator 0. 155 if (Candidate.size() == 1) 156 return Candidate[0] == '@' || (Candidate[0] >= '0' && Candidate[0] <= '9'); 157 158 // If it's not 0-9, then it's an encoded number terminated with an @ 159 if (Candidate.back() != '@') 160 return false; 161 Candidate = Candidate.dropBack(); 162 163 // An encoded number starts with B-P and all subsequent digits are in A-P. 164 // Note that the reason the first digit cannot be A is two fold. First, it 165 // would create an ambiguity with ?A which delimits the beginning of an 166 // anonymous namespace. Second, A represents 0, and you don't start a multi 167 // digit number with a leading 0. Presumably the anonymous namespace 168 // ambiguity is also why single digit encoded numbers use 0-9 rather than A-J. 169 if (Candidate[0] < 'B' || Candidate[0] > 'P') 170 return false; 171 Candidate = Candidate.dropFront(); 172 while (!Candidate.empty()) { 173 if (Candidate[0] < 'A' || Candidate[0] > 'P') 174 return false; 175 Candidate = Candidate.dropFront(); 176 } 177 178 return true; 179 } 180 181 static bool isTagType(StringView S) { 182 switch (S.front()) { 183 case 'T': // union 184 case 'U': // struct 185 case 'V': // class 186 case 'W': // enum 187 return true; 188 } 189 return false; 190 } 191 192 static bool isCustomType(StringView S) { return S[0] == '?'; } 193 194 static bool isPointerType(StringView S) { 195 if (S.startsWith("$$Q")) // foo && 196 return true; 197 198 switch (S.front()) { 199 case 'A': // foo & 200 case 'P': // foo * 201 case 'Q': // foo *const 202 case 'R': // foo *volatile 203 case 'S': // foo *const volatile 204 return true; 205 } 206 return false; 207 } 208 209 static bool isArrayType(StringView S) { return S[0] == 'Y'; } 210 211 static bool isFunctionType(StringView S) { 212 return S.startsWith("$$A8@@") || S.startsWith("$$A6"); 213 } 214 215 static FunctionRefQualifier 216 demangleFunctionRefQualifier(StringView &MangledName) { 217 if (MangledName.consumeFront('G')) 218 return FunctionRefQualifier::Reference; 219 else if (MangledName.consumeFront('H')) 220 return FunctionRefQualifier::RValueReference; 221 return FunctionRefQualifier::None; 222 } 223 224 static std::pair<Qualifiers, PointerAffinity> 225 demanglePointerCVQualifiers(StringView &MangledName) { 226 if (MangledName.consumeFront("$$Q")) 227 return std::make_pair(Q_None, PointerAffinity::RValueReference); 228 229 switch (MangledName.popFront()) { 230 case 'A': 231 return std::make_pair(Q_None, PointerAffinity::Reference); 232 case 'P': 233 return std::make_pair(Q_None, PointerAffinity::Pointer); 234 case 'Q': 235 return std::make_pair(Q_Const, PointerAffinity::Pointer); 236 case 'R': 237 return std::make_pair(Q_Volatile, PointerAffinity::Pointer); 238 case 'S': 239 return std::make_pair(Qualifiers(Q_Const | Q_Volatile), 240 PointerAffinity::Pointer); 241 } 242 // This function is only called if isPointerType() returns true, 243 // and it only returns true for the six cases listed above. 244 DEMANGLE_UNREACHABLE; 245 } 246 247 StringView Demangler::copyString(StringView Borrowed) { 248 char *Stable = Arena.allocUnalignedBuffer(Borrowed.size() + 1); 249 std::strcpy(Stable, Borrowed.begin()); 250 251 return {Stable, Borrowed.size()}; 252 } 253 254 SpecialTableSymbolNode * 255 Demangler::demangleSpecialTableSymbolNode(StringView &MangledName, 256 SpecialIntrinsicKind K) { 257 NamedIdentifierNode *NI = Arena.alloc<NamedIdentifierNode>(); 258 switch (K) { 259 case SpecialIntrinsicKind::Vftable: 260 NI->Name = "`vftable'"; 261 break; 262 case SpecialIntrinsicKind::Vbtable: 263 NI->Name = "`vbtable'"; 264 break; 265 case SpecialIntrinsicKind::LocalVftable: 266 NI->Name = "`local vftable'"; 267 break; 268 case SpecialIntrinsicKind::RttiCompleteObjLocator: 269 NI->Name = "`RTTI Complete Object Locator'"; 270 break; 271 default: 272 DEMANGLE_UNREACHABLE; 273 } 274 QualifiedNameNode *QN = demangleNameScopeChain(MangledName, NI); 275 SpecialTableSymbolNode *STSN = Arena.alloc<SpecialTableSymbolNode>(); 276 STSN->Name = QN; 277 bool IsMember = false; 278 if (MangledName.empty()) { 279 Error = true; 280 return nullptr; 281 } 282 char Front = MangledName.popFront(); 283 if (Front != '6' && Front != '7') { 284 Error = true; 285 return nullptr; 286 } 287 288 std::tie(STSN->Quals, IsMember) = demangleQualifiers(MangledName); 289 if (!MangledName.consumeFront('@')) 290 STSN->TargetName = demangleFullyQualifiedTypeName(MangledName); 291 return STSN; 292 } 293 294 LocalStaticGuardVariableNode * 295 Demangler::demangleLocalStaticGuard(StringView &MangledName, bool IsThread) { 296 LocalStaticGuardIdentifierNode *LSGI = 297 Arena.alloc<LocalStaticGuardIdentifierNode>(); 298 LSGI->IsThread = IsThread; 299 QualifiedNameNode *QN = demangleNameScopeChain(MangledName, LSGI); 300 LocalStaticGuardVariableNode *LSGVN = 301 Arena.alloc<LocalStaticGuardVariableNode>(); 302 LSGVN->Name = QN; 303 304 if (MangledName.consumeFront("4IA")) 305 LSGVN->IsVisible = false; 306 else if (MangledName.consumeFront("5")) 307 LSGVN->IsVisible = true; 308 else { 309 Error = true; 310 return nullptr; 311 } 312 313 if (!MangledName.empty()) 314 LSGI->ScopeIndex = demangleUnsigned(MangledName); 315 return LSGVN; 316 } 317 318 static NamedIdentifierNode *synthesizeNamedIdentifier(ArenaAllocator &Arena, 319 StringView Name) { 320 NamedIdentifierNode *Id = Arena.alloc<NamedIdentifierNode>(); 321 Id->Name = Name; 322 return Id; 323 } 324 325 static QualifiedNameNode *synthesizeQualifiedName(ArenaAllocator &Arena, 326 IdentifierNode *Identifier) { 327 QualifiedNameNode *QN = Arena.alloc<QualifiedNameNode>(); 328 QN->Components = Arena.alloc<NodeArrayNode>(); 329 QN->Components->Count = 1; 330 QN->Components->Nodes = Arena.allocArray<Node *>(1); 331 QN->Components->Nodes[0] = Identifier; 332 return QN; 333 } 334 335 static QualifiedNameNode *synthesizeQualifiedName(ArenaAllocator &Arena, 336 StringView Name) { 337 NamedIdentifierNode *Id = synthesizeNamedIdentifier(Arena, Name); 338 return synthesizeQualifiedName(Arena, Id); 339 } 340 341 static VariableSymbolNode *synthesizeVariable(ArenaAllocator &Arena, 342 TypeNode *Type, 343 StringView VariableName) { 344 VariableSymbolNode *VSN = Arena.alloc<VariableSymbolNode>(); 345 VSN->Type = Type; 346 VSN->Name = synthesizeQualifiedName(Arena, VariableName); 347 return VSN; 348 } 349 350 VariableSymbolNode *Demangler::demangleUntypedVariable( 351 ArenaAllocator &Arena, StringView &MangledName, StringView VariableName) { 352 NamedIdentifierNode *NI = synthesizeNamedIdentifier(Arena, VariableName); 353 QualifiedNameNode *QN = demangleNameScopeChain(MangledName, NI); 354 VariableSymbolNode *VSN = Arena.alloc<VariableSymbolNode>(); 355 VSN->Name = QN; 356 if (MangledName.consumeFront("8")) 357 return VSN; 358 359 Error = true; 360 return nullptr; 361 } 362 363 VariableSymbolNode * 364 Demangler::demangleRttiBaseClassDescriptorNode(ArenaAllocator &Arena, 365 StringView &MangledName) { 366 RttiBaseClassDescriptorNode *RBCDN = 367 Arena.alloc<RttiBaseClassDescriptorNode>(); 368 RBCDN->NVOffset = demangleUnsigned(MangledName); 369 RBCDN->VBPtrOffset = demangleSigned(MangledName); 370 RBCDN->VBTableOffset = demangleUnsigned(MangledName); 371 RBCDN->Flags = demangleUnsigned(MangledName); 372 if (Error) 373 return nullptr; 374 375 VariableSymbolNode *VSN = Arena.alloc<VariableSymbolNode>(); 376 VSN->Name = demangleNameScopeChain(MangledName, RBCDN); 377 MangledName.consumeFront('8'); 378 return VSN; 379 } 380 381 FunctionSymbolNode *Demangler::demangleInitFiniStub(StringView &MangledName, 382 bool IsDestructor) { 383 DynamicStructorIdentifierNode *DSIN = 384 Arena.alloc<DynamicStructorIdentifierNode>(); 385 DSIN->IsDestructor = IsDestructor; 386 387 bool IsKnownStaticDataMember = false; 388 if (MangledName.consumeFront('?')) 389 IsKnownStaticDataMember = true; 390 391 SymbolNode *Symbol = demangleDeclarator(MangledName); 392 if (Error) 393 return nullptr; 394 395 FunctionSymbolNode *FSN = nullptr; 396 397 if (Symbol->kind() == NodeKind::VariableSymbol) { 398 DSIN->Variable = static_cast<VariableSymbolNode *>(Symbol); 399 400 // Older versions of clang mangled this type of symbol incorrectly. They 401 // would omit the leading ? and they would only emit a single @ at the end. 402 // The correct mangling is a leading ? and 2 trailing @ signs. Handle 403 // both cases. 404 int AtCount = IsKnownStaticDataMember ? 2 : 1; 405 for (int I = 0; I < AtCount; ++I) { 406 if (MangledName.consumeFront('@')) 407 continue; 408 Error = true; 409 return nullptr; 410 } 411 412 FSN = demangleFunctionEncoding(MangledName); 413 if (FSN) 414 FSN->Name = synthesizeQualifiedName(Arena, DSIN); 415 } else { 416 if (IsKnownStaticDataMember) { 417 // This was supposed to be a static data member, but we got a function. 418 Error = true; 419 return nullptr; 420 } 421 422 FSN = static_cast<FunctionSymbolNode *>(Symbol); 423 DSIN->Name = Symbol->Name; 424 FSN->Name = synthesizeQualifiedName(Arena, DSIN); 425 } 426 427 return FSN; 428 } 429 430 SymbolNode *Demangler::demangleSpecialIntrinsic(StringView &MangledName) { 431 SpecialIntrinsicKind SIK = consumeSpecialIntrinsicKind(MangledName); 432 433 switch (SIK) { 434 case SpecialIntrinsicKind::None: 435 return nullptr; 436 case SpecialIntrinsicKind::StringLiteralSymbol: 437 return demangleStringLiteral(MangledName); 438 case SpecialIntrinsicKind::Vftable: 439 case SpecialIntrinsicKind::Vbtable: 440 case SpecialIntrinsicKind::LocalVftable: 441 case SpecialIntrinsicKind::RttiCompleteObjLocator: 442 return demangleSpecialTableSymbolNode(MangledName, SIK); 443 case SpecialIntrinsicKind::VcallThunk: 444 return demangleVcallThunkNode(MangledName); 445 case SpecialIntrinsicKind::LocalStaticGuard: 446 return demangleLocalStaticGuard(MangledName, /*IsThread=*/false); 447 case SpecialIntrinsicKind::LocalStaticThreadGuard: 448 return demangleLocalStaticGuard(MangledName, /*IsThread=*/true); 449 case SpecialIntrinsicKind::RttiTypeDescriptor: { 450 TypeNode *T = demangleType(MangledName, QualifierMangleMode::Result); 451 if (Error) 452 break; 453 if (!MangledName.consumeFront("@8")) 454 break; 455 if (!MangledName.empty()) 456 break; 457 return synthesizeVariable(Arena, T, "`RTTI Type Descriptor'"); 458 } 459 case SpecialIntrinsicKind::RttiBaseClassArray: 460 return demangleUntypedVariable(Arena, MangledName, 461 "`RTTI Base Class Array'"); 462 case SpecialIntrinsicKind::RttiClassHierarchyDescriptor: 463 return demangleUntypedVariable(Arena, MangledName, 464 "`RTTI Class Hierarchy Descriptor'"); 465 case SpecialIntrinsicKind::RttiBaseClassDescriptor: 466 return demangleRttiBaseClassDescriptorNode(Arena, MangledName); 467 case SpecialIntrinsicKind::DynamicInitializer: 468 return demangleInitFiniStub(MangledName, /*IsDestructor=*/false); 469 case SpecialIntrinsicKind::DynamicAtexitDestructor: 470 return demangleInitFiniStub(MangledName, /*IsDestructor=*/true); 471 case SpecialIntrinsicKind::Typeof: 472 case SpecialIntrinsicKind::UdtReturning: 473 // It's unclear which tools produces these manglings, so demangling 474 // support is not (yet?) implemented. 475 break; 476 case SpecialIntrinsicKind::Unknown: 477 DEMANGLE_UNREACHABLE; // Never returned by consumeSpecialIntrinsicKind. 478 } 479 Error = true; 480 return nullptr; 481 } 482 483 IdentifierNode * 484 Demangler::demangleFunctionIdentifierCode(StringView &MangledName) { 485 assert(MangledName.startsWith('?')); 486 MangledName = MangledName.dropFront(); 487 if (MangledName.empty()) { 488 Error = true; 489 return nullptr; 490 } 491 492 if (MangledName.consumeFront("__")) 493 return demangleFunctionIdentifierCode( 494 MangledName, FunctionIdentifierCodeGroup::DoubleUnder); 495 if (MangledName.consumeFront("_")) 496 return demangleFunctionIdentifierCode(MangledName, 497 FunctionIdentifierCodeGroup::Under); 498 return demangleFunctionIdentifierCode(MangledName, 499 FunctionIdentifierCodeGroup::Basic); 500 } 501 502 StructorIdentifierNode * 503 Demangler::demangleStructorIdentifier(StringView &MangledName, 504 bool IsDestructor) { 505 StructorIdentifierNode *N = Arena.alloc<StructorIdentifierNode>(); 506 N->IsDestructor = IsDestructor; 507 return N; 508 } 509 510 ConversionOperatorIdentifierNode * 511 Demangler::demangleConversionOperatorIdentifier(StringView &MangledName) { 512 ConversionOperatorIdentifierNode *N = 513 Arena.alloc<ConversionOperatorIdentifierNode>(); 514 return N; 515 } 516 517 LiteralOperatorIdentifierNode * 518 Demangler::demangleLiteralOperatorIdentifier(StringView &MangledName) { 519 LiteralOperatorIdentifierNode *N = 520 Arena.alloc<LiteralOperatorIdentifierNode>(); 521 N->Name = demangleSimpleString(MangledName, /*Memorize=*/false); 522 return N; 523 } 524 525 IntrinsicFunctionKind 526 Demangler::translateIntrinsicFunctionCode(char CH, 527 FunctionIdentifierCodeGroup Group) { 528 using IFK = IntrinsicFunctionKind; 529 if (!(CH >= '0' && CH <= '9') && !(CH >= 'A' && CH <= 'Z')) { 530 Error = true; 531 return IFK::None; 532 } 533 534 // Not all ? identifiers are intrinsics *functions*. This function only maps 535 // operator codes for the special functions, all others are handled elsewhere, 536 // hence the IFK::None entries in the table. 537 static IFK Basic[36] = { 538 IFK::None, // ?0 # Foo::Foo() 539 IFK::None, // ?1 # Foo::~Foo() 540 IFK::New, // ?2 # operator new 541 IFK::Delete, // ?3 # operator delete 542 IFK::Assign, // ?4 # operator= 543 IFK::RightShift, // ?5 # operator>> 544 IFK::LeftShift, // ?6 # operator<< 545 IFK::LogicalNot, // ?7 # operator! 546 IFK::Equals, // ?8 # operator== 547 IFK::NotEquals, // ?9 # operator!= 548 IFK::ArraySubscript, // ?A # operator[] 549 IFK::None, // ?B # Foo::operator <type>() 550 IFK::Pointer, // ?C # operator-> 551 IFK::Dereference, // ?D # operator* 552 IFK::Increment, // ?E # operator++ 553 IFK::Decrement, // ?F # operator-- 554 IFK::Minus, // ?G # operator- 555 IFK::Plus, // ?H # operator+ 556 IFK::BitwiseAnd, // ?I # operator& 557 IFK::MemberPointer, // ?J # operator->* 558 IFK::Divide, // ?K # operator/ 559 IFK::Modulus, // ?L # operator% 560 IFK::LessThan, // ?M operator< 561 IFK::LessThanEqual, // ?N operator<= 562 IFK::GreaterThan, // ?O operator> 563 IFK::GreaterThanEqual, // ?P operator>= 564 IFK::Comma, // ?Q operator, 565 IFK::Parens, // ?R operator() 566 IFK::BitwiseNot, // ?S operator~ 567 IFK::BitwiseXor, // ?T operator^ 568 IFK::BitwiseOr, // ?U operator| 569 IFK::LogicalAnd, // ?V operator&& 570 IFK::LogicalOr, // ?W operator|| 571 IFK::TimesEqual, // ?X operator*= 572 IFK::PlusEqual, // ?Y operator+= 573 IFK::MinusEqual, // ?Z operator-= 574 }; 575 static IFK Under[36] = { 576 IFK::DivEqual, // ?_0 operator/= 577 IFK::ModEqual, // ?_1 operator%= 578 IFK::RshEqual, // ?_2 operator>>= 579 IFK::LshEqual, // ?_3 operator<<= 580 IFK::BitwiseAndEqual, // ?_4 operator&= 581 IFK::BitwiseOrEqual, // ?_5 operator|= 582 IFK::BitwiseXorEqual, // ?_6 operator^= 583 IFK::None, // ?_7 # vftable 584 IFK::None, // ?_8 # vbtable 585 IFK::None, // ?_9 # vcall 586 IFK::None, // ?_A # typeof 587 IFK::None, // ?_B # local static guard 588 IFK::None, // ?_C # string literal 589 IFK::VbaseDtor, // ?_D # vbase destructor 590 IFK::VecDelDtor, // ?_E # vector deleting destructor 591 IFK::DefaultCtorClosure, // ?_F # default constructor closure 592 IFK::ScalarDelDtor, // ?_G # scalar deleting destructor 593 IFK::VecCtorIter, // ?_H # vector constructor iterator 594 IFK::VecDtorIter, // ?_I # vector destructor iterator 595 IFK::VecVbaseCtorIter, // ?_J # vector vbase constructor iterator 596 IFK::VdispMap, // ?_K # virtual displacement map 597 IFK::EHVecCtorIter, // ?_L # eh vector constructor iterator 598 IFK::EHVecDtorIter, // ?_M # eh vector destructor iterator 599 IFK::EHVecVbaseCtorIter, // ?_N # eh vector vbase constructor iterator 600 IFK::CopyCtorClosure, // ?_O # copy constructor closure 601 IFK::None, // ?_P<name> # udt returning <name> 602 IFK::None, // ?_Q # <unknown> 603 IFK::None, // ?_R0 - ?_R4 # RTTI Codes 604 IFK::None, // ?_S # local vftable 605 IFK::LocalVftableCtorClosure, // ?_T # local vftable constructor closure 606 IFK::ArrayNew, // ?_U operator new[] 607 IFK::ArrayDelete, // ?_V operator delete[] 608 IFK::None, // ?_W <unused> 609 IFK::None, // ?_X <unused> 610 IFK::None, // ?_Y <unused> 611 IFK::None, // ?_Z <unused> 612 }; 613 static IFK DoubleUnder[36] = { 614 IFK::None, // ?__0 <unused> 615 IFK::None, // ?__1 <unused> 616 IFK::None, // ?__2 <unused> 617 IFK::None, // ?__3 <unused> 618 IFK::None, // ?__4 <unused> 619 IFK::None, // ?__5 <unused> 620 IFK::None, // ?__6 <unused> 621 IFK::None, // ?__7 <unused> 622 IFK::None, // ?__8 <unused> 623 IFK::None, // ?__9 <unused> 624 IFK::ManVectorCtorIter, // ?__A managed vector ctor iterator 625 IFK::ManVectorDtorIter, // ?__B managed vector dtor iterator 626 IFK::EHVectorCopyCtorIter, // ?__C EH vector copy ctor iterator 627 IFK::EHVectorVbaseCopyCtorIter, // ?__D EH vector vbase copy ctor iter 628 IFK::None, // ?__E dynamic initializer for `T' 629 IFK::None, // ?__F dynamic atexit destructor for `T' 630 IFK::VectorCopyCtorIter, // ?__G vector copy constructor iter 631 IFK::VectorVbaseCopyCtorIter, // ?__H vector vbase copy ctor iter 632 IFK::ManVectorVbaseCopyCtorIter, // ?__I managed vector vbase copy ctor 633 // iter 634 IFK::None, // ?__J local static thread guard 635 IFK::None, // ?__K operator ""_name 636 IFK::CoAwait, // ?__L operator co_await 637 IFK::Spaceship, // ?__M operator<=> 638 IFK::None, // ?__N <unused> 639 IFK::None, // ?__O <unused> 640 IFK::None, // ?__P <unused> 641 IFK::None, // ?__Q <unused> 642 IFK::None, // ?__R <unused> 643 IFK::None, // ?__S <unused> 644 IFK::None, // ?__T <unused> 645 IFK::None, // ?__U <unused> 646 IFK::None, // ?__V <unused> 647 IFK::None, // ?__W <unused> 648 IFK::None, // ?__X <unused> 649 IFK::None, // ?__Y <unused> 650 IFK::None, // ?__Z <unused> 651 }; 652 653 int Index = (CH >= '0' && CH <= '9') ? (CH - '0') : (CH - 'A' + 10); 654 switch (Group) { 655 case FunctionIdentifierCodeGroup::Basic: 656 return Basic[Index]; 657 case FunctionIdentifierCodeGroup::Under: 658 return Under[Index]; 659 case FunctionIdentifierCodeGroup::DoubleUnder: 660 return DoubleUnder[Index]; 661 } 662 DEMANGLE_UNREACHABLE; 663 } 664 665 IdentifierNode * 666 Demangler::demangleFunctionIdentifierCode(StringView &MangledName, 667 FunctionIdentifierCodeGroup Group) { 668 if (MangledName.empty()) { 669 Error = true; 670 return nullptr; 671 } 672 switch (Group) { 673 case FunctionIdentifierCodeGroup::Basic: 674 switch (char CH = MangledName.popFront()) { 675 case '0': 676 case '1': 677 return demangleStructorIdentifier(MangledName, CH == '1'); 678 case 'B': 679 return demangleConversionOperatorIdentifier(MangledName); 680 default: 681 return Arena.alloc<IntrinsicFunctionIdentifierNode>( 682 translateIntrinsicFunctionCode(CH, Group)); 683 } 684 case FunctionIdentifierCodeGroup::Under: 685 return Arena.alloc<IntrinsicFunctionIdentifierNode>( 686 translateIntrinsicFunctionCode(MangledName.popFront(), Group)); 687 case FunctionIdentifierCodeGroup::DoubleUnder: 688 switch (char CH = MangledName.popFront()) { 689 case 'K': 690 return demangleLiteralOperatorIdentifier(MangledName); 691 default: 692 return Arena.alloc<IntrinsicFunctionIdentifierNode>( 693 translateIntrinsicFunctionCode(CH, Group)); 694 } 695 } 696 697 DEMANGLE_UNREACHABLE; 698 } 699 700 SymbolNode *Demangler::demangleEncodedSymbol(StringView &MangledName, 701 QualifiedNameNode *Name) { 702 if (MangledName.empty()) { 703 Error = true; 704 return nullptr; 705 } 706 707 // Read a variable. 708 switch (MangledName.front()) { 709 case '0': 710 case '1': 711 case '2': 712 case '3': 713 case '4': { 714 StorageClass SC = demangleVariableStorageClass(MangledName); 715 return demangleVariableEncoding(MangledName, SC); 716 } 717 } 718 FunctionSymbolNode *FSN = demangleFunctionEncoding(MangledName); 719 720 IdentifierNode *UQN = Name->getUnqualifiedIdentifier(); 721 if (UQN->kind() == NodeKind::ConversionOperatorIdentifier) { 722 ConversionOperatorIdentifierNode *COIN = 723 static_cast<ConversionOperatorIdentifierNode *>(UQN); 724 if (FSN) 725 COIN->TargetType = FSN->Signature->ReturnType; 726 } 727 return FSN; 728 } 729 730 SymbolNode *Demangler::demangleDeclarator(StringView &MangledName) { 731 // What follows is a main symbol name. This may include namespaces or class 732 // back references. 733 QualifiedNameNode *QN = demangleFullyQualifiedSymbolName(MangledName); 734 if (Error) 735 return nullptr; 736 737 SymbolNode *Symbol = demangleEncodedSymbol(MangledName, QN); 738 if (Error) 739 return nullptr; 740 Symbol->Name = QN; 741 742 IdentifierNode *UQN = QN->getUnqualifiedIdentifier(); 743 if (UQN->kind() == NodeKind::ConversionOperatorIdentifier) { 744 ConversionOperatorIdentifierNode *COIN = 745 static_cast<ConversionOperatorIdentifierNode *>(UQN); 746 if (!COIN->TargetType) { 747 Error = true; 748 return nullptr; 749 } 750 } 751 return Symbol; 752 } 753 754 SymbolNode *Demangler::demangleMD5Name(StringView &MangledName) { 755 assert(MangledName.startsWith("??@")); 756 // This is an MD5 mangled name. We can't demangle it, just return the 757 // mangled name. 758 // An MD5 mangled name is ??@ followed by 32 characters and a terminating @. 759 size_t MD5Last = MangledName.find('@', strlen("??@")); 760 if (MD5Last == StringView::npos) { 761 Error = true; 762 return nullptr; 763 } 764 const char *Start = MangledName.begin(); 765 MangledName = MangledName.dropFront(MD5Last + 1); 766 767 // There are two additional special cases for MD5 names: 768 // 1. For complete object locators where the object name is long enough 769 // for the object to have an MD5 name, the complete object locator is 770 // called ??@...@??_R4@ (with a trailing "??_R4@" instead of the usual 771 // leading "??_R4". This is handled here. 772 // 2. For catchable types, in versions of MSVC before 2015 (<1900) or after 773 // 2017.2 (>= 1914), the catchable type mangling is _CT??@...@??@...@8 774 // instead of_CT??@...@8 with just one MD5 name. Since we don't yet 775 // demangle catchable types anywhere, this isn't handled for MD5 names 776 // either. 777 MangledName.consumeFront("??_R4@"); 778 779 StringView MD5(Start, MangledName.begin()); 780 SymbolNode *S = Arena.alloc<SymbolNode>(NodeKind::Md5Symbol); 781 S->Name = synthesizeQualifiedName(Arena, MD5); 782 783 return S; 784 } 785 786 // Parser entry point. 787 SymbolNode *Demangler::parse(StringView &MangledName) { 788 if (MangledName.startsWith("??@")) 789 return demangleMD5Name(MangledName); 790 791 // MSVC-style mangled symbols must start with '?'. 792 if (!MangledName.startsWith('?')) { 793 Error = true; 794 return nullptr; 795 } 796 797 MangledName.consumeFront('?'); 798 799 // ?$ is a template instantiation, but all other names that start with ? are 800 // operators / special names. 801 if (SymbolNode *SI = demangleSpecialIntrinsic(MangledName)) 802 return SI; 803 804 return demangleDeclarator(MangledName); 805 } 806 807 TagTypeNode *Demangler::parseTagUniqueName(StringView &MangledName) { 808 if (!MangledName.consumeFront(".?A")) 809 return nullptr; 810 MangledName.consumeFront(".?A"); 811 if (MangledName.empty()) 812 return nullptr; 813 814 return demangleClassType(MangledName); 815 } 816 817 // <type-encoding> ::= <storage-class> <variable-type> 818 // <storage-class> ::= 0 # private static member 819 // ::= 1 # protected static member 820 // ::= 2 # public static member 821 // ::= 3 # global 822 // ::= 4 # static local 823 824 VariableSymbolNode *Demangler::demangleVariableEncoding(StringView &MangledName, 825 StorageClass SC) { 826 VariableSymbolNode *VSN = Arena.alloc<VariableSymbolNode>(); 827 828 VSN->Type = demangleType(MangledName, QualifierMangleMode::Drop); 829 VSN->SC = SC; 830 831 if (Error) 832 return nullptr; 833 834 // <variable-type> ::= <type> <cvr-qualifiers> 835 // ::= <type> <pointee-cvr-qualifiers> # pointers, references 836 switch (VSN->Type->kind()) { 837 case NodeKind::PointerType: { 838 PointerTypeNode *PTN = static_cast<PointerTypeNode *>(VSN->Type); 839 840 Qualifiers ExtraChildQuals = Q_None; 841 PTN->Quals = Qualifiers(VSN->Type->Quals | 842 demanglePointerExtQualifiers(MangledName)); 843 844 bool IsMember = false; 845 std::tie(ExtraChildQuals, IsMember) = demangleQualifiers(MangledName); 846 847 if (PTN->ClassParent) { 848 QualifiedNameNode *BackRefName = 849 demangleFullyQualifiedTypeName(MangledName); 850 (void)BackRefName; 851 } 852 PTN->Pointee->Quals = Qualifiers(PTN->Pointee->Quals | ExtraChildQuals); 853 854 break; 855 } 856 default: 857 VSN->Type->Quals = demangleQualifiers(MangledName).first; 858 break; 859 } 860 861 return VSN; 862 } 863 864 // Sometimes numbers are encoded in mangled symbols. For example, 865 // "int (*x)[20]" is a valid C type (x is a pointer to an array of 866 // length 20), so we need some way to embed numbers as part of symbols. 867 // This function parses it. 868 // 869 // <number> ::= [?] <non-negative integer> 870 // 871 // <non-negative integer> ::= <decimal digit> # when 1 <= Number <= 10 872 // ::= <hex digit>+ @ # when Number == 0 or >= 10 873 // 874 // <hex-digit> ::= [A-P] # A = 0, B = 1, ... 875 std::pair<uint64_t, bool> Demangler::demangleNumber(StringView &MangledName) { 876 bool IsNegative = MangledName.consumeFront('?'); 877 878 if (startsWithDigit(MangledName)) { 879 uint64_t Ret = MangledName[0] - '0' + 1; 880 MangledName = MangledName.dropFront(1); 881 return {Ret, IsNegative}; 882 } 883 884 uint64_t Ret = 0; 885 for (size_t i = 0; i < MangledName.size(); ++i) { 886 char C = MangledName[i]; 887 if (C == '@') { 888 MangledName = MangledName.dropFront(i + 1); 889 return {Ret, IsNegative}; 890 } 891 if ('A' <= C && C <= 'P') { 892 Ret = (Ret << 4) + (C - 'A'); 893 continue; 894 } 895 break; 896 } 897 898 Error = true; 899 return {0ULL, false}; 900 } 901 902 uint64_t Demangler::demangleUnsigned(StringView &MangledName) { 903 bool IsNegative = false; 904 uint64_t Number = 0; 905 std::tie(Number, IsNegative) = demangleNumber(MangledName); 906 if (IsNegative) 907 Error = true; 908 return Number; 909 } 910 911 int64_t Demangler::demangleSigned(StringView &MangledName) { 912 bool IsNegative = false; 913 uint64_t Number = 0; 914 std::tie(Number, IsNegative) = demangleNumber(MangledName); 915 if (Number > INT64_MAX) 916 Error = true; 917 int64_t I = static_cast<int64_t>(Number); 918 return IsNegative ? -I : I; 919 } 920 921 // First 10 strings can be referenced by special BackReferences ?0, ?1, ..., ?9. 922 // Memorize it. 923 void Demangler::memorizeString(StringView S) { 924 if (Backrefs.NamesCount >= BackrefContext::Max) 925 return; 926 for (size_t i = 0; i < Backrefs.NamesCount; ++i) 927 if (S == Backrefs.Names[i]->Name) 928 return; 929 NamedIdentifierNode *N = Arena.alloc<NamedIdentifierNode>(); 930 N->Name = S; 931 Backrefs.Names[Backrefs.NamesCount++] = N; 932 } 933 934 NamedIdentifierNode *Demangler::demangleBackRefName(StringView &MangledName) { 935 assert(startsWithDigit(MangledName)); 936 937 size_t I = MangledName[0] - '0'; 938 if (I >= Backrefs.NamesCount) { 939 Error = true; 940 return nullptr; 941 } 942 943 MangledName = MangledName.dropFront(); 944 return Backrefs.Names[I]; 945 } 946 947 void Demangler::memorizeIdentifier(IdentifierNode *Identifier) { 948 // Render this class template name into a string buffer so that we can 949 // memorize it for the purpose of back-referencing. 950 OutputStream OS; 951 if (!initializeOutputStream(nullptr, nullptr, OS, 1024)) 952 // FIXME: Propagate out-of-memory as an error? 953 std::terminate(); 954 Identifier->output(OS, OF_Default); 955 OS << '\0'; 956 char *Name = OS.getBuffer(); 957 958 StringView Owned = copyString(Name); 959 memorizeString(Owned); 960 std::free(Name); 961 } 962 963 IdentifierNode * 964 Demangler::demangleTemplateInstantiationName(StringView &MangledName, 965 NameBackrefBehavior NBB) { 966 assert(MangledName.startsWith("?$")); 967 MangledName.consumeFront("?$"); 968 969 BackrefContext OuterContext; 970 std::swap(OuterContext, Backrefs); 971 972 IdentifierNode *Identifier = 973 demangleUnqualifiedSymbolName(MangledName, NBB_Simple); 974 if (!Error) 975 Identifier->TemplateParams = demangleTemplateParameterList(MangledName); 976 977 std::swap(OuterContext, Backrefs); 978 if (Error) 979 return nullptr; 980 981 if (NBB & NBB_Template) { 982 // NBB_Template is only set for types and non-leaf names ("a::" in "a::b"). 983 // Structors and conversion operators only makes sense in a leaf name, so 984 // reject them in NBB_Template contexts. 985 if (Identifier->kind() == NodeKind::ConversionOperatorIdentifier || 986 Identifier->kind() == NodeKind::StructorIdentifier) { 987 Error = true; 988 return nullptr; 989 } 990 991 memorizeIdentifier(Identifier); 992 } 993 994 return Identifier; 995 } 996 997 NamedIdentifierNode *Demangler::demangleSimpleName(StringView &MangledName, 998 bool Memorize) { 999 StringView S = demangleSimpleString(MangledName, Memorize); 1000 if (Error) 1001 return nullptr; 1002 1003 NamedIdentifierNode *Name = Arena.alloc<NamedIdentifierNode>(); 1004 Name->Name = S; 1005 return Name; 1006 } 1007 1008 static bool isRebasedHexDigit(char C) { return (C >= 'A' && C <= 'P'); } 1009 1010 static uint8_t rebasedHexDigitToNumber(char C) { 1011 assert(isRebasedHexDigit(C)); 1012 return (C <= 'J') ? (C - 'A') : (10 + C - 'K'); 1013 } 1014 1015 uint8_t Demangler::demangleCharLiteral(StringView &MangledName) { 1016 assert(!MangledName.empty()); 1017 if (!MangledName.startsWith('?')) 1018 return MangledName.popFront(); 1019 1020 MangledName = MangledName.dropFront(); 1021 if (MangledName.empty()) 1022 goto CharLiteralError; 1023 1024 if (MangledName.consumeFront('$')) { 1025 // Two hex digits 1026 if (MangledName.size() < 2) 1027 goto CharLiteralError; 1028 StringView Nibbles = MangledName.substr(0, 2); 1029 if (!isRebasedHexDigit(Nibbles[0]) || !isRebasedHexDigit(Nibbles[1])) 1030 goto CharLiteralError; 1031 // Don't append the null terminator. 1032 uint8_t C1 = rebasedHexDigitToNumber(Nibbles[0]); 1033 uint8_t C2 = rebasedHexDigitToNumber(Nibbles[1]); 1034 MangledName = MangledName.dropFront(2); 1035 return (C1 << 4) | C2; 1036 } 1037 1038 if (startsWithDigit(MangledName)) { 1039 const char *Lookup = ",/\\:. \n\t'-"; 1040 char C = Lookup[MangledName[0] - '0']; 1041 MangledName = MangledName.dropFront(); 1042 return C; 1043 } 1044 1045 if (MangledName[0] >= 'a' && MangledName[0] <= 'z') { 1046 char Lookup[26] = {'\xE1', '\xE2', '\xE3', '\xE4', '\xE5', '\xE6', '\xE7', 1047 '\xE8', '\xE9', '\xEA', '\xEB', '\xEC', '\xED', '\xEE', 1048 '\xEF', '\xF0', '\xF1', '\xF2', '\xF3', '\xF4', '\xF5', 1049 '\xF6', '\xF7', '\xF8', '\xF9', '\xFA'}; 1050 char C = Lookup[MangledName[0] - 'a']; 1051 MangledName = MangledName.dropFront(); 1052 return C; 1053 } 1054 1055 if (MangledName[0] >= 'A' && MangledName[0] <= 'Z') { 1056 char Lookup[26] = {'\xC1', '\xC2', '\xC3', '\xC4', '\xC5', '\xC6', '\xC7', 1057 '\xC8', '\xC9', '\xCA', '\xCB', '\xCC', '\xCD', '\xCE', 1058 '\xCF', '\xD0', '\xD1', '\xD2', '\xD3', '\xD4', '\xD5', 1059 '\xD6', '\xD7', '\xD8', '\xD9', '\xDA'}; 1060 char C = Lookup[MangledName[0] - 'A']; 1061 MangledName = MangledName.dropFront(); 1062 return C; 1063 } 1064 1065 CharLiteralError: 1066 Error = true; 1067 return '\0'; 1068 } 1069 1070 wchar_t Demangler::demangleWcharLiteral(StringView &MangledName) { 1071 uint8_t C1, C2; 1072 1073 C1 = demangleCharLiteral(MangledName); 1074 if (Error || MangledName.empty()) 1075 goto WCharLiteralError; 1076 C2 = demangleCharLiteral(MangledName); 1077 if (Error) 1078 goto WCharLiteralError; 1079 1080 return ((wchar_t)C1 << 8) | (wchar_t)C2; 1081 1082 WCharLiteralError: 1083 Error = true; 1084 return L'\0'; 1085 } 1086 1087 static void writeHexDigit(char *Buffer, uint8_t Digit) { 1088 assert(Digit <= 15); 1089 *Buffer = (Digit < 10) ? ('0' + Digit) : ('A' + Digit - 10); 1090 } 1091 1092 static void outputHex(OutputStream &OS, unsigned C) { 1093 assert (C != 0); 1094 1095 // It's easier to do the math if we can work from right to left, but we need 1096 // to print the numbers from left to right. So render this into a temporary 1097 // buffer first, then output the temporary buffer. Each byte is of the form 1098 // \xAB, which means that each byte needs 4 characters. Since there are at 1099 // most 4 bytes, we need a 4*4+1 = 17 character temporary buffer. 1100 char TempBuffer[17]; 1101 1102 ::memset(TempBuffer, 0, sizeof(TempBuffer)); 1103 constexpr int MaxPos = sizeof(TempBuffer) - 1; 1104 1105 int Pos = MaxPos - 1; // TempBuffer[MaxPos] is the terminating \0. 1106 while (C != 0) { 1107 for (int I = 0; I < 2; ++I) { 1108 writeHexDigit(&TempBuffer[Pos--], C % 16); 1109 C /= 16; 1110 } 1111 } 1112 TempBuffer[Pos--] = 'x'; 1113 assert(Pos >= 0); 1114 TempBuffer[Pos--] = '\\'; 1115 OS << StringView(&TempBuffer[Pos + 1]); 1116 } 1117 1118 static void outputEscapedChar(OutputStream &OS, unsigned C) { 1119 switch (C) { 1120 case '\0': // nul 1121 OS << "\\0"; 1122 return; 1123 case '\'': // single quote 1124 OS << "\\\'"; 1125 return; 1126 case '\"': // double quote 1127 OS << "\\\""; 1128 return; 1129 case '\\': // backslash 1130 OS << "\\\\"; 1131 return; 1132 case '\a': // bell 1133 OS << "\\a"; 1134 return; 1135 case '\b': // backspace 1136 OS << "\\b"; 1137 return; 1138 case '\f': // form feed 1139 OS << "\\f"; 1140 return; 1141 case '\n': // new line 1142 OS << "\\n"; 1143 return; 1144 case '\r': // carriage return 1145 OS << "\\r"; 1146 return; 1147 case '\t': // tab 1148 OS << "\\t"; 1149 return; 1150 case '\v': // vertical tab 1151 OS << "\\v"; 1152 return; 1153 default: 1154 break; 1155 } 1156 1157 if (C > 0x1F && C < 0x7F) { 1158 // Standard ascii char. 1159 OS << (char)C; 1160 return; 1161 } 1162 1163 outputHex(OS, C); 1164 } 1165 1166 static unsigned countTrailingNullBytes(const uint8_t *StringBytes, int Length) { 1167 const uint8_t *End = StringBytes + Length - 1; 1168 unsigned Count = 0; 1169 while (Length > 0 && *End == 0) { 1170 --Length; 1171 --End; 1172 ++Count; 1173 } 1174 return Count; 1175 } 1176 1177 static unsigned countEmbeddedNulls(const uint8_t *StringBytes, 1178 unsigned Length) { 1179 unsigned Result = 0; 1180 for (unsigned I = 0; I < Length; ++I) { 1181 if (*StringBytes++ == 0) 1182 ++Result; 1183 } 1184 return Result; 1185 } 1186 1187 // A mangled (non-wide) string literal stores the total length of the string it 1188 // refers to (passed in NumBytes), and it contains up to 32 bytes of actual text 1189 // (passed in StringBytes, NumChars). 1190 static unsigned guessCharByteSize(const uint8_t *StringBytes, unsigned NumChars, 1191 uint64_t NumBytes) { 1192 assert(NumBytes > 0); 1193 1194 // If the number of bytes is odd, this is guaranteed to be a char string. 1195 if (NumBytes % 2 == 1) 1196 return 1; 1197 1198 // All strings can encode at most 32 bytes of data. If it's less than that, 1199 // then we encoded the entire string. In this case we check for a 1-byte, 1200 // 2-byte, or 4-byte null terminator. 1201 if (NumBytes < 32) { 1202 unsigned TrailingNulls = countTrailingNullBytes(StringBytes, NumChars); 1203 if (TrailingNulls >= 4 && NumBytes % 4 == 0) 1204 return 4; 1205 if (TrailingNulls >= 2) 1206 return 2; 1207 return 1; 1208 } 1209 1210 // The whole string was not able to be encoded. Try to look at embedded null 1211 // terminators to guess. The heuristic is that we count all embedded null 1212 // terminators. If more than 2/3 are null, it's a char32. If more than 1/3 1213 // are null, it's a char16. Otherwise it's a char8. This obviously isn't 1214 // perfect and is biased towards languages that have ascii alphabets, but this 1215 // was always going to be best effort since the encoding is lossy. 1216 unsigned Nulls = countEmbeddedNulls(StringBytes, NumChars); 1217 if (Nulls >= 2 * NumChars / 3 && NumBytes % 4 == 0) 1218 return 4; 1219 if (Nulls >= NumChars / 3) 1220 return 2; 1221 return 1; 1222 } 1223 1224 static unsigned decodeMultiByteChar(const uint8_t *StringBytes, 1225 unsigned CharIndex, unsigned CharBytes) { 1226 assert(CharBytes == 1 || CharBytes == 2 || CharBytes == 4); 1227 unsigned Offset = CharIndex * CharBytes; 1228 unsigned Result = 0; 1229 StringBytes = StringBytes + Offset; 1230 for (unsigned I = 0; I < CharBytes; ++I) { 1231 unsigned C = static_cast<unsigned>(StringBytes[I]); 1232 Result |= C << (8 * I); 1233 } 1234 return Result; 1235 } 1236 1237 FunctionSymbolNode *Demangler::demangleVcallThunkNode(StringView &MangledName) { 1238 FunctionSymbolNode *FSN = Arena.alloc<FunctionSymbolNode>(); 1239 VcallThunkIdentifierNode *VTIN = Arena.alloc<VcallThunkIdentifierNode>(); 1240 FSN->Signature = Arena.alloc<ThunkSignatureNode>(); 1241 FSN->Signature->FunctionClass = FC_NoParameterList; 1242 1243 FSN->Name = demangleNameScopeChain(MangledName, VTIN); 1244 if (!Error) 1245 Error = !MangledName.consumeFront("$B"); 1246 if (!Error) 1247 VTIN->OffsetInVTable = demangleUnsigned(MangledName); 1248 if (!Error) 1249 Error = !MangledName.consumeFront('A'); 1250 if (!Error) 1251 FSN->Signature->CallConvention = demangleCallingConvention(MangledName); 1252 return (Error) ? nullptr : FSN; 1253 } 1254 1255 EncodedStringLiteralNode * 1256 Demangler::demangleStringLiteral(StringView &MangledName) { 1257 // This function uses goto, so declare all variables up front. 1258 OutputStream OS; 1259 StringView CRC; 1260 uint64_t StringByteSize; 1261 bool IsWcharT = false; 1262 bool IsNegative = false; 1263 size_t CrcEndPos = 0; 1264 char *ResultBuffer = nullptr; 1265 1266 EncodedStringLiteralNode *Result = Arena.alloc<EncodedStringLiteralNode>(); 1267 1268 // Must happen before the first `goto StringLiteralError`. 1269 if (!initializeOutputStream(nullptr, nullptr, OS, 1024)) 1270 // FIXME: Propagate out-of-memory as an error? 1271 std::terminate(); 1272 1273 // Prefix indicating the beginning of a string literal 1274 if (!MangledName.consumeFront("@_")) 1275 goto StringLiteralError; 1276 if (MangledName.empty()) 1277 goto StringLiteralError; 1278 1279 // Char Type (regular or wchar_t) 1280 switch (MangledName.popFront()) { 1281 case '1': 1282 IsWcharT = true; 1283 DEMANGLE_FALLTHROUGH; 1284 case '0': 1285 break; 1286 default: 1287 goto StringLiteralError; 1288 } 1289 1290 // Encoded Length 1291 std::tie(StringByteSize, IsNegative) = demangleNumber(MangledName); 1292 if (Error || IsNegative || StringByteSize < (IsWcharT ? 2 : 1)) 1293 goto StringLiteralError; 1294 1295 // CRC 32 (always 8 characters plus a terminator) 1296 CrcEndPos = MangledName.find('@'); 1297 if (CrcEndPos == StringView::npos) 1298 goto StringLiteralError; 1299 CRC = MangledName.substr(0, CrcEndPos); 1300 MangledName = MangledName.dropFront(CrcEndPos + 1); 1301 if (MangledName.empty()) 1302 goto StringLiteralError; 1303 1304 if (IsWcharT) { 1305 Result->Char = CharKind::Wchar; 1306 if (StringByteSize > 64) 1307 Result->IsTruncated = true; 1308 1309 while (!MangledName.consumeFront('@')) { 1310 if (MangledName.size() < 2) 1311 goto StringLiteralError; 1312 wchar_t W = demangleWcharLiteral(MangledName); 1313 if (StringByteSize != 2 || Result->IsTruncated) 1314 outputEscapedChar(OS, W); 1315 StringByteSize -= 2; 1316 if (Error) 1317 goto StringLiteralError; 1318 } 1319 } else { 1320 // The max byte length is actually 32, but some compilers mangled strings 1321 // incorrectly, so we have to assume it can go higher. 1322 constexpr unsigned MaxStringByteLength = 32 * 4; 1323 uint8_t StringBytes[MaxStringByteLength]; 1324 1325 unsigned BytesDecoded = 0; 1326 while (!MangledName.consumeFront('@')) { 1327 if (MangledName.size() < 1 || BytesDecoded >= MaxStringByteLength) 1328 goto StringLiteralError; 1329 StringBytes[BytesDecoded++] = demangleCharLiteral(MangledName); 1330 } 1331 1332 if (StringByteSize > BytesDecoded) 1333 Result->IsTruncated = true; 1334 1335 unsigned CharBytes = 1336 guessCharByteSize(StringBytes, BytesDecoded, StringByteSize); 1337 assert(StringByteSize % CharBytes == 0); 1338 switch (CharBytes) { 1339 case 1: 1340 Result->Char = CharKind::Char; 1341 break; 1342 case 2: 1343 Result->Char = CharKind::Char16; 1344 break; 1345 case 4: 1346 Result->Char = CharKind::Char32; 1347 break; 1348 default: 1349 DEMANGLE_UNREACHABLE; 1350 } 1351 const unsigned NumChars = BytesDecoded / CharBytes; 1352 for (unsigned CharIndex = 0; CharIndex < NumChars; ++CharIndex) { 1353 unsigned NextChar = 1354 decodeMultiByteChar(StringBytes, CharIndex, CharBytes); 1355 if (CharIndex + 1 < NumChars || Result->IsTruncated) 1356 outputEscapedChar(OS, NextChar); 1357 } 1358 } 1359 1360 OS << '\0'; 1361 ResultBuffer = OS.getBuffer(); 1362 Result->DecodedString = copyString(ResultBuffer); 1363 std::free(ResultBuffer); 1364 return Result; 1365 1366 StringLiteralError: 1367 Error = true; 1368 std::free(OS.getBuffer()); 1369 return nullptr; 1370 } 1371 1372 // Returns MangledName's prefix before the first '@', or an error if 1373 // MangledName contains no '@' or the prefix has length 0. 1374 StringView Demangler::demangleSimpleString(StringView &MangledName, 1375 bool Memorize) { 1376 StringView S; 1377 for (size_t i = 0; i < MangledName.size(); ++i) { 1378 if (MangledName[i] != '@') 1379 continue; 1380 if (i == 0) 1381 break; 1382 S = MangledName.substr(0, i); 1383 MangledName = MangledName.dropFront(i + 1); 1384 1385 if (Memorize) 1386 memorizeString(S); 1387 return S; 1388 } 1389 1390 Error = true; 1391 return {}; 1392 } 1393 1394 NamedIdentifierNode * 1395 Demangler::demangleAnonymousNamespaceName(StringView &MangledName) { 1396 assert(MangledName.startsWith("?A")); 1397 MangledName.consumeFront("?A"); 1398 1399 NamedIdentifierNode *Node = Arena.alloc<NamedIdentifierNode>(); 1400 Node->Name = "`anonymous namespace'"; 1401 size_t EndPos = MangledName.find('@'); 1402 if (EndPos == StringView::npos) { 1403 Error = true; 1404 return nullptr; 1405 } 1406 StringView NamespaceKey = MangledName.substr(0, EndPos); 1407 memorizeString(NamespaceKey); 1408 MangledName = MangledName.substr(EndPos + 1); 1409 return Node; 1410 } 1411 1412 NamedIdentifierNode * 1413 Demangler::demangleLocallyScopedNamePiece(StringView &MangledName) { 1414 assert(startsWithLocalScopePattern(MangledName)); 1415 1416 NamedIdentifierNode *Identifier = Arena.alloc<NamedIdentifierNode>(); 1417 MangledName.consumeFront('?'); 1418 uint64_t Number = 0; 1419 bool IsNegative = false; 1420 std::tie(Number, IsNegative) = demangleNumber(MangledName); 1421 assert(!IsNegative); 1422 1423 // One ? to terminate the number 1424 MangledName.consumeFront('?'); 1425 1426 assert(!Error); 1427 Node *Scope = parse(MangledName); 1428 if (Error) 1429 return nullptr; 1430 1431 // Render the parent symbol's name into a buffer. 1432 OutputStream OS; 1433 if (!initializeOutputStream(nullptr, nullptr, OS, 1024)) 1434 // FIXME: Propagate out-of-memory as an error? 1435 std::terminate(); 1436 OS << '`'; 1437 Scope->output(OS, OF_Default); 1438 OS << '\''; 1439 OS << "::`" << Number << "'"; 1440 OS << '\0'; 1441 char *Result = OS.getBuffer(); 1442 Identifier->Name = copyString(Result); 1443 std::free(Result); 1444 return Identifier; 1445 } 1446 1447 // Parses a type name in the form of A@B@C@@ which represents C::B::A. 1448 QualifiedNameNode * 1449 Demangler::demangleFullyQualifiedTypeName(StringView &MangledName) { 1450 IdentifierNode *Identifier = 1451 demangleUnqualifiedTypeName(MangledName, /*Memorize=*/true); 1452 if (Error) 1453 return nullptr; 1454 assert(Identifier); 1455 1456 QualifiedNameNode *QN = demangleNameScopeChain(MangledName, Identifier); 1457 if (Error) 1458 return nullptr; 1459 assert(QN); 1460 return QN; 1461 } 1462 1463 // Parses a symbol name in the form of A@B@C@@ which represents C::B::A. 1464 // Symbol names have slightly different rules regarding what can appear 1465 // so we separate out the implementations for flexibility. 1466 QualifiedNameNode * 1467 Demangler::demangleFullyQualifiedSymbolName(StringView &MangledName) { 1468 // This is the final component of a symbol name (i.e. the leftmost component 1469 // of a mangled name. Since the only possible template instantiation that 1470 // can appear in this context is a function template, and since those are 1471 // not saved for the purposes of name backreferences, only backref simple 1472 // names. 1473 IdentifierNode *Identifier = 1474 demangleUnqualifiedSymbolName(MangledName, NBB_Simple); 1475 if (Error) 1476 return nullptr; 1477 1478 QualifiedNameNode *QN = demangleNameScopeChain(MangledName, Identifier); 1479 if (Error) 1480 return nullptr; 1481 1482 if (Identifier->kind() == NodeKind::StructorIdentifier) { 1483 if (QN->Components->Count < 2) { 1484 Error = true; 1485 return nullptr; 1486 } 1487 StructorIdentifierNode *SIN = 1488 static_cast<StructorIdentifierNode *>(Identifier); 1489 Node *ClassNode = QN->Components->Nodes[QN->Components->Count - 2]; 1490 SIN->Class = static_cast<IdentifierNode *>(ClassNode); 1491 } 1492 assert(QN); 1493 return QN; 1494 } 1495 1496 IdentifierNode *Demangler::demangleUnqualifiedTypeName(StringView &MangledName, 1497 bool Memorize) { 1498 // An inner-most name can be a back-reference, because a fully-qualified name 1499 // (e.g. Scope + Inner) can contain other fully qualified names inside of 1500 // them (for example template parameters), and these nested parameters can 1501 // refer to previously mangled types. 1502 if (startsWithDigit(MangledName)) 1503 return demangleBackRefName(MangledName); 1504 1505 if (MangledName.startsWith("?$")) 1506 return demangleTemplateInstantiationName(MangledName, NBB_Template); 1507 1508 return demangleSimpleName(MangledName, Memorize); 1509 } 1510 1511 IdentifierNode * 1512 Demangler::demangleUnqualifiedSymbolName(StringView &MangledName, 1513 NameBackrefBehavior NBB) { 1514 if (startsWithDigit(MangledName)) 1515 return demangleBackRefName(MangledName); 1516 if (MangledName.startsWith("?$")) 1517 return demangleTemplateInstantiationName(MangledName, NBB); 1518 if (MangledName.startsWith('?')) 1519 return demangleFunctionIdentifierCode(MangledName); 1520 return demangleSimpleName(MangledName, /*Memorize=*/(NBB & NBB_Simple) != 0); 1521 } 1522 1523 IdentifierNode *Demangler::demangleNameScopePiece(StringView &MangledName) { 1524 if (startsWithDigit(MangledName)) 1525 return demangleBackRefName(MangledName); 1526 1527 if (MangledName.startsWith("?$")) 1528 return demangleTemplateInstantiationName(MangledName, NBB_Template); 1529 1530 if (MangledName.startsWith("?A")) 1531 return demangleAnonymousNamespaceName(MangledName); 1532 1533 if (startsWithLocalScopePattern(MangledName)) 1534 return demangleLocallyScopedNamePiece(MangledName); 1535 1536 return demangleSimpleName(MangledName, /*Memorize=*/true); 1537 } 1538 1539 static NodeArrayNode *nodeListToNodeArray(ArenaAllocator &Arena, NodeList *Head, 1540 size_t Count) { 1541 NodeArrayNode *N = Arena.alloc<NodeArrayNode>(); 1542 N->Count = Count; 1543 N->Nodes = Arena.allocArray<Node *>(Count); 1544 for (size_t I = 0; I < Count; ++I) { 1545 N->Nodes[I] = Head->N; 1546 Head = Head->Next; 1547 } 1548 return N; 1549 } 1550 1551 QualifiedNameNode * 1552 Demangler::demangleNameScopeChain(StringView &MangledName, 1553 IdentifierNode *UnqualifiedName) { 1554 NodeList *Head = Arena.alloc<NodeList>(); 1555 1556 Head->N = UnqualifiedName; 1557 1558 size_t Count = 1; 1559 while (!MangledName.consumeFront("@")) { 1560 ++Count; 1561 NodeList *NewHead = Arena.alloc<NodeList>(); 1562 NewHead->Next = Head; 1563 Head = NewHead; 1564 1565 if (MangledName.empty()) { 1566 Error = true; 1567 return nullptr; 1568 } 1569 1570 assert(!Error); 1571 IdentifierNode *Elem = demangleNameScopePiece(MangledName); 1572 if (Error) 1573 return nullptr; 1574 1575 Head->N = Elem; 1576 } 1577 1578 QualifiedNameNode *QN = Arena.alloc<QualifiedNameNode>(); 1579 QN->Components = nodeListToNodeArray(Arena, Head, Count); 1580 return QN; 1581 } 1582 1583 FuncClass Demangler::demangleFunctionClass(StringView &MangledName) { 1584 switch (MangledName.popFront()) { 1585 case '9': 1586 return FuncClass(FC_ExternC | FC_NoParameterList); 1587 case 'A': 1588 return FC_Private; 1589 case 'B': 1590 return FuncClass(FC_Private | FC_Far); 1591 case 'C': 1592 return FuncClass(FC_Private | FC_Static); 1593 case 'D': 1594 return FuncClass(FC_Private | FC_Static | FC_Far); 1595 case 'E': 1596 return FuncClass(FC_Private | FC_Virtual); 1597 case 'F': 1598 return FuncClass(FC_Private | FC_Virtual | FC_Far); 1599 case 'G': 1600 return FuncClass(FC_Private | FC_StaticThisAdjust); 1601 case 'H': 1602 return FuncClass(FC_Private | FC_StaticThisAdjust | FC_Far); 1603 case 'I': 1604 return FuncClass(FC_Protected); 1605 case 'J': 1606 return FuncClass(FC_Protected | FC_Far); 1607 case 'K': 1608 return FuncClass(FC_Protected | FC_Static); 1609 case 'L': 1610 return FuncClass(FC_Protected | FC_Static | FC_Far); 1611 case 'M': 1612 return FuncClass(FC_Protected | FC_Virtual); 1613 case 'N': 1614 return FuncClass(FC_Protected | FC_Virtual | FC_Far); 1615 case 'O': 1616 return FuncClass(FC_Protected | FC_Virtual | FC_StaticThisAdjust); 1617 case 'P': 1618 return FuncClass(FC_Protected | FC_Virtual | FC_StaticThisAdjust | FC_Far); 1619 case 'Q': 1620 return FuncClass(FC_Public); 1621 case 'R': 1622 return FuncClass(FC_Public | FC_Far); 1623 case 'S': 1624 return FuncClass(FC_Public | FC_Static); 1625 case 'T': 1626 return FuncClass(FC_Public | FC_Static | FC_Far); 1627 case 'U': 1628 return FuncClass(FC_Public | FC_Virtual); 1629 case 'V': 1630 return FuncClass(FC_Public | FC_Virtual | FC_Far); 1631 case 'W': 1632 return FuncClass(FC_Public | FC_Virtual | FC_StaticThisAdjust); 1633 case 'X': 1634 return FuncClass(FC_Public | FC_Virtual | FC_StaticThisAdjust | FC_Far); 1635 case 'Y': 1636 return FuncClass(FC_Global); 1637 case 'Z': 1638 return FuncClass(FC_Global | FC_Far); 1639 case '$': { 1640 FuncClass VFlag = FC_VirtualThisAdjust; 1641 if (MangledName.consumeFront('R')) 1642 VFlag = FuncClass(VFlag | FC_VirtualThisAdjustEx); 1643 if (MangledName.empty()) 1644 break; 1645 switch (MangledName.popFront()) { 1646 case '0': 1647 return FuncClass(FC_Private | FC_Virtual | VFlag); 1648 case '1': 1649 return FuncClass(FC_Private | FC_Virtual | VFlag | FC_Far); 1650 case '2': 1651 return FuncClass(FC_Protected | FC_Virtual | VFlag); 1652 case '3': 1653 return FuncClass(FC_Protected | FC_Virtual | VFlag | FC_Far); 1654 case '4': 1655 return FuncClass(FC_Public | FC_Virtual | VFlag); 1656 case '5': 1657 return FuncClass(FC_Public | FC_Virtual | VFlag | FC_Far); 1658 } 1659 } 1660 } 1661 1662 Error = true; 1663 return FC_Public; 1664 } 1665 1666 CallingConv Demangler::demangleCallingConvention(StringView &MangledName) { 1667 if (MangledName.empty()) { 1668 Error = true; 1669 return CallingConv::None; 1670 } 1671 1672 switch (MangledName.popFront()) { 1673 case 'A': 1674 case 'B': 1675 return CallingConv::Cdecl; 1676 case 'C': 1677 case 'D': 1678 return CallingConv::Pascal; 1679 case 'E': 1680 case 'F': 1681 return CallingConv::Thiscall; 1682 case 'G': 1683 case 'H': 1684 return CallingConv::Stdcall; 1685 case 'I': 1686 case 'J': 1687 return CallingConv::Fastcall; 1688 case 'M': 1689 case 'N': 1690 return CallingConv::Clrcall; 1691 case 'O': 1692 case 'P': 1693 return CallingConv::Eabi; 1694 case 'Q': 1695 return CallingConv::Vectorcall; 1696 } 1697 1698 return CallingConv::None; 1699 } 1700 1701 StorageClass Demangler::demangleVariableStorageClass(StringView &MangledName) { 1702 assert(MangledName.front() >= '0' && MangledName.front() <= '4'); 1703 1704 switch (MangledName.popFront()) { 1705 case '0': 1706 return StorageClass::PrivateStatic; 1707 case '1': 1708 return StorageClass::ProtectedStatic; 1709 case '2': 1710 return StorageClass::PublicStatic; 1711 case '3': 1712 return StorageClass::Global; 1713 case '4': 1714 return StorageClass::FunctionLocalStatic; 1715 } 1716 DEMANGLE_UNREACHABLE; 1717 } 1718 1719 std::pair<Qualifiers, bool> 1720 Demangler::demangleQualifiers(StringView &MangledName) { 1721 if (MangledName.empty()) { 1722 Error = true; 1723 return std::make_pair(Q_None, false); 1724 } 1725 1726 switch (MangledName.popFront()) { 1727 // Member qualifiers 1728 case 'Q': 1729 return std::make_pair(Q_None, true); 1730 case 'R': 1731 return std::make_pair(Q_Const, true); 1732 case 'S': 1733 return std::make_pair(Q_Volatile, true); 1734 case 'T': 1735 return std::make_pair(Qualifiers(Q_Const | Q_Volatile), true); 1736 // Non-Member qualifiers 1737 case 'A': 1738 return std::make_pair(Q_None, false); 1739 case 'B': 1740 return std::make_pair(Q_Const, false); 1741 case 'C': 1742 return std::make_pair(Q_Volatile, false); 1743 case 'D': 1744 return std::make_pair(Qualifiers(Q_Const | Q_Volatile), false); 1745 } 1746 Error = true; 1747 return std::make_pair(Q_None, false); 1748 } 1749 1750 // <variable-type> ::= <type> <cvr-qualifiers> 1751 // ::= <type> <pointee-cvr-qualifiers> # pointers, references 1752 TypeNode *Demangler::demangleType(StringView &MangledName, 1753 QualifierMangleMode QMM) { 1754 Qualifiers Quals = Q_None; 1755 bool IsMember = false; 1756 if (QMM == QualifierMangleMode::Mangle) { 1757 std::tie(Quals, IsMember) = demangleQualifiers(MangledName); 1758 } else if (QMM == QualifierMangleMode::Result) { 1759 if (MangledName.consumeFront('?')) 1760 std::tie(Quals, IsMember) = demangleQualifiers(MangledName); 1761 } 1762 1763 if (MangledName.empty()) { 1764 Error = true; 1765 return nullptr; 1766 } 1767 1768 TypeNode *Ty = nullptr; 1769 if (isTagType(MangledName)) 1770 Ty = demangleClassType(MangledName); 1771 else if (isPointerType(MangledName)) { 1772 if (isMemberPointer(MangledName, Error)) 1773 Ty = demangleMemberPointerType(MangledName); 1774 else if (!Error) 1775 Ty = demanglePointerType(MangledName); 1776 else 1777 return nullptr; 1778 } else if (isArrayType(MangledName)) 1779 Ty = demangleArrayType(MangledName); 1780 else if (isFunctionType(MangledName)) { 1781 if (MangledName.consumeFront("$$A8@@")) 1782 Ty = demangleFunctionType(MangledName, true); 1783 else { 1784 assert(MangledName.startsWith("$$A6")); 1785 MangledName.consumeFront("$$A6"); 1786 Ty = demangleFunctionType(MangledName, false); 1787 } 1788 } else if (isCustomType(MangledName)) { 1789 Ty = demangleCustomType(MangledName); 1790 } else { 1791 Ty = demanglePrimitiveType(MangledName); 1792 } 1793 1794 if (!Ty || Error) 1795 return Ty; 1796 Ty->Quals = Qualifiers(Ty->Quals | Quals); 1797 return Ty; 1798 } 1799 1800 bool Demangler::demangleThrowSpecification(StringView &MangledName) { 1801 if (MangledName.consumeFront("_E")) 1802 return true; 1803 if (MangledName.consumeFront('Z')) 1804 return false; 1805 1806 Error = true; 1807 return false; 1808 } 1809 1810 FunctionSignatureNode *Demangler::demangleFunctionType(StringView &MangledName, 1811 bool HasThisQuals) { 1812 FunctionSignatureNode *FTy = Arena.alloc<FunctionSignatureNode>(); 1813 1814 if (HasThisQuals) { 1815 FTy->Quals = demanglePointerExtQualifiers(MangledName); 1816 FTy->RefQualifier = demangleFunctionRefQualifier(MangledName); 1817 FTy->Quals = Qualifiers(FTy->Quals | demangleQualifiers(MangledName).first); 1818 } 1819 1820 // Fields that appear on both member and non-member functions. 1821 FTy->CallConvention = demangleCallingConvention(MangledName); 1822 1823 // <return-type> ::= <type> 1824 // ::= @ # structors (they have no declared return type) 1825 bool IsStructor = MangledName.consumeFront('@'); 1826 if (!IsStructor) 1827 FTy->ReturnType = demangleType(MangledName, QualifierMangleMode::Result); 1828 1829 FTy->Params = demangleFunctionParameterList(MangledName, FTy->IsVariadic); 1830 1831 FTy->IsNoexcept = demangleThrowSpecification(MangledName); 1832 1833 return FTy; 1834 } 1835 1836 FunctionSymbolNode * 1837 Demangler::demangleFunctionEncoding(StringView &MangledName) { 1838 FuncClass ExtraFlags = FC_None; 1839 if (MangledName.consumeFront("$$J0")) 1840 ExtraFlags = FC_ExternC; 1841 1842 if (MangledName.empty()) { 1843 Error = true; 1844 return nullptr; 1845 } 1846 1847 FuncClass FC = demangleFunctionClass(MangledName); 1848 FC = FuncClass(ExtraFlags | FC); 1849 1850 FunctionSignatureNode *FSN = nullptr; 1851 ThunkSignatureNode *TTN = nullptr; 1852 if (FC & FC_StaticThisAdjust) { 1853 TTN = Arena.alloc<ThunkSignatureNode>(); 1854 TTN->ThisAdjust.StaticOffset = demangleSigned(MangledName); 1855 } else if (FC & FC_VirtualThisAdjust) { 1856 TTN = Arena.alloc<ThunkSignatureNode>(); 1857 if (FC & FC_VirtualThisAdjustEx) { 1858 TTN->ThisAdjust.VBPtrOffset = demangleSigned(MangledName); 1859 TTN->ThisAdjust.VBOffsetOffset = demangleSigned(MangledName); 1860 } 1861 TTN->ThisAdjust.VtordispOffset = demangleSigned(MangledName); 1862 TTN->ThisAdjust.StaticOffset = demangleSigned(MangledName); 1863 } 1864 1865 if (FC & FC_NoParameterList) { 1866 // This is an extern "C" function whose full signature hasn't been mangled. 1867 // This happens when we need to mangle a local symbol inside of an extern 1868 // "C" function. 1869 FSN = Arena.alloc<FunctionSignatureNode>(); 1870 } else { 1871 bool HasThisQuals = !(FC & (FC_Global | FC_Static)); 1872 FSN = demangleFunctionType(MangledName, HasThisQuals); 1873 } 1874 1875 if (Error) 1876 return nullptr; 1877 1878 if (TTN) { 1879 *static_cast<FunctionSignatureNode *>(TTN) = *FSN; 1880 FSN = TTN; 1881 } 1882 FSN->FunctionClass = FC; 1883 1884 FunctionSymbolNode *Symbol = Arena.alloc<FunctionSymbolNode>(); 1885 Symbol->Signature = FSN; 1886 return Symbol; 1887 } 1888 1889 CustomTypeNode *Demangler::demangleCustomType(StringView &MangledName) { 1890 assert(MangledName.startsWith('?')); 1891 MangledName.popFront(); 1892 1893 CustomTypeNode *CTN = Arena.alloc<CustomTypeNode>(); 1894 CTN->Identifier = demangleUnqualifiedTypeName(MangledName, /*Memorize=*/true); 1895 if (!MangledName.consumeFront('@')) 1896 Error = true; 1897 if (Error) 1898 return nullptr; 1899 return CTN; 1900 } 1901 1902 // Reads a primitive type. 1903 PrimitiveTypeNode *Demangler::demanglePrimitiveType(StringView &MangledName) { 1904 if (MangledName.consumeFront("$$T")) 1905 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Nullptr); 1906 1907 switch (MangledName.popFront()) { 1908 case 'X': 1909 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Void); 1910 case 'D': 1911 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Char); 1912 case 'C': 1913 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Schar); 1914 case 'E': 1915 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Uchar); 1916 case 'F': 1917 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Short); 1918 case 'G': 1919 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Ushort); 1920 case 'H': 1921 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Int); 1922 case 'I': 1923 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Uint); 1924 case 'J': 1925 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Long); 1926 case 'K': 1927 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Ulong); 1928 case 'M': 1929 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Float); 1930 case 'N': 1931 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Double); 1932 case 'O': 1933 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Ldouble); 1934 case '_': { 1935 if (MangledName.empty()) { 1936 Error = true; 1937 return nullptr; 1938 } 1939 switch (MangledName.popFront()) { 1940 case 'N': 1941 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Bool); 1942 case 'J': 1943 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Int64); 1944 case 'K': 1945 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Uint64); 1946 case 'W': 1947 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Wchar); 1948 case 'Q': 1949 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Char8); 1950 case 'S': 1951 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Char16); 1952 case 'U': 1953 return Arena.alloc<PrimitiveTypeNode>(PrimitiveKind::Char32); 1954 } 1955 break; 1956 } 1957 } 1958 Error = true; 1959 return nullptr; 1960 } 1961 1962 TagTypeNode *Demangler::demangleClassType(StringView &MangledName) { 1963 TagTypeNode *TT = nullptr; 1964 1965 switch (MangledName.popFront()) { 1966 case 'T': 1967 TT = Arena.alloc<TagTypeNode>(TagKind::Union); 1968 break; 1969 case 'U': 1970 TT = Arena.alloc<TagTypeNode>(TagKind::Struct); 1971 break; 1972 case 'V': 1973 TT = Arena.alloc<TagTypeNode>(TagKind::Class); 1974 break; 1975 case 'W': 1976 if (!MangledName.consumeFront('4')) { 1977 Error = true; 1978 return nullptr; 1979 } 1980 TT = Arena.alloc<TagTypeNode>(TagKind::Enum); 1981 break; 1982 default: 1983 assert(false); 1984 } 1985 1986 TT->QualifiedName = demangleFullyQualifiedTypeName(MangledName); 1987 return TT; 1988 } 1989 1990 // <pointer-type> ::= E? <pointer-cvr-qualifiers> <ext-qualifiers> <type> 1991 // # the E is required for 64-bit non-static pointers 1992 PointerTypeNode *Demangler::demanglePointerType(StringView &MangledName) { 1993 PointerTypeNode *Pointer = Arena.alloc<PointerTypeNode>(); 1994 1995 std::tie(Pointer->Quals, Pointer->Affinity) = 1996 demanglePointerCVQualifiers(MangledName); 1997 1998 if (MangledName.consumeFront("6")) { 1999 Pointer->Pointee = demangleFunctionType(MangledName, false); 2000 return Pointer; 2001 } 2002 2003 Qualifiers ExtQuals = demanglePointerExtQualifiers(MangledName); 2004 Pointer->Quals = Qualifiers(Pointer->Quals | ExtQuals); 2005 2006 Pointer->Pointee = demangleType(MangledName, QualifierMangleMode::Mangle); 2007 return Pointer; 2008 } 2009 2010 PointerTypeNode *Demangler::demangleMemberPointerType(StringView &MangledName) { 2011 PointerTypeNode *Pointer = Arena.alloc<PointerTypeNode>(); 2012 2013 std::tie(Pointer->Quals, Pointer->Affinity) = 2014 demanglePointerCVQualifiers(MangledName); 2015 assert(Pointer->Affinity == PointerAffinity::Pointer); 2016 2017 Qualifiers ExtQuals = demanglePointerExtQualifiers(MangledName); 2018 Pointer->Quals = Qualifiers(Pointer->Quals | ExtQuals); 2019 2020 // isMemberPointer() only returns true if there is at least one character 2021 // after the qualifiers. 2022 if (MangledName.consumeFront("8")) { 2023 Pointer->ClassParent = demangleFullyQualifiedTypeName(MangledName); 2024 Pointer->Pointee = demangleFunctionType(MangledName, true); 2025 } else { 2026 Qualifiers PointeeQuals = Q_None; 2027 bool IsMember = false; 2028 std::tie(PointeeQuals, IsMember) = demangleQualifiers(MangledName); 2029 assert(IsMember || Error); 2030 Pointer->ClassParent = demangleFullyQualifiedTypeName(MangledName); 2031 2032 Pointer->Pointee = demangleType(MangledName, QualifierMangleMode::Drop); 2033 if (Pointer->Pointee) 2034 Pointer->Pointee->Quals = PointeeQuals; 2035 } 2036 2037 return Pointer; 2038 } 2039 2040 Qualifiers Demangler::demanglePointerExtQualifiers(StringView &MangledName) { 2041 Qualifiers Quals = Q_None; 2042 if (MangledName.consumeFront('E')) 2043 Quals = Qualifiers(Quals | Q_Pointer64); 2044 if (MangledName.consumeFront('I')) 2045 Quals = Qualifiers(Quals | Q_Restrict); 2046 if (MangledName.consumeFront('F')) 2047 Quals = Qualifiers(Quals | Q_Unaligned); 2048 2049 return Quals; 2050 } 2051 2052 ArrayTypeNode *Demangler::demangleArrayType(StringView &MangledName) { 2053 assert(MangledName.front() == 'Y'); 2054 MangledName.popFront(); 2055 2056 uint64_t Rank = 0; 2057 bool IsNegative = false; 2058 std::tie(Rank, IsNegative) = demangleNumber(MangledName); 2059 if (IsNegative || Rank == 0) { 2060 Error = true; 2061 return nullptr; 2062 } 2063 2064 ArrayTypeNode *ATy = Arena.alloc<ArrayTypeNode>(); 2065 NodeList *Head = Arena.alloc<NodeList>(); 2066 NodeList *Tail = Head; 2067 2068 for (uint64_t I = 0; I < Rank; ++I) { 2069 uint64_t D = 0; 2070 std::tie(D, IsNegative) = demangleNumber(MangledName); 2071 if (Error || IsNegative) { 2072 Error = true; 2073 return nullptr; 2074 } 2075 Tail->N = Arena.alloc<IntegerLiteralNode>(D, IsNegative); 2076 if (I + 1 < Rank) { 2077 Tail->Next = Arena.alloc<NodeList>(); 2078 Tail = Tail->Next; 2079 } 2080 } 2081 ATy->Dimensions = nodeListToNodeArray(Arena, Head, Rank); 2082 2083 if (MangledName.consumeFront("$$C")) { 2084 bool IsMember = false; 2085 std::tie(ATy->Quals, IsMember) = demangleQualifiers(MangledName); 2086 if (IsMember) { 2087 Error = true; 2088 return nullptr; 2089 } 2090 } 2091 2092 ATy->ElementType = demangleType(MangledName, QualifierMangleMode::Drop); 2093 return ATy; 2094 } 2095 2096 // Reads a function's parameters. 2097 NodeArrayNode *Demangler::demangleFunctionParameterList(StringView &MangledName, 2098 bool &IsVariadic) { 2099 // Empty parameter list. 2100 if (MangledName.consumeFront('X')) 2101 return nullptr; 2102 2103 NodeList *Head = Arena.alloc<NodeList>(); 2104 NodeList **Current = &Head; 2105 size_t Count = 0; 2106 while (!Error && !MangledName.startsWith('@') && 2107 !MangledName.startsWith('Z')) { 2108 ++Count; 2109 2110 if (startsWithDigit(MangledName)) { 2111 size_t N = MangledName[0] - '0'; 2112 if (N >= Backrefs.FunctionParamCount) { 2113 Error = true; 2114 return nullptr; 2115 } 2116 MangledName = MangledName.dropFront(); 2117 2118 *Current = Arena.alloc<NodeList>(); 2119 (*Current)->N = Backrefs.FunctionParams[N]; 2120 Current = &(*Current)->Next; 2121 continue; 2122 } 2123 2124 size_t OldSize = MangledName.size(); 2125 2126 *Current = Arena.alloc<NodeList>(); 2127 TypeNode *TN = demangleType(MangledName, QualifierMangleMode::Drop); 2128 if (!TN || Error) 2129 return nullptr; 2130 2131 (*Current)->N = TN; 2132 2133 size_t CharsConsumed = OldSize - MangledName.size(); 2134 assert(CharsConsumed != 0); 2135 2136 // Single-letter types are ignored for backreferences because memorizing 2137 // them doesn't save anything. 2138 if (Backrefs.FunctionParamCount <= 9 && CharsConsumed > 1) 2139 Backrefs.FunctionParams[Backrefs.FunctionParamCount++] = TN; 2140 2141 Current = &(*Current)->Next; 2142 } 2143 2144 if (Error) 2145 return nullptr; 2146 2147 NodeArrayNode *NA = nodeListToNodeArray(Arena, Head, Count); 2148 // A non-empty parameter list is terminated by either 'Z' (variadic) parameter 2149 // list or '@' (non variadic). Careful not to consume "@Z", as in that case 2150 // the following Z could be a throw specifier. 2151 if (MangledName.consumeFront('@')) 2152 return NA; 2153 2154 if (MangledName.consumeFront('Z')) { 2155 IsVariadic = true; 2156 return NA; 2157 } 2158 2159 DEMANGLE_UNREACHABLE; 2160 } 2161 2162 NodeArrayNode * 2163 Demangler::demangleTemplateParameterList(StringView &MangledName) { 2164 NodeList *Head; 2165 NodeList **Current = &Head; 2166 size_t Count = 0; 2167 2168 while (!MangledName.startsWith('@')) { 2169 if (MangledName.consumeFront("$S") || MangledName.consumeFront("$$V") || 2170 MangledName.consumeFront("$$$V") || MangledName.consumeFront("$$Z")) { 2171 // parameter pack separator 2172 continue; 2173 } 2174 2175 ++Count; 2176 2177 // Template parameter lists don't participate in back-referencing. 2178 *Current = Arena.alloc<NodeList>(); 2179 2180 NodeList &TP = **Current; 2181 2182 TemplateParameterReferenceNode *TPRN = nullptr; 2183 if (MangledName.consumeFront("$$Y")) { 2184 // Template alias 2185 TP.N = demangleFullyQualifiedTypeName(MangledName); 2186 } else if (MangledName.consumeFront("$$B")) { 2187 // Array 2188 TP.N = demangleType(MangledName, QualifierMangleMode::Drop); 2189 } else if (MangledName.consumeFront("$$C")) { 2190 // Type has qualifiers. 2191 TP.N = demangleType(MangledName, QualifierMangleMode::Mangle); 2192 } else if (MangledName.startsWith("$1") || MangledName.startsWith("$H") || 2193 MangledName.startsWith("$I") || MangledName.startsWith("$J")) { 2194 // Pointer to member 2195 TP.N = TPRN = Arena.alloc<TemplateParameterReferenceNode>(); 2196 TPRN->IsMemberPointer = true; 2197 2198 MangledName = MangledName.dropFront(); 2199 // 1 - single inheritance <name> 2200 // H - multiple inheritance <name> <number> 2201 // I - virtual inheritance <name> <number> <number> 2202 // J - unspecified inheritance <name> <number> <number> <number> 2203 char InheritanceSpecifier = MangledName.popFront(); 2204 SymbolNode *S = nullptr; 2205 if (MangledName.startsWith('?')) { 2206 S = parse(MangledName); 2207 if (Error || !S->Name) { 2208 Error = true; 2209 return nullptr; 2210 } 2211 memorizeIdentifier(S->Name->getUnqualifiedIdentifier()); 2212 } 2213 2214 switch (InheritanceSpecifier) { 2215 case 'J': 2216 TPRN->ThunkOffsets[TPRN->ThunkOffsetCount++] = 2217 demangleSigned(MangledName); 2218 DEMANGLE_FALLTHROUGH; 2219 case 'I': 2220 TPRN->ThunkOffsets[TPRN->ThunkOffsetCount++] = 2221 demangleSigned(MangledName); 2222 DEMANGLE_FALLTHROUGH; 2223 case 'H': 2224 TPRN->ThunkOffsets[TPRN->ThunkOffsetCount++] = 2225 demangleSigned(MangledName); 2226 DEMANGLE_FALLTHROUGH; 2227 case '1': 2228 break; 2229 default: 2230 DEMANGLE_UNREACHABLE; 2231 } 2232 TPRN->Affinity = PointerAffinity::Pointer; 2233 TPRN->Symbol = S; 2234 } else if (MangledName.startsWith("$E?")) { 2235 MangledName.consumeFront("$E"); 2236 // Reference to symbol 2237 TP.N = TPRN = Arena.alloc<TemplateParameterReferenceNode>(); 2238 TPRN->Symbol = parse(MangledName); 2239 TPRN->Affinity = PointerAffinity::Reference; 2240 } else if (MangledName.startsWith("$F") || MangledName.startsWith("$G")) { 2241 TP.N = TPRN = Arena.alloc<TemplateParameterReferenceNode>(); 2242 2243 // Data member pointer. 2244 MangledName = MangledName.dropFront(); 2245 char InheritanceSpecifier = MangledName.popFront(); 2246 2247 switch (InheritanceSpecifier) { 2248 case 'G': 2249 TPRN->ThunkOffsets[TPRN->ThunkOffsetCount++] = 2250 demangleSigned(MangledName); 2251 DEMANGLE_FALLTHROUGH; 2252 case 'F': 2253 TPRN->ThunkOffsets[TPRN->ThunkOffsetCount++] = 2254 demangleSigned(MangledName); 2255 TPRN->ThunkOffsets[TPRN->ThunkOffsetCount++] = 2256 demangleSigned(MangledName); 2257 break; 2258 default: 2259 DEMANGLE_UNREACHABLE; 2260 } 2261 TPRN->IsMemberPointer = true; 2262 2263 } else if (MangledName.consumeFront("$0")) { 2264 // Integral non-type template parameter 2265 bool IsNegative = false; 2266 uint64_t Value = 0; 2267 std::tie(Value, IsNegative) = demangleNumber(MangledName); 2268 2269 TP.N = Arena.alloc<IntegerLiteralNode>(Value, IsNegative); 2270 } else { 2271 TP.N = demangleType(MangledName, QualifierMangleMode::Drop); 2272 } 2273 if (Error) 2274 return nullptr; 2275 2276 Current = &TP.Next; 2277 } 2278 2279 // The loop above returns nullptr on Error. 2280 assert(!Error); 2281 2282 // Template parameter lists cannot be variadic, so it can only be terminated 2283 // by @ (as opposed to 'Z' in the function parameter case). 2284 assert(MangledName.startsWith('@')); // The above loop exits only on '@'. 2285 MangledName.consumeFront('@'); 2286 return nodeListToNodeArray(Arena, Head, Count); 2287 } 2288 2289 void Demangler::dumpBackReferences() { 2290 std::printf("%d function parameter backreferences\n", 2291 (int)Backrefs.FunctionParamCount); 2292 2293 // Create an output stream so we can render each type. 2294 OutputStream OS; 2295 if (!initializeOutputStream(nullptr, nullptr, OS, 1024)) 2296 std::terminate(); 2297 for (size_t I = 0; I < Backrefs.FunctionParamCount; ++I) { 2298 OS.setCurrentPosition(0); 2299 2300 TypeNode *T = Backrefs.FunctionParams[I]; 2301 T->output(OS, OF_Default); 2302 2303 std::printf(" [%d] - %.*s\n", (int)I, (int)OS.getCurrentPosition(), 2304 OS.getBuffer()); 2305 } 2306 std::free(OS.getBuffer()); 2307 2308 if (Backrefs.FunctionParamCount > 0) 2309 std::printf("\n"); 2310 std::printf("%d name backreferences\n", (int)Backrefs.NamesCount); 2311 for (size_t I = 0; I < Backrefs.NamesCount; ++I) { 2312 std::printf(" [%d] - %.*s\n", (int)I, (int)Backrefs.Names[I]->Name.size(), 2313 Backrefs.Names[I]->Name.begin()); 2314 } 2315 if (Backrefs.NamesCount > 0) 2316 std::printf("\n"); 2317 } 2318 2319 char *llvm::microsoftDemangle(const char *MangledName, char *Buf, size_t *N, 2320 int *Status, MSDemangleFlags Flags) { 2321 int InternalStatus = demangle_success; 2322 Demangler D; 2323 OutputStream S; 2324 2325 StringView Name{MangledName}; 2326 SymbolNode *AST = D.parse(Name); 2327 2328 if (Flags & MSDF_DumpBackrefs) 2329 D.dumpBackReferences(); 2330 2331 if (D.Error) 2332 InternalStatus = demangle_invalid_mangled_name; 2333 else if (!initializeOutputStream(Buf, N, S, 1024)) 2334 InternalStatus = demangle_memory_alloc_failure; 2335 else { 2336 AST->output(S, OF_Default); 2337 S += '\0'; 2338 if (N != nullptr) 2339 *N = S.getCurrentPosition(); 2340 Buf = S.getBuffer(); 2341 } 2342 2343 if (Status) 2344 *Status = InternalStatus; 2345 return InternalStatus == demangle_success ? Buf : nullptr; 2346 } 2347