1//===- ExternalASTSource.h - Abstract External AST Interface ----*- 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// This file defines the ExternalASTSource interface, which enables
10// construction of AST nodes from some external source.
11//
12//===----------------------------------------------------------------------===//
13
14#ifndef LLVM_CLANG_AST_EXTERNALASTSOURCE_H
15#define LLVM_CLANG_AST_EXTERNALASTSOURCE_H
16
17#include "clang/AST/CharUnits.h"
18#include "clang/AST/DeclBase.h"
19#include "clang/Basic/LLVM.h"
20#include "llvm/ADT/ArrayRef.h"
21#include "llvm/ADT/DenseMap.h"
22#include "llvm/ADT/IntrusiveRefCntPtr.h"
23#include "llvm/ADT/Optional.h"
24#include "llvm/ADT/PointerUnion.h"
25#include "llvm/ADT/STLExtras.h"
26#include "llvm/ADT/SmallVector.h"
27#include "llvm/ADT/StringRef.h"
28#include "llvm/ADT/iterator.h"
29#include "llvm/Support/PointerLikeTypeTraits.h"
30#include <cassert>
31#include <cstddef>
32#include <cstdint>
33#include <iterator>
34#include <string>
35#include <utility>
36
37namespace clang {
38
39class ASTConsumer;
40class ASTContext;
41class ASTSourceDescriptor;
42class CXXBaseSpecifier;
43class CXXCtorInitializer;
44class CXXRecordDecl;
45class DeclarationName;
46class FieldDecl;
47class IdentifierInfo;
48class NamedDecl;
49class ObjCInterfaceDecl;
50class RecordDecl;
51class Selector;
52class Stmt;
53class TagDecl;
54
55/// Abstract interface for external sources of AST nodes.
56///
57/// External AST sources provide AST nodes constructed from some
58/// external source, such as a precompiled header. External AST
59/// sources can resolve types and declarations from abstract IDs into
60/// actual type and declaration nodes, and read parts of declaration
61/// contexts.
62class ExternalASTSource : public RefCountedBase<ExternalASTSource> {
63 friend class ExternalSemaSource;
64
65 /// Generation number for this external AST source. Must be increased
66 /// whenever we might have added new redeclarations for existing decls.
67 uint32_t CurrentGeneration = 0;
68
69 /// LLVM-style RTTI.
70 static char ID;
71
72public:
73 ExternalASTSource() = default;
74 virtual ~ExternalASTSource();
75
76 /// RAII class for safely pairing a StartedDeserializing call
77 /// with FinishedDeserializing.
78 class Deserializing {
79 ExternalASTSource *Source;
80
81 public:
82 explicit Deserializing(ExternalASTSource *source) : Source(source) {
83 assert(Source);
84 Source->StartedDeserializing();
85 }
86
87 ~Deserializing() {
88 Source->FinishedDeserializing();
89 }
90 };
91
92 /// Get the current generation of this AST source. This number
93 /// is incremented each time the AST source lazily extends an existing
94 /// entity.
95 uint32_t getGeneration() const { return CurrentGeneration; }
96
97 /// Resolve a declaration ID into a declaration, potentially
98 /// building a new declaration.
99 ///
100 /// This method only needs to be implemented if the AST source ever
101 /// passes back decl sets as VisibleDeclaration objects.
102 ///
103 /// The default implementation of this method is a no-op.
104 virtual Decl *GetExternalDecl(uint32_t ID);
105
106 /// Resolve a selector ID into a selector.
107 ///
108 /// This operation only needs to be implemented if the AST source
109 /// returns non-zero for GetNumKnownSelectors().
110 ///
111 /// The default implementation of this method is a no-op.
112 virtual Selector GetExternalSelector(uint32_t ID);
113
114 /// Returns the number of selectors known to the external AST
115 /// source.
116 ///
117 /// The default implementation of this method is a no-op.
118 virtual uint32_t GetNumExternalSelectors();
119
120 /// Resolve the offset of a statement in the decl stream into
121 /// a statement.
122 ///
123 /// This operation is meant to be used via a LazyOffsetPtr. It only
124 /// needs to be implemented if the AST source uses methods like
125 /// FunctionDecl::setLazyBody when building decls.
126 ///
127 /// The default implementation of this method is a no-op.
128 virtual Stmt *GetExternalDeclStmt(uint64_t Offset);
129
130 /// Resolve the offset of a set of C++ constructor initializers in
131 /// the decl stream into an array of initializers.
132 ///
133 /// The default implementation of this method is a no-op.
134 virtual CXXCtorInitializer **GetExternalCXXCtorInitializers(uint64_t Offset);
135
136 /// Resolve the offset of a set of C++ base specifiers in the decl
137 /// stream into an array of specifiers.
138 ///
139 /// The default implementation of this method is a no-op.
140 virtual CXXBaseSpecifier *GetExternalCXXBaseSpecifiers(uint64_t Offset);
141
142 /// Update an out-of-date identifier.
143 virtual void updateOutOfDateIdentifier(IdentifierInfo &II) {}
144
145 /// Find all declarations with the given name in the given context,
146 /// and add them to the context by calling SetExternalVisibleDeclsForName
147 /// or SetNoExternalVisibleDeclsForName.
148 /// \return \c true if any declarations might have been found, \c false if
149 /// we definitely have no declarations with tbis name.
150 ///
151 /// The default implementation of this method is a no-op returning \c false.
152 virtual bool
153 FindExternalVisibleDeclsByName(const DeclContext *DC, DeclarationName Name);
154
155 /// Ensures that the table of all visible declarations inside this
156 /// context is up to date.
157 ///
158 /// The default implementation of this function is a no-op.
159 virtual void completeVisibleDeclsMap(const DeclContext *DC);
160
161 /// Retrieve the module that corresponds to the given module ID.
162 virtual Module *getModule(unsigned ID) { return nullptr; }
163
164 /// Return a descriptor for the corresponding module, if one exists.
165 virtual llvm::Optional<ASTSourceDescriptor> getSourceDescriptor(unsigned ID);
166
167 enum ExtKind { EK_Always, EK_Never, EK_ReplyHazy };
168
169 virtual ExtKind hasExternalDefinitions(const Decl *D);
170
171 /// Finds all declarations lexically contained within the given
172 /// DeclContext, after applying an optional filter predicate.
173 ///
174 /// \param IsKindWeWant a predicate function that returns true if the passed
175 /// declaration kind is one we are looking for.
176 ///
177 /// The default implementation of this method is a no-op.
178 virtual void
179 FindExternalLexicalDecls(const DeclContext *DC,
180 llvm::function_ref<bool(Decl::Kind)> IsKindWeWant,
181 SmallVectorImpl<Decl *> &Result);
182
183 /// Finds all declarations lexically contained within the given
184 /// DeclContext.
185 void FindExternalLexicalDecls(const DeclContext *DC,
186 SmallVectorImpl<Decl *> &Result) {
187 FindExternalLexicalDecls(DC, [](Decl::Kind) { return true; }, Result);
188 }
189
190 /// Get the decls that are contained in a file in the Offset/Length
191 /// range. \p Length can be 0 to indicate a point at \p Offset instead of
192 /// a range.
193 virtual void FindFileRegionDecls(FileID File, unsigned Offset,
194 unsigned Length,
195 SmallVectorImpl<Decl *> &Decls);
196
197 /// Gives the external AST source an opportunity to complete
198 /// the redeclaration chain for a declaration. Called each time we
199 /// need the most recent declaration of a declaration after the
200 /// generation count is incremented.
201 virtual void CompleteRedeclChain(const Decl *D);
202
203 /// Gives the external AST source an opportunity to complete
204 /// an incomplete type.
205 virtual void CompleteType(TagDecl *Tag);
206
207 /// Gives the external AST source an opportunity to complete an
208 /// incomplete Objective-C class.
209 ///
210 /// This routine will only be invoked if the "externally completed" bit is
211 /// set on the ObjCInterfaceDecl via the function
212 /// \c ObjCInterfaceDecl::setExternallyCompleted().
213 virtual void CompleteType(ObjCInterfaceDecl *Class);
214
215 /// Loads comment ranges.
216 virtual void ReadComments();
217
218 /// Notify ExternalASTSource that we started deserialization of
219 /// a decl or type so until FinishedDeserializing is called there may be
220 /// decls that are initializing. Must be paired with FinishedDeserializing.
221 ///
222 /// The default implementation of this method is a no-op.
223 virtual void StartedDeserializing();
224
225 /// Notify ExternalASTSource that we finished the deserialization of
226 /// a decl or type. Must be paired with StartedDeserializing.
227 ///
228 /// The default implementation of this method is a no-op.
229 virtual void FinishedDeserializing();
230
231 /// Function that will be invoked when we begin parsing a new
232 /// translation unit involving this external AST source.
233 ///
234 /// The default implementation of this method is a no-op.
235 virtual void StartTranslationUnit(ASTConsumer *Consumer);
236
237 /// Print any statistics that have been gathered regarding
238 /// the external AST source.
239 ///
240 /// The default implementation of this method is a no-op.
241 virtual void PrintStats();
242
243 /// Perform layout on the given record.
244 ///
245 /// This routine allows the external AST source to provide an specific
246 /// layout for a record, overriding the layout that would normally be
247 /// constructed. It is intended for clients who receive specific layout
248 /// details rather than source code (such as LLDB). The client is expected
249 /// to fill in the field offsets, base offsets, virtual base offsets, and
250 /// complete object size.
251 ///
252 /// \param Record The record whose layout is being requested.
253 ///
254 /// \param Size The final size of the record, in bits.
255 ///
256 /// \param Alignment The final alignment of the record, in bits.
257 ///
258 /// \param FieldOffsets The offset of each of the fields within the record,
259 /// expressed in bits. All of the fields must be provided with offsets.
260 ///
261 /// \param BaseOffsets The offset of each of the direct, non-virtual base
262 /// classes. If any bases are not given offsets, the bases will be laid
263 /// out according to the ABI.
264 ///
265 /// \param VirtualBaseOffsets The offset of each of the virtual base classes
266 /// (either direct or not). If any bases are not given offsets, the bases will be laid
267 /// out according to the ABI.
268 ///
269 /// \returns true if the record layout was provided, false otherwise.
270 virtual bool layoutRecordType(
271 const RecordDecl *Record, uint64_t &Size, uint64_t &Alignment,
272 llvm::DenseMap<const FieldDecl *, uint64_t> &FieldOffsets,
273 llvm::DenseMap<const CXXRecordDecl *, CharUnits> &BaseOffsets,
274 llvm::DenseMap<const CXXRecordDecl *, CharUnits> &VirtualBaseOffsets);
275
276 //===--------------------------------------------------------------------===//
277 // Queries for performance analysis.
278 //===--------------------------------------------------------------------===//
279
280 struct MemoryBufferSizes {
281 size_t malloc_bytes;
282 size_t mmap_bytes;
283
284 MemoryBufferSizes(size_t malloc_bytes, size_t mmap_bytes)
285 : malloc_bytes(malloc_bytes), mmap_bytes(mmap_bytes) {}
286 };
287
288 /// Return the amount of memory used by memory buffers, breaking down
289 /// by heap-backed versus mmap'ed memory.
290 MemoryBufferSizes getMemoryBufferSizes() const {
291 MemoryBufferSizes sizes(0, 0);
292 getMemoryBufferSizes(sizes);
293 return sizes;
294 }
295
296 virtual void getMemoryBufferSizes(MemoryBufferSizes &sizes) const;
297
298 /// LLVM-style RTTI.
299 /// \{
300 virtual bool isA(const void *ClassID) const { return ClassID == &ID; }
301 static bool classof(const ExternalASTSource *S) { return S->isA(&ID); }
302 /// \}
303
304protected:
305 static DeclContextLookupResult
306 SetExternalVisibleDeclsForName(const DeclContext *DC,
307 DeclarationName Name,
308 ArrayRef<NamedDecl*> Decls);
309
310 static DeclContextLookupResult
311 SetNoExternalVisibleDeclsForName(const DeclContext *DC,
312 DeclarationName Name);
313
314 /// Increment the current generation.
315 uint32_t incrementGeneration(ASTContext &C);
316};
317
318/// A lazy pointer to an AST node (of base type T) that resides
319/// within an external AST source.
320///
321/// The AST node is identified within the external AST source by a
322/// 63-bit offset, and can be retrieved via an operation on the
323/// external AST source itself.
324template<typename T, typename OffsT, T* (ExternalASTSource::*Get)(OffsT Offset)>
325struct LazyOffsetPtr {
326 /// Either a pointer to an AST node or the offset within the
327 /// external AST source where the AST node can be found.
328 ///
329 /// If the low bit is clear, a pointer to the AST node. If the low
330 /// bit is set, the upper 63 bits are the offset.
331 mutable uint64_t Ptr = 0;
332
333public:
334 LazyOffsetPtr() = default;
335 explicit LazyOffsetPtr(T *Ptr) : Ptr(reinterpret_cast<uint64_t>(Ptr)) {}
336
337 explicit LazyOffsetPtr(uint64_t Offset) : Ptr((Offset << 1) | 0x01) {
338 assert((Offset << 1 >> 1) == Offset && "Offsets must require < 63 bits");
339 if (Offset == 0)
340 Ptr = 0;
341 }
342
343 LazyOffsetPtr &operator=(T *Ptr) {
344 this->Ptr = reinterpret_cast<uint64_t>(Ptr);
345 return *this;
346 }
347
348 LazyOffsetPtr &operator=(uint64_t Offset) {
349 assert((Offset << 1 >> 1) == Offset && "Offsets must require < 63 bits");
350 if (Offset == 0)
351 Ptr = 0;
352 else
353 Ptr = (Offset << 1) | 0x01;
354
355 return *this;
356 }
357
358 /// Whether this pointer is non-NULL.
359 ///
360 /// This operation does not require the AST node to be deserialized.
361 explicit operator bool() const { return Ptr != 0; }
362
363 /// Whether this pointer is non-NULL.
364 ///
365 /// This operation does not require the AST node to be deserialized.
366 bool isValid() const { return Ptr != 0; }
367
368 /// Whether this pointer is currently stored as an offset.
369 bool isOffset() const { return Ptr & 0x01; }
370
371 /// Retrieve the pointer to the AST node that this lazy pointer points to.
372 ///
373 /// \param Source the external AST source.
374 ///
375 /// \returns a pointer to the AST node.
376 T* get(ExternalASTSource *Source) const {
377 if (isOffset()) {
378 assert(Source &&
379 "Cannot deserialize a lazy pointer without an AST source");
380 Ptr = reinterpret_cast<uint64_t>((Source->*Get)(Ptr >> 1));
381 }
382 return reinterpret_cast<T*>(Ptr);
383 }
384};
385
386/// A lazy value (of type T) that is within an AST node of type Owner,
387/// where the value might change in later generations of the external AST
388/// source.
389template<typename Owner, typename T, void (ExternalASTSource::*Update)(Owner)>
390struct LazyGenerationalUpdatePtr {
391 /// A cache of the value of this pointer, in the most recent generation in
392 /// which we queried it.
393 struct LazyData {
394 ExternalASTSource *ExternalSource;
395 uint32_t LastGeneration = 0;
396 T LastValue;
397
398 LazyData(ExternalASTSource *Source, T Value)
399 : ExternalSource(Source), LastValue(Value) {}
400 };
401
402 // Our value is represented as simply T if there is no external AST source.
403 using ValueType = llvm::PointerUnion<T, LazyData*>;
404 ValueType Value;
405
406 LazyGenerationalUpdatePtr(ValueType V) : Value(V) {}
407
408 // Defined in ASTContext.h
409 static ValueType makeValue(const ASTContext &Ctx, T Value);
410
411public:
412 explicit LazyGenerationalUpdatePtr(const ASTContext &Ctx, T Value = T())
413 : Value(makeValue(Ctx, Value)) {}
414
415 /// Create a pointer that is not potentially updated by later generations of
416 /// the external AST source.
417 enum NotUpdatedTag { NotUpdated };
418 LazyGenerationalUpdatePtr(NotUpdatedTag, T Value = T())
419 : Value(Value) {}
420
421 /// Forcibly set this pointer (which must be lazy) as needing updates.
422 void markIncomplete() {
423 Value.template get<LazyData *>()->LastGeneration = 0;
424 }
425
426 /// Set the value of this pointer, in the current generation.
427 void set(T NewValue) {
428 if (auto *LazyVal = Value.template dyn_cast<LazyData *>()) {
429 LazyVal->LastValue = NewValue;
430 return;
431 }
432 Value = NewValue;
433 }
434
435 /// Set the value of this pointer, for this and all future generations.
436 void setNotUpdated(T NewValue) { Value = NewValue; }
437
438 /// Get the value of this pointer, updating its owner if necessary.
439 T get(Owner O) {
440 if (auto *LazyVal = Value.template dyn_cast<LazyData *>()) {
441 if (LazyVal->LastGeneration != LazyVal->ExternalSource->getGeneration()) {
442 LazyVal->LastGeneration = LazyVal->ExternalSource->getGeneration();
443 (LazyVal->ExternalSource->*Update)(O);
444 }
445 return LazyVal->LastValue;
446 }
447 return Value.template get<T>();
448 }
449
450 /// Get the most recently computed value of this pointer without updating it.
451 T getNotUpdated() const {
452 if (auto *LazyVal = Value.template dyn_cast<LazyData *>())
453 return LazyVal->LastValue;
454 return Value.template get<T>();
455 }
456
457 void *getOpaqueValue() { return Value.getOpaqueValue(); }
458 static LazyGenerationalUpdatePtr getFromOpaqueValue(void *Ptr) {
459 return LazyGenerationalUpdatePtr(ValueType::getFromOpaqueValue(Ptr));
460 }
461};
462
463} // namespace clang
464
465namespace llvm {
466
467/// Specialize PointerLikeTypeTraits to allow LazyGenerationalUpdatePtr to be
468/// placed into a PointerUnion.
469template<typename Owner, typename T,
470 void (clang::ExternalASTSource::*Update)(Owner)>
471struct PointerLikeTypeTraits<
472 clang::LazyGenerationalUpdatePtr<Owner, T, Update>> {
473 using Ptr = clang::LazyGenerationalUpdatePtr<Owner, T, Update>;
474
475 static void *getAsVoidPointer(Ptr P) { return P.getOpaqueValue(); }
476 static Ptr getFromVoidPointer(void *P) { return Ptr::getFromOpaqueValue(P); }
477
478 static constexpr int NumLowBitsAvailable =
479 PointerLikeTypeTraits<T>::NumLowBitsAvailable - 1;
480};
481
482} // namespace llvm
483
484namespace clang {
485
486/// Represents a lazily-loaded vector of data.
487///
488/// The lazily-loaded vector of data contains data that is partially loaded
489/// from an external source and partially added by local translation. The
490/// items loaded from the external source are loaded lazily, when needed for
491/// iteration over the complete vector.
492template<typename T, typename Source,
493 void (Source::*Loader)(SmallVectorImpl<T>&),
494 unsigned LoadedStorage = 2, unsigned LocalStorage = 4>
495class LazyVector {
496 SmallVector<T, LoadedStorage> Loaded;
497 SmallVector<T, LocalStorage> Local;
498
499public:
500 /// Iteration over the elements in the vector.
501 ///
502 /// In a complete iteration, the iterator walks the range [-M, N),
503 /// where negative values are used to indicate elements
504 /// loaded from the external source while non-negative values are used to
505 /// indicate elements added via \c push_back().
506 /// However, to provide iteration in source order (for, e.g., chained
507 /// precompiled headers), dereferencing the iterator flips the negative
508 /// values (corresponding to loaded entities), so that position -M
509 /// corresponds to element 0 in the loaded entities vector, position -M+1
510 /// corresponds to element 1 in the loaded entities vector, etc. This
511 /// gives us a reasonably efficient, source-order walk.
512 ///
513 /// We define this as a wrapping iterator around an int. The
514 /// iterator_adaptor_base class forwards the iterator methods to basic integer
515 /// arithmetic.
516 class iterator
517 : public llvm::iterator_adaptor_base<
518 iterator, int, std::random_access_iterator_tag, T, int, T *, T &> {
519 friend class LazyVector;
520
521 LazyVector *Self;
522
523 iterator(LazyVector *Self, int Position)
524 : iterator::iterator_adaptor_base(Position), Self(Self) {}
525
526 bool isLoaded() const { return this->I < 0; }
527
528 public:
529 iterator() : iterator(nullptr, 0) {}
530
531 typename iterator::reference operator*() const {
532 if (isLoaded())
533 return Self->Loaded.end()[this->I];
534 return Self->Local.begin()[this->I];
535 }
536 };
537
538 iterator begin(Source *source, bool LocalOnly = false) {
539 if (LocalOnly)
540 return iterator(this, 0);
541
542 if (source)
543 (source->*Loader)(Loaded);
544 return iterator(this, -(int)Loaded.size());
545 }
546
547 iterator end() {
548 return iterator(this, Local.size());
549 }
550
551 void push_back(const T& LocalValue) {
552 Local.push_back(LocalValue);
553 }
554
555 void erase(iterator From, iterator To) {
556 if (From.isLoaded() && To.isLoaded()) {
557 Loaded.erase(&*From, &*To);
558 return;
559 }
560
561 if (From.isLoaded()) {
562 Loaded.erase(&*From, Loaded.end());
563 From = begin(nullptr, true);
564 }
565
566 Local.erase(&*From, &*To);
567 }
568};
569
570/// A lazy pointer to a statement.
571using LazyDeclStmtPtr =
572 LazyOffsetPtr<Stmt, uint64_t, &ExternalASTSource::GetExternalDeclStmt>;
573
574/// A lazy pointer to a declaration.
575using LazyDeclPtr =
576 LazyOffsetPtr<Decl, uint32_t, &ExternalASTSource::GetExternalDecl>;
577
578/// A lazy pointer to a set of CXXCtorInitializers.
579using LazyCXXCtorInitializersPtr =
580 LazyOffsetPtr<CXXCtorInitializer *, uint64_t,
581 &ExternalASTSource::GetExternalCXXCtorInitializers>;
582
583/// A lazy pointer to a set of CXXBaseSpecifiers.
584using LazyCXXBaseSpecifiersPtr =
585 LazyOffsetPtr<CXXBaseSpecifier, uint64_t,
586 &ExternalASTSource::GetExternalCXXBaseSpecifiers>;
587
588} // namespace clang
589
590#endif // LLVM_CLANG_AST_EXTERNALASTSOURCE_H
591