hanchenye-llvm-project/clang/lib/AST/DeclCXX.cpp

1960 lines
78 KiB
C++
Raw Normal View History

Change struct forward declarations and definitions to use unique RecordDecls, as opposed to creating a single RecordDecl and reusing it. This change effects both RecordDecls and CXXRecordDecls, but does not effect EnumDecls (yet). The motivation of this patch is as follows: - Capture more source information, necessary for refactoring/rewriting clients. - Pave the way to resolve ownership issues with RecordDecls with the forthcoming addition of DeclGroups. Current caveats: - Until DeclGroups are in place, we will leak RecordDecls not explicitly referenced by the AST. For example: typedef struct { ... } x; The RecordDecl for the struct will be leaked because the TypedefDecl doesn't refer to it. This will be solved with DeclGroups. - This patch also (temporarily) breaks CodeGen. More below. High-level changes: - As before, TagType still refers to a TagDecl, but it doesn't own it. When a struct/union/class is first referenced, a RecordType and RecordDecl are created for it, and the RecordType refers to that RecordDecl. Later, if a new RecordDecl is created, the pointer to a RecordDecl in RecordType is updated to point to the RecordDecl that defines the struct/union/class. - TagDecl and RecordDecl now how a method 'getDefinition()' to return the TagDecl*/RecordDecl* that refers to the TagDecl* that defines a particular enum/struct/class/union. This is useful from going from a RecordDecl* that defines a forward declaration to the RecordDecl* that provides the actual definition. Note that this also works for EnumDecls, except that in this case there is no distinction between forward declarations and definitions (yet). - Clients should no longer assume that 'isDefinition()' returns true from a RecordDecl if the corresponding struct/union/class has been defined. isDefinition() only returns true if a particular RecordDecl is the defining Decl. Use 'getDefinition()' instead to determine if a struct has been defined. - The main changes to Sema happen in ActOnTag. To make the changes more incremental, I split off the processing of enums and structs et al into two code paths. Enums use the original code path (which is in ActOnTag) and structs use the ActOnTagStruct. Eventually the two code paths will be merged, but the idea was to preserve the original logic both for comparison and not to change the logic for both enums and structs all at once. - There is NO CHAINING of RecordDecls for the same RecordType. All RecordDecls that correspond to the same type simply have a pointer to that type. If we need to figure out what are all the RecordDecls for a given type we can build a backmap. - The diff in CXXRecordDecl.[cpp,h] is actually very small; it just mimics the changes to RecordDecl. For some reason 'svn' marks the entire file as changed. Why is CodeGen broken: - Codegen assumes that there is an equivalence between RecordDecl* and RecordType*. This was true before because we only created one RecordDecl* for a given RecordType*, but it is no longer true. I believe this shouldn't be too hard to change, but the patch was big enough as it is. I have tested this patch on both the clang test suite, and by running the static analyzer over Postgresql and a large Apple-internal project (mix of Objective-C and C). llvm-svn: 55839
2008-09-06 01:16:31 +08:00
//===--- DeclCXX.cpp - C++ Declaration AST Node Implementation ------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements the C++ related Decl classes.
//
//===----------------------------------------------------------------------===//
#include "clang/AST/DeclCXX.h"
#include "clang/AST/DeclTemplate.h"
Change struct forward declarations and definitions to use unique RecordDecls, as opposed to creating a single RecordDecl and reusing it. This change effects both RecordDecls and CXXRecordDecls, but does not effect EnumDecls (yet). The motivation of this patch is as follows: - Capture more source information, necessary for refactoring/rewriting clients. - Pave the way to resolve ownership issues with RecordDecls with the forthcoming addition of DeclGroups. Current caveats: - Until DeclGroups are in place, we will leak RecordDecls not explicitly referenced by the AST. For example: typedef struct { ... } x; The RecordDecl for the struct will be leaked because the TypedefDecl doesn't refer to it. This will be solved with DeclGroups. - This patch also (temporarily) breaks CodeGen. More below. High-level changes: - As before, TagType still refers to a TagDecl, but it doesn't own it. When a struct/union/class is first referenced, a RecordType and RecordDecl are created for it, and the RecordType refers to that RecordDecl. Later, if a new RecordDecl is created, the pointer to a RecordDecl in RecordType is updated to point to the RecordDecl that defines the struct/union/class. - TagDecl and RecordDecl now how a method 'getDefinition()' to return the TagDecl*/RecordDecl* that refers to the TagDecl* that defines a particular enum/struct/class/union. This is useful from going from a RecordDecl* that defines a forward declaration to the RecordDecl* that provides the actual definition. Note that this also works for EnumDecls, except that in this case there is no distinction between forward declarations and definitions (yet). - Clients should no longer assume that 'isDefinition()' returns true from a RecordDecl if the corresponding struct/union/class has been defined. isDefinition() only returns true if a particular RecordDecl is the defining Decl. Use 'getDefinition()' instead to determine if a struct has been defined. - The main changes to Sema happen in ActOnTag. To make the changes more incremental, I split off the processing of enums and structs et al into two code paths. Enums use the original code path (which is in ActOnTag) and structs use the ActOnTagStruct. Eventually the two code paths will be merged, but the idea was to preserve the original logic both for comparison and not to change the logic for both enums and structs all at once. - There is NO CHAINING of RecordDecls for the same RecordType. All RecordDecls that correspond to the same type simply have a pointer to that type. If we need to figure out what are all the RecordDecls for a given type we can build a backmap. - The diff in CXXRecordDecl.[cpp,h] is actually very small; it just mimics the changes to RecordDecl. For some reason 'svn' marks the entire file as changed. Why is CodeGen broken: - Codegen assumes that there is an equivalence between RecordDecl* and RecordType*. This was true before because we only created one RecordDecl* for a given RecordType*, but it is no longer true. I believe this shouldn't be too hard to change, but the patch was big enough as it is. I have tested this patch on both the clang test suite, and by running the static analyzer over Postgresql and a large Apple-internal project (mix of Objective-C and C). llvm-svn: 55839
2008-09-06 01:16:31 +08:00
#include "clang/AST/ASTContext.h"
#include "clang/AST/ASTMutationListener.h"
#include "clang/AST/CXXInheritance.h"
#include "clang/AST/Expr.h"
#include "clang/AST/ExprCXX.h"
#include "clang/AST/TypeLoc.h"
#include "clang/Basic/IdentifierTable.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallPtrSet.h"
Change struct forward declarations and definitions to use unique RecordDecls, as opposed to creating a single RecordDecl and reusing it. This change effects both RecordDecls and CXXRecordDecls, but does not effect EnumDecls (yet). The motivation of this patch is as follows: - Capture more source information, necessary for refactoring/rewriting clients. - Pave the way to resolve ownership issues with RecordDecls with the forthcoming addition of DeclGroups. Current caveats: - Until DeclGroups are in place, we will leak RecordDecls not explicitly referenced by the AST. For example: typedef struct { ... } x; The RecordDecl for the struct will be leaked because the TypedefDecl doesn't refer to it. This will be solved with DeclGroups. - This patch also (temporarily) breaks CodeGen. More below. High-level changes: - As before, TagType still refers to a TagDecl, but it doesn't own it. When a struct/union/class is first referenced, a RecordType and RecordDecl are created for it, and the RecordType refers to that RecordDecl. Later, if a new RecordDecl is created, the pointer to a RecordDecl in RecordType is updated to point to the RecordDecl that defines the struct/union/class. - TagDecl and RecordDecl now how a method 'getDefinition()' to return the TagDecl*/RecordDecl* that refers to the TagDecl* that defines a particular enum/struct/class/union. This is useful from going from a RecordDecl* that defines a forward declaration to the RecordDecl* that provides the actual definition. Note that this also works for EnumDecls, except that in this case there is no distinction between forward declarations and definitions (yet). - Clients should no longer assume that 'isDefinition()' returns true from a RecordDecl if the corresponding struct/union/class has been defined. isDefinition() only returns true if a particular RecordDecl is the defining Decl. Use 'getDefinition()' instead to determine if a struct has been defined. - The main changes to Sema happen in ActOnTag. To make the changes more incremental, I split off the processing of enums and structs et al into two code paths. Enums use the original code path (which is in ActOnTag) and structs use the ActOnTagStruct. Eventually the two code paths will be merged, but the idea was to preserve the original logic both for comparison and not to change the logic for both enums and structs all at once. - There is NO CHAINING of RecordDecls for the same RecordType. All RecordDecls that correspond to the same type simply have a pointer to that type. If we need to figure out what are all the RecordDecls for a given type we can build a backmap. - The diff in CXXRecordDecl.[cpp,h] is actually very small; it just mimics the changes to RecordDecl. For some reason 'svn' marks the entire file as changed. Why is CodeGen broken: - Codegen assumes that there is an equivalence between RecordDecl* and RecordType*. This was true before because we only created one RecordDecl* for a given RecordType*, but it is no longer true. I believe this shouldn't be too hard to change, but the patch was big enough as it is. I have tested this patch on both the clang test suite, and by running the static analyzer over Postgresql and a large Apple-internal project (mix of Objective-C and C). llvm-svn: 55839
2008-09-06 01:16:31 +08:00
using namespace clang;
//===----------------------------------------------------------------------===//
// Decl Allocation/Deallocation Method Implementations
//===----------------------------------------------------------------------===//
void AccessSpecDecl::anchor() { }
AccessSpecDecl *AccessSpecDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(AccessSpecDecl));
return new (Mem) AccessSpecDecl(EmptyShell());
}
CXXRecordDecl::DefinitionData::DefinitionData(CXXRecordDecl *D)
: UserDeclaredConstructor(false), UserDeclaredCopyConstructor(false),
UserDeclaredMoveConstructor(false), UserDeclaredCopyAssignment(false),
UserDeclaredMoveAssignment(false), UserDeclaredDestructor(false),
Aggregate(true), PlainOldData(true), Empty(true), Polymorphic(false),
Abstract(false), IsStandardLayout(true), HasNoNonEmptyBases(true),
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
HasPrivateFields(false), HasProtectedFields(false), HasPublicFields(false),
HasMutableFields(false), HasTrivialDefaultConstructor(true),
HasConstexprNonCopyMoveConstructor(false),
DefaultedDefaultConstructorIsConstexpr(true),
DefaultedCopyConstructorIsConstexpr(true),
DefaultedMoveConstructorIsConstexpr(true),
HasConstexprDefaultConstructor(false), HasConstexprCopyConstructor(false),
HasConstexprMoveConstructor(false), HasTrivialCopyConstructor(true),
HasTrivialMoveConstructor(true), HasTrivialCopyAssignment(true),
HasTrivialMoveAssignment(true), HasTrivialDestructor(true),
HasNonLiteralTypeFieldsOrBases(false), ComputedVisibleConversions(false),
UserProvidedDefaultConstructor(false), DeclaredDefaultConstructor(false),
DeclaredCopyConstructor(false), DeclaredMoveConstructor(false),
DeclaredCopyAssignment(false), DeclaredMoveAssignment(false),
DeclaredDestructor(false), FailedImplicitMoveConstructor(false),
FailedImplicitMoveAssignment(false), IsLambda(false), NumBases(0),
NumVBases(0), Bases(), VBases(), Definition(D), FirstFriend(0) {
}
CXXRecordDecl::CXXRecordDecl(Kind K, TagKind TK, DeclContext *DC,
2011-03-09 22:09:51 +08:00
SourceLocation StartLoc, SourceLocation IdLoc,
IdentifierInfo *Id, CXXRecordDecl *PrevDecl)
: RecordDecl(K, TK, DC, StartLoc, IdLoc, Id, PrevDecl),
DefinitionData(PrevDecl ? PrevDecl->DefinitionData : 0),
TemplateOrInstantiation() { }
CXXRecordDecl *CXXRecordDecl::Create(const ASTContext &C, TagKind TK,
2011-03-09 22:09:51 +08:00
DeclContext *DC, SourceLocation StartLoc,
SourceLocation IdLoc, IdentifierInfo *Id,
CXXRecordDecl* PrevDecl,
bool DelayTypeCreation) {
2011-03-09 22:09:51 +08:00
CXXRecordDecl* R = new (C) CXXRecordDecl(CXXRecord, TK, DC, StartLoc, IdLoc,
Id, PrevDecl);
// FIXME: DelayTypeCreation seems like such a hack
if (!DelayTypeCreation)
C.getTypeDeclType(R, PrevDecl);
Change struct forward declarations and definitions to use unique RecordDecls, as opposed to creating a single RecordDecl and reusing it. This change effects both RecordDecls and CXXRecordDecls, but does not effect EnumDecls (yet). The motivation of this patch is as follows: - Capture more source information, necessary for refactoring/rewriting clients. - Pave the way to resolve ownership issues with RecordDecls with the forthcoming addition of DeclGroups. Current caveats: - Until DeclGroups are in place, we will leak RecordDecls not explicitly referenced by the AST. For example: typedef struct { ... } x; The RecordDecl for the struct will be leaked because the TypedefDecl doesn't refer to it. This will be solved with DeclGroups. - This patch also (temporarily) breaks CodeGen. More below. High-level changes: - As before, TagType still refers to a TagDecl, but it doesn't own it. When a struct/union/class is first referenced, a RecordType and RecordDecl are created for it, and the RecordType refers to that RecordDecl. Later, if a new RecordDecl is created, the pointer to a RecordDecl in RecordType is updated to point to the RecordDecl that defines the struct/union/class. - TagDecl and RecordDecl now how a method 'getDefinition()' to return the TagDecl*/RecordDecl* that refers to the TagDecl* that defines a particular enum/struct/class/union. This is useful from going from a RecordDecl* that defines a forward declaration to the RecordDecl* that provides the actual definition. Note that this also works for EnumDecls, except that in this case there is no distinction between forward declarations and definitions (yet). - Clients should no longer assume that 'isDefinition()' returns true from a RecordDecl if the corresponding struct/union/class has been defined. isDefinition() only returns true if a particular RecordDecl is the defining Decl. Use 'getDefinition()' instead to determine if a struct has been defined. - The main changes to Sema happen in ActOnTag. To make the changes more incremental, I split off the processing of enums and structs et al into two code paths. Enums use the original code path (which is in ActOnTag) and structs use the ActOnTagStruct. Eventually the two code paths will be merged, but the idea was to preserve the original logic both for comparison and not to change the logic for both enums and structs all at once. - There is NO CHAINING of RecordDecls for the same RecordType. All RecordDecls that correspond to the same type simply have a pointer to that type. If we need to figure out what are all the RecordDecls for a given type we can build a backmap. - The diff in CXXRecordDecl.[cpp,h] is actually very small; it just mimics the changes to RecordDecl. For some reason 'svn' marks the entire file as changed. Why is CodeGen broken: - Codegen assumes that there is an equivalence between RecordDecl* and RecordType*. This was true before because we only created one RecordDecl* for a given RecordType*, but it is no longer true. I believe this shouldn't be too hard to change, but the patch was big enough as it is. I have tested this patch on both the clang test suite, and by running the static analyzer over Postgresql and a large Apple-internal project (mix of Objective-C and C). llvm-svn: 55839
2008-09-06 01:16:31 +08:00
return R;
}
CXXRecordDecl *
CXXRecordDecl::CreateDeserialized(const ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(CXXRecordDecl));
return new (Mem) CXXRecordDecl(CXXRecord, TTK_Struct, 0, SourceLocation(),
SourceLocation(), 0, 0);
}
void
CXXRecordDecl::setBases(CXXBaseSpecifier const * const *Bases,
unsigned NumBases) {
ASTContext &C = getASTContext();
if (!data().Bases.isOffset() && data().NumBases > 0)
C.Deallocate(data().getBases());
if (NumBases) {
// C++ [dcl.init.aggr]p1:
// An aggregate is [...] a class with [...] no base classes [...].
data().Aggregate = false;
// C++ [class]p4:
// A POD-struct is an aggregate class...
data().PlainOldData = false;
}
// The set of seen virtual base types.
llvm::SmallPtrSet<CanQualType, 8> SeenVBaseTypes;
// The virtual bases of this class.
SmallVector<const CXXBaseSpecifier *, 8> VBases;
data().Bases = new(C) CXXBaseSpecifier [NumBases];
data().NumBases = NumBases;
for (unsigned i = 0; i < NumBases; ++i) {
data().getBases()[i] = *Bases[i];
// Keep track of inherited vbases for this base class.
const CXXBaseSpecifier *Base = Bases[i];
QualType BaseType = Base->getType();
// Skip dependent types; we can't do any checking on them now.
if (BaseType->isDependentType())
continue;
CXXRecordDecl *BaseClassDecl
= cast<CXXRecordDecl>(BaseType->getAs<RecordType>()->getDecl());
// A class with a non-empty base class is not empty.
// FIXME: Standard ref?
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
if (!BaseClassDecl->isEmpty()) {
if (!data().Empty) {
// C++0x [class]p7:
// A standard-layout class is a class that:
// [...]
// -- either has no non-static data members in the most derived
// class and at most one base class with non-static data members,
// or has no base classes with non-static data members, and
// If this is the second non-empty base, then neither of these two
// clauses can be true.
data().IsStandardLayout = false;
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
}
data().Empty = false;
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
data().HasNoNonEmptyBases = false;
}
// C++ [class.virtual]p1:
// A class that declares or inherits a virtual function is called a
// polymorphic class.
if (BaseClassDecl->isPolymorphic())
data().Polymorphic = true;
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
// C++0x [class]p7:
// A standard-layout class is a class that: [...]
// -- has no non-standard-layout base classes
if (!BaseClassDecl->isStandardLayout())
data().IsStandardLayout = false;
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
// Record if this base is the first non-literal field or base.
if (!hasNonLiteralTypeFieldsOrBases() && !BaseType->isLiteralType())
data().HasNonLiteralTypeFieldsOrBases = true;
// Now go through all virtual bases of this base and add them.
for (CXXRecordDecl::base_class_iterator VBase =
BaseClassDecl->vbases_begin(),
E = BaseClassDecl->vbases_end(); VBase != E; ++VBase) {
// Add this base if it's not already in the list.
if (SeenVBaseTypes.insert(C.getCanonicalType(VBase->getType())))
VBases.push_back(VBase);
}
if (Base->isVirtual()) {
// Add this base if it's not already in the list.
if (SeenVBaseTypes.insert(C.getCanonicalType(BaseType)))
VBases.push_back(Base);
// C++0x [meta.unary.prop] is_empty:
// T is a class type, but not a union type, with ... no virtual base
// classes
data().Empty = false;
// C++ [class.ctor]p5:
// A default constructor is trivial [...] if:
// -- its class has [...] no virtual bases
data().HasTrivialDefaultConstructor = false;
// C++0x [class.copy]p13:
// A copy/move constructor for class X is trivial if it is neither
// user-provided nor deleted and if
// -- class X has no virtual functions and no virtual base classes, and
data().HasTrivialCopyConstructor = false;
data().HasTrivialMoveConstructor = false;
// C++0x [class.copy]p27:
// A copy/move assignment operator for class X is trivial if it is
// neither user-provided nor deleted and if
// -- class X has no virtual functions and no virtual base classes, and
data().HasTrivialCopyAssignment = false;
data().HasTrivialMoveAssignment = false;
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
// C++0x [class]p7:
// A standard-layout class is a class that: [...]
// -- has [...] no virtual base classes
data().IsStandardLayout = false;
// C++11 [dcl.constexpr]p4:
// In the definition of a constexpr constructor [...]
// -- the class shall not have any virtual base classes
data().DefaultedDefaultConstructorIsConstexpr = false;
data().DefaultedCopyConstructorIsConstexpr = false;
data().DefaultedMoveConstructorIsConstexpr = false;
} else {
// C++ [class.ctor]p5:
// A default constructor is trivial [...] if:
// -- all the direct base classes of its class have trivial default
// constructors.
if (!BaseClassDecl->hasTrivialDefaultConstructor())
data().HasTrivialDefaultConstructor = false;
// C++0x [class.copy]p13:
// A copy/move constructor for class X is trivial if [...]
// [...]
// -- the constructor selected to copy/move each direct base class
// subobject is trivial, and
// FIXME: C++0x: We need to only consider the selected constructor
// instead of all of them.
if (!BaseClassDecl->hasTrivialCopyConstructor())
data().HasTrivialCopyConstructor = false;
if (!BaseClassDecl->hasTrivialMoveConstructor())
data().HasTrivialMoveConstructor = false;
// C++0x [class.copy]p27:
// A copy/move assignment operator for class X is trivial if [...]
// [...]
// -- the assignment operator selected to copy/move each direct base
// class subobject is trivial, and
// FIXME: C++0x: We need to only consider the selected operator instead
// of all of them.
if (!BaseClassDecl->hasTrivialCopyAssignment())
data().HasTrivialCopyAssignment = false;
if (!BaseClassDecl->hasTrivialMoveAssignment())
data().HasTrivialMoveAssignment = false;
// C++11 [class.ctor]p6:
// If that user-written default cosntructor would satisfy the
// requirements of a constexpr constructor, the implicitly-defined
// default constructor is constexpr.
if (!BaseClassDecl->hasConstexprDefaultConstructor())
data().DefaultedDefaultConstructorIsConstexpr = false;
// C++11 [class.copy]p13:
// If the implicitly-defined constructor would satisfy the requirements
// of a constexpr constructor, the implicitly-defined constructor is
// constexpr.
// C++11 [dcl.constexpr]p4:
// -- every constructor involved in initializing [...] base class
// sub-objects shall be a constexpr constructor
if (!BaseClassDecl->hasConstexprCopyConstructor())
data().DefaultedCopyConstructorIsConstexpr = false;
if (BaseClassDecl->hasDeclaredMoveConstructor() ||
BaseClassDecl->needsImplicitMoveConstructor())
// FIXME: If the implicit move constructor generated for the base class
// would be ill-formed, the implicit move constructor generated for the
// derived class calls the base class' copy constructor.
data().DefaultedMoveConstructorIsConstexpr &=
!BaseClassDecl->hasConstexprMoveConstructor();
else if (!BaseClassDecl->hasConstexprCopyConstructor())
data().DefaultedMoveConstructorIsConstexpr = false;
}
// C++ [class.ctor]p3:
// A destructor is trivial if all the direct base classes of its class
// have trivial destructors.
if (!BaseClassDecl->hasTrivialDestructor())
data().HasTrivialDestructor = false;
// A class has an Objective-C object member if... or any of its bases
// has an Objective-C object member.
if (BaseClassDecl->hasObjectMember())
setHasObjectMember(true);
// Keep track of the presence of mutable fields.
if (BaseClassDecl->hasMutableFields())
data().HasMutableFields = true;
}
if (VBases.empty())
return;
// Create base specifier for any direct or indirect virtual bases.
data().VBases = new (C) CXXBaseSpecifier[VBases.size()];
data().NumVBases = VBases.size();
for (int I = 0, E = VBases.size(); I != E; ++I)
data().getVBases()[I] = *VBases[I];
}
/// Callback function for CXXRecordDecl::forallBases that acknowledges
/// that it saw a base class.
static bool SawBase(const CXXRecordDecl *, void *) {
return true;
}
bool CXXRecordDecl::hasAnyDependentBases() const {
if (!isDependentContext())
return false;
return !forallBases(SawBase, 0);
}
bool CXXRecordDecl::hasConstCopyConstructor() const {
return getCopyConstructor(Qualifiers::Const) != 0;
}
bool CXXRecordDecl::isTriviallyCopyable() const {
// C++0x [class]p5:
// A trivially copyable class is a class that:
// -- has no non-trivial copy constructors,
if (!hasTrivialCopyConstructor()) return false;
// -- has no non-trivial move constructors,
if (!hasTrivialMoveConstructor()) return false;
// -- has no non-trivial copy assignment operators,
if (!hasTrivialCopyAssignment()) return false;
// -- has no non-trivial move assignment operators, and
if (!hasTrivialMoveAssignment()) return false;
// -- has a trivial destructor.
if (!hasTrivialDestructor()) return false;
return true;
}
/// \brief Perform a simplistic form of overload resolution that only considers
/// cv-qualifiers on a single parameter, and return the best overload candidate
/// (if there is one).
static CXXMethodDecl *
GetBestOverloadCandidateSimple(
const SmallVectorImpl<std::pair<CXXMethodDecl *, Qualifiers> > &Cands) {
if (Cands.empty())
return 0;
if (Cands.size() == 1)
return Cands[0].first;
unsigned Best = 0, N = Cands.size();
for (unsigned I = 1; I != N; ++I)
if (Cands[Best].second.compatiblyIncludes(Cands[I].second))
Best = I;
for (unsigned I = 1; I != N; ++I)
if (Cands[Best].second.compatiblyIncludes(Cands[I].second))
return 0;
return Cands[Best].first;
}
CXXConstructorDecl *CXXRecordDecl::getCopyConstructor(unsigned TypeQuals) const{
ASTContext &Context = getASTContext();
QualType ClassType
= Context.getTypeDeclType(const_cast<CXXRecordDecl*>(this));
DeclarationName ConstructorName
= Context.DeclarationNames.getCXXConstructorName(
Context.getCanonicalType(ClassType));
unsigned FoundTQs;
SmallVector<std::pair<CXXMethodDecl *, Qualifiers>, 4> Found;
DeclContext::lookup_const_iterator Con, ConEnd;
for (llvm::tie(Con, ConEnd) = this->lookup(ConstructorName);
Con != ConEnd; ++Con) {
// C++ [class.copy]p2:
// A non-template constructor for class X is a copy constructor if [...]
if (isa<FunctionTemplateDecl>(*Con))
continue;
CXXConstructorDecl *Constructor = cast<CXXConstructorDecl>(*Con);
if (Constructor->isCopyConstructor(FoundTQs)) {
if (((TypeQuals & Qualifiers::Const) == (FoundTQs & Qualifiers::Const)) ||
(!(TypeQuals & Qualifiers::Const) && (FoundTQs & Qualifiers::Const)))
Found.push_back(std::make_pair(
const_cast<CXXConstructorDecl *>(Constructor),
Qualifiers::fromCVRMask(FoundTQs)));
}
}
return cast_or_null<CXXConstructorDecl>(
GetBestOverloadCandidateSimple(Found));
}
CXXConstructorDecl *CXXRecordDecl::getMoveConstructor() const {
for (ctor_iterator I = ctor_begin(), E = ctor_end(); I != E; ++I)
if (I->isMoveConstructor())
return *I;
return 0;
}
CXXMethodDecl *CXXRecordDecl::getCopyAssignmentOperator(bool ArgIsConst) const {
ASTContext &Context = getASTContext();
QualType Class = Context.getTypeDeclType(const_cast<CXXRecordDecl *>(this));
DeclarationName Name = Context.DeclarationNames.getCXXOperatorName(OO_Equal);
SmallVector<std::pair<CXXMethodDecl *, Qualifiers>, 4> Found;
DeclContext::lookup_const_iterator Op, OpEnd;
for (llvm::tie(Op, OpEnd) = this->lookup(Name); Op != OpEnd; ++Op) {
// C++ [class.copy]p9:
// A user-declared copy assignment operator is a non-static non-template
// member function of class X with exactly one parameter of type X, X&,
// const X&, volatile X& or const volatile X&.
const CXXMethodDecl* Method = dyn_cast<CXXMethodDecl>(*Op);
if (!Method || Method->isStatic() || Method->getPrimaryTemplate())
continue;
const FunctionProtoType *FnType
= Method->getType()->getAs<FunctionProtoType>();
assert(FnType && "Overloaded operator has no prototype.");
// Don't assert on this; an invalid decl might have been left in the AST.
if (FnType->getNumArgs() != 1 || FnType->isVariadic())
continue;
QualType ArgType = FnType->getArgType(0);
Qualifiers Quals;
if (const LValueReferenceType *Ref = ArgType->getAs<LValueReferenceType>()) {
ArgType = Ref->getPointeeType();
// If we have a const argument and we have a reference to a non-const,
// this function does not match.
if (ArgIsConst && !ArgType.isConstQualified())
continue;
Quals = ArgType.getQualifiers();
} else {
// By-value copy-assignment operators are treated like const X&
// copy-assignment operators.
Quals = Qualifiers::fromCVRMask(Qualifiers::Const);
}
if (!Context.hasSameUnqualifiedType(ArgType, Class))
continue;
// Save this copy-assignment operator. It might be "the one".
Found.push_back(std::make_pair(const_cast<CXXMethodDecl *>(Method), Quals));
}
// Use a simplistic form of overload resolution to find the candidate.
return GetBestOverloadCandidateSimple(Found);
}
CXXMethodDecl *CXXRecordDecl::getMoveAssignmentOperator() const {
for (method_iterator I = method_begin(), E = method_end(); I != E; ++I)
if (I->isMoveAssignmentOperator())
return *I;
return 0;
}
void CXXRecordDecl::markedVirtualFunctionPure() {
// C++ [class.abstract]p2:
// A class is abstract if it has at least one pure virtual function.
data().Abstract = true;
}
void CXXRecordDecl::addedMember(Decl *D) {
// Ignore friends and invalid declarations.
if (D->getFriendObjectKind() || D->isInvalidDecl())
return;
FunctionTemplateDecl *FunTmpl = dyn_cast<FunctionTemplateDecl>(D);
if (FunTmpl)
D = FunTmpl->getTemplatedDecl();
if (CXXMethodDecl *Method = dyn_cast<CXXMethodDecl>(D)) {
if (Method->isVirtual()) {
// C++ [dcl.init.aggr]p1:
// An aggregate is an array or a class with [...] no virtual functions.
data().Aggregate = false;
// C++ [class]p4:
// A POD-struct is an aggregate class...
data().PlainOldData = false;
// Virtual functions make the class non-empty.
// FIXME: Standard ref?
data().Empty = false;
// C++ [class.virtual]p1:
// A class that declares or inherits a virtual function is called a
// polymorphic class.
data().Polymorphic = true;
// C++0x [class.ctor]p5
// A default constructor is trivial [...] if:
// -- its class has no virtual functions [...]
data().HasTrivialDefaultConstructor = false;
// C++0x [class.copy]p13:
// A copy/move constructor for class X is trivial if [...]
// -- class X has no virtual functions [...]
data().HasTrivialCopyConstructor = false;
data().HasTrivialMoveConstructor = false;
// C++0x [class.copy]p27:
// A copy/move assignment operator for class X is trivial if [...]
// -- class X has no virtual functions [...]
data().HasTrivialCopyAssignment = false;
data().HasTrivialMoveAssignment = false;
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
// C++0x [class]p7:
// A standard-layout class is a class that: [...]
// -- has no virtual functions
data().IsStandardLayout = false;
}
}
if (D->isImplicit()) {
// Notify that an implicit member was added after the definition
// was completed.
if (!isBeingDefined())
if (ASTMutationListener *L = getASTMutationListener())
L->AddedCXXImplicitMember(data().Definition, D);
// If this is a special member function, note that it was added and then
// return early.
if (CXXConstructorDecl *Constructor = dyn_cast<CXXConstructorDecl>(D)) {
if (Constructor->isDefaultConstructor()) {
data().DeclaredDefaultConstructor = true;
if (Constructor->isConstexpr()) {
data().HasConstexprDefaultConstructor = true;
data().HasConstexprNonCopyMoveConstructor = true;
}
} else if (Constructor->isCopyConstructor()) {
data().DeclaredCopyConstructor = true;
if (Constructor->isConstexpr())
data().HasConstexprCopyConstructor = true;
} else if (Constructor->isMoveConstructor()) {
data().DeclaredMoveConstructor = true;
if (Constructor->isConstexpr())
data().HasConstexprMoveConstructor = true;
} else
goto NotASpecialMember;
return;
} else if (isa<CXXDestructorDecl>(D)) {
data().DeclaredDestructor = true;
return;
} else if (CXXMethodDecl *Method = dyn_cast<CXXMethodDecl>(D)) {
if (Method->isCopyAssignmentOperator())
data().DeclaredCopyAssignment = true;
else if (Method->isMoveAssignmentOperator())
data().DeclaredMoveAssignment = true;
else
goto NotASpecialMember;
return;
}
NotASpecialMember:;
// Any other implicit declarations are handled like normal declarations.
}
// Handle (user-declared) constructors.
if (CXXConstructorDecl *Constructor = dyn_cast<CXXConstructorDecl>(D)) {
// Note that we have a user-declared constructor.
data().UserDeclaredConstructor = true;
// Technically, "user-provided" is only defined for special member
// functions, but the intent of the standard is clearly that it should apply
// to all functions.
bool UserProvided = Constructor->isUserProvided();
if (Constructor->isDefaultConstructor()) {
data().DeclaredDefaultConstructor = true;
if (UserProvided) {
// C++0x [class.ctor]p5:
// A default constructor is trivial if it is not user-provided [...]
data().HasTrivialDefaultConstructor = false;
data().UserProvidedDefaultConstructor = true;
}
if (Constructor->isConstexpr()) {
data().HasConstexprDefaultConstructor = true;
data().HasConstexprNonCopyMoveConstructor = true;
}
}
// Note when we have a user-declared copy or move constructor, which will
// suppress the implicit declaration of those constructors.
if (!FunTmpl) {
if (Constructor->isCopyConstructor()) {
data().UserDeclaredCopyConstructor = true;
data().DeclaredCopyConstructor = true;
// C++0x [class.copy]p13:
// A copy/move constructor for class X is trivial if it is not
// user-provided [...]
if (UserProvided)
data().HasTrivialCopyConstructor = false;
if (Constructor->isConstexpr())
data().HasConstexprCopyConstructor = true;
} else if (Constructor->isMoveConstructor()) {
data().UserDeclaredMoveConstructor = true;
data().DeclaredMoveConstructor = true;
// C++0x [class.copy]p13:
// A copy/move constructor for class X is trivial if it is not
// user-provided [...]
if (UserProvided)
data().HasTrivialMoveConstructor = false;
if (Constructor->isConstexpr())
data().HasConstexprMoveConstructor = true;
}
}
if (Constructor->isConstexpr() && !Constructor->isCopyOrMoveConstructor()) {
// Record if we see any constexpr constructors which are neither copy
// nor move constructors.
data().HasConstexprNonCopyMoveConstructor = true;
}
// C++ [dcl.init.aggr]p1:
// An aggregate is an array or a class with no user-declared
// constructors [...].
// C++0x [dcl.init.aggr]p1:
// An aggregate is an array or a class with no user-provided
// constructors [...].
if (!getASTContext().getLangOptions().CPlusPlus0x || UserProvided)
data().Aggregate = false;
// C++ [class]p4:
// A POD-struct is an aggregate class [...]
// Since the POD bit is meant to be C++03 POD-ness, clear it even if the
// type is technically an aggregate in C++0x since it wouldn't be in 03.
data().PlainOldData = false;
return;
}
// Handle (user-declared) destructors.
if (CXXDestructorDecl *DD = dyn_cast<CXXDestructorDecl>(D)) {
data().DeclaredDestructor = true;
data().UserDeclaredDestructor = true;
// C++ [class]p4:
// A POD-struct is an aggregate class that has [...] no user-defined
// destructor.
// This bit is the C++03 POD bit, not the 0x one.
data().PlainOldData = false;
// C++11 [class.dtor]p5:
// A destructor is trivial if it is not user-provided and if
// -- the destructor is not virtual.
if (DD->isUserProvided() || DD->isVirtual()) {
data().HasTrivialDestructor = false;
// C++11 [dcl.constexpr]p1:
// The constexpr specifier shall be applied only to [...] the
// declaration of a static data member of a literal type.
// C++11 [basic.types]p10:
// A type is a literal type if it is [...] a class type that [...] has
// a trivial destructor.
data().DefaultedDefaultConstructorIsConstexpr = false;
data().DefaultedCopyConstructorIsConstexpr = false;
data().DefaultedMoveConstructorIsConstexpr = false;
}
return;
}
// Handle (user-declared) member functions.
if (CXXMethodDecl *Method = dyn_cast<CXXMethodDecl>(D)) {
if (Method->isCopyAssignmentOperator()) {
// C++ [class]p4:
// A POD-struct is an aggregate class that [...] has no user-defined
// copy assignment operator [...].
// This is the C++03 bit only.
data().PlainOldData = false;
// This is a copy assignment operator.
// Suppress the implicit declaration of a copy constructor.
data().UserDeclaredCopyAssignment = true;
data().DeclaredCopyAssignment = true;
// C++0x [class.copy]p27:
// A copy/move assignment operator for class X is trivial if it is
// neither user-provided nor deleted [...]
if (Method->isUserProvided())
data().HasTrivialCopyAssignment = false;
return;
}
if (Method->isMoveAssignmentOperator()) {
// This is an extension in C++03 mode, but we'll keep consistency by
// taking a move assignment operator to induce non-POD-ness
data().PlainOldData = false;
// This is a move assignment operator.
data().UserDeclaredMoveAssignment = true;
data().DeclaredMoveAssignment = true;
// C++0x [class.copy]p27:
// A copy/move assignment operator for class X is trivial if it is
// neither user-provided nor deleted [...]
if (Method->isUserProvided())
data().HasTrivialMoveAssignment = false;
}
// Keep the list of conversion functions up-to-date.
if (CXXConversionDecl *Conversion = dyn_cast<CXXConversionDecl>(D)) {
// We don't record specializations.
if (Conversion->getPrimaryTemplate())
return;
// FIXME: We intentionally don't use the decl's access here because it
// hasn't been set yet. That's really just a misdesign in Sema.
if (FunTmpl) {
if (FunTmpl->getPreviousDeclaration())
data().Conversions.replace(FunTmpl->getPreviousDeclaration(),
FunTmpl);
else
data().Conversions.addDecl(FunTmpl);
} else {
if (Conversion->getPreviousDeclaration())
data().Conversions.replace(Conversion->getPreviousDeclaration(),
Conversion);
else
data().Conversions.addDecl(Conversion);
}
}
return;
}
// Handle non-static data members.
if (FieldDecl *Field = dyn_cast<FieldDecl>(D)) {
// C++ [class.bit]p2:
// A declaration for a bit-field that omits the identifier declares an
// unnamed bit-field. Unnamed bit-fields are not members and cannot be
// initialized.
if (Field->isUnnamedBitfield())
return;
// C++ [dcl.init.aggr]p1:
// An aggregate is an array or a class (clause 9) with [...] no
// private or protected non-static data members (clause 11).
//
// A POD must be an aggregate.
if (D->getAccess() == AS_private || D->getAccess() == AS_protected) {
data().Aggregate = false;
data().PlainOldData = false;
}
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
// C++0x [class]p7:
// A standard-layout class is a class that:
// [...]
// -- has the same access control for all non-static data members,
switch (D->getAccess()) {
case AS_private: data().HasPrivateFields = true; break;
case AS_protected: data().HasProtectedFields = true; break;
case AS_public: data().HasPublicFields = true; break;
case AS_none: llvm_unreachable("Invalid access specifier");
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
};
if ((data().HasPrivateFields + data().HasProtectedFields +
data().HasPublicFields) > 1)
data().IsStandardLayout = false;
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
// Keep track of the presence of mutable fields.
if (Field->isMutable())
data().HasMutableFields = true;
// C++0x [class]p9:
// A POD struct is a class that is both a trivial class and a
// standard-layout class, and has no non-static data members of type
// non-POD struct, non-POD union (or array of such types).
//
// Automatic Reference Counting: the presence of a member of Objective-C pointer type
// that does not explicitly have no lifetime makes the class a non-POD.
// However, we delay setting PlainOldData to false in this case so that
// Sema has a chance to diagnostic causes where the same class will be
// non-POD with Automatic Reference Counting but a POD without Instant Objects.
// In this case, the class will become a non-POD class when we complete
// the definition.
ASTContext &Context = getASTContext();
QualType T = Context.getBaseElementType(Field->getType());
if (T->isObjCRetainableType() || T.isObjCGCStrong()) {
if (!Context.getLangOptions().ObjCAutoRefCount ||
T.getObjCLifetime() != Qualifiers::OCL_ExplicitNone)
setHasObjectMember(true);
} else if (!T.isPODType(Context))
data().PlainOldData = false;
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
if (T->isReferenceType()) {
data().HasTrivialDefaultConstructor = false;
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
// C++0x [class]p7:
// A standard-layout class is a class that:
// -- has no non-static data members of type [...] reference,
data().IsStandardLayout = false;
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
}
// Record if this field is the first non-literal field or base.
// As a slight variation on the standard, we regard mutable members as being
// non-literal, since mutating a constexpr variable would break C++11
// constant expression semantics.
if ((!hasNonLiteralTypeFieldsOrBases() && !T->isLiteralType()) ||
Field->isMutable())
data().HasNonLiteralTypeFieldsOrBases = true;
if (Field->hasInClassInitializer()) {
// C++0x [class]p5:
// A default constructor is trivial if [...] no non-static data member
// of its class has a brace-or-equal-initializer.
data().HasTrivialDefaultConstructor = false;
// C++0x [dcl.init.aggr]p1:
// An aggregate is a [...] class with [...] no
// brace-or-equal-initializers for non-static data members.
data().Aggregate = false;
// C++0x [class]p10:
// A POD struct is [...] a trivial class.
data().PlainOldData = false;
}
if (const RecordType *RecordTy = T->getAs<RecordType>()) {
CXXRecordDecl* FieldRec = cast<CXXRecordDecl>(RecordTy->getDecl());
if (FieldRec->getDefinition()) {
// C++0x [class.ctor]p5:
// A default constructor is trivial [...] if:
// -- for all the non-static data members of its class that are of
// class type (or array thereof), each such class has a trivial
// default constructor.
if (!FieldRec->hasTrivialDefaultConstructor())
data().HasTrivialDefaultConstructor = false;
// C++0x [class.copy]p13:
// A copy/move constructor for class X is trivial if [...]
// [...]
// -- for each non-static data member of X that is of class type (or
// an array thereof), the constructor selected to copy/move that
// member is trivial;
// FIXME: C++0x: We don't correctly model 'selected' constructors.
if (!FieldRec->hasTrivialCopyConstructor())
data().HasTrivialCopyConstructor = false;
if (!FieldRec->hasTrivialMoveConstructor())
data().HasTrivialMoveConstructor = false;
// C++0x [class.copy]p27:
// A copy/move assignment operator for class X is trivial if [...]
// [...]
// -- for each non-static data member of X that is of class type (or
// an array thereof), the assignment operator selected to
// copy/move that member is trivial;
// FIXME: C++0x: We don't correctly model 'selected' operators.
if (!FieldRec->hasTrivialCopyAssignment())
data().HasTrivialCopyAssignment = false;
if (!FieldRec->hasTrivialMoveAssignment())
data().HasTrivialMoveAssignment = false;
if (!FieldRec->hasTrivialDestructor())
data().HasTrivialDestructor = false;
if (FieldRec->hasObjectMember())
setHasObjectMember(true);
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
// C++0x [class]p7:
// A standard-layout class is a class that:
// -- has no non-static data members of type non-standard-layout
// class (or array of such types) [...]
if (!FieldRec->isStandardLayout())
data().IsStandardLayout = false;
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
// C++0x [class]p7:
// A standard-layout class is a class that:
// [...]
// -- has no base classes of the same type as the first non-static
// data member.
// We don't want to expend bits in the state of the record decl
// tracking whether this is the first non-static data member so we
// cheat a bit and use some of the existing state: the empty bit.
// Virtual bases and virtual methods make a class non-empty, but they
// also make it non-standard-layout so we needn't check here.
// A non-empty base class may leave the class standard-layout, but not
// if we have arrived here, and have at least on non-static data
// member. If IsStandardLayout remains true, then the first non-static
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
// data member must come through here with Empty still true, and Empty
// will subsequently be set to false below.
if (data().IsStandardLayout && data().Empty) {
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
for (CXXRecordDecl::base_class_const_iterator BI = bases_begin(),
BE = bases_end();
BI != BE; ++BI) {
if (Context.hasSameUnqualifiedType(BI->getType(), T)) {
data().IsStandardLayout = false;
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
break;
}
}
}
// Keep track of the presence of mutable fields.
if (FieldRec->hasMutableFields())
data().HasMutableFields = true;
// C++11 [class.copy]p13:
// If the implicitly-defined constructor would satisfy the
// requirements of a constexpr constructor, the implicitly-defined
// constructor is constexpr.
// C++11 [dcl.constexpr]p4:
// -- every constructor involved in initializing non-static data
// members [...] shall be a constexpr constructor
if (!Field->hasInClassInitializer() &&
!FieldRec->hasConstexprDefaultConstructor())
// The standard requires any in-class initializer to be a constant
// expression. We consider this to be a defect.
data().DefaultedDefaultConstructorIsConstexpr = false;
if (!FieldRec->hasConstexprCopyConstructor())
data().DefaultedCopyConstructorIsConstexpr = false;
if (FieldRec->hasDeclaredMoveConstructor() ||
FieldRec->needsImplicitMoveConstructor())
// FIXME: If the implicit move constructor generated for the member's
// class would be ill-formed, the implicit move constructor generated
// for this class calls the member's copy constructor.
data().DefaultedMoveConstructorIsConstexpr &=
FieldRec->hasConstexprMoveConstructor();
else if (!FieldRec->hasConstexprCopyConstructor())
data().DefaultedMoveConstructorIsConstexpr = false;
}
} else {
// Base element type of field is a non-class type.
if (!T->isLiteralType()) {
data().DefaultedDefaultConstructorIsConstexpr = false;
data().DefaultedCopyConstructorIsConstexpr = false;
data().DefaultedMoveConstructorIsConstexpr = false;
} else if (!Field->hasInClassInitializer())
data().DefaultedDefaultConstructorIsConstexpr = false;
}
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
// C++0x [class]p7:
// A standard-layout class is a class that:
// [...]
// -- either has no non-static data members in the most derived
// class and at most one base class with non-static data members,
// or has no base classes with non-static data members, and
// At this point we know that we have a non-static data member, so the last
// clause holds.
if (!data().HasNoNonEmptyBases)
data().IsStandardLayout = false;
Completely re-implement the core logic behind the __is_standard_layout type trait. The previous implementation suffered from several problems: 1) It implemented all of the logic in RecordType by walking over every base and field in a CXXRecordDecl and validating the constraints of the standard. This made for very straightforward code, but is extremely inefficient. It also is conceptually wrong, the logic tied to the C++ definition of standard-layout classes should be in CXXRecordDecl, not RecordType. 2) To address the performance problems with #1, a cache bit was added to CXXRecordDecl, and at the completion of every C++ class, the RecordType was queried to determine if it was a standard layout class, and that state was cached. Two things went very very wrong with this. First, the caching version of the query *was never called*. Even within the recursive steps of the walk over all fields and bases the caching variant was not called, making each query a full *recursive* walk. Second, despite the cache not being used, it was computed for every class declared, even when the trait was never used in the program. This probably significantly regressed compile time performance for edge-case files. 3) An ASTContext was required merely to query the type trait because querying it performed the actual computations. 4) The caching bit wasn't managed correctly (uninitialized). The new implementation follows the system for all the other traits on C++ classes by encoding all the state needed in the definition data and building up the trait incrementally as each base and member are added to the definition of the class. The idiosyncracies of the specification of standard-layout classes requires more state than I would like; currently 5 bits. I could eliminate one of the bits easily at the expense of both clarity and resilience of the code. I might be able to eliminate one of the other bits by computing its state in terms of other state bits in the definition. I've already done that in one place where there was a fairly simple way to achieve it. It's possible some of the bits could be moved out of the definition data and into some other structure which isn't serialized if the serialized bloat is a problem. That would preclude serialization of a partial class declaration, but that's likely already precluded. Comments on any of these issues welcome. llvm-svn: 130601
2011-04-30 17:17:45 +08:00
// If this is not a zero-length bit-field, then the class is not empty.
if (data().Empty) {
if (!Field->isBitField() ||
(!Field->getBitWidth()->isTypeDependent() &&
!Field->getBitWidth()->isValueDependent() &&
Field->getBitWidthValue(Context) != 0))
data().Empty = false;
}
}
// Handle using declarations of conversion functions.
if (UsingShadowDecl *Shadow = dyn_cast<UsingShadowDecl>(D))
if (Shadow->getDeclName().getNameKind()
== DeclarationName::CXXConversionFunctionName)
data().Conversions.addDecl(Shadow, Shadow->getAccess());
}
static CanQualType GetConversionType(ASTContext &Context, NamedDecl *Conv) {
QualType T;
if (isa<UsingShadowDecl>(Conv))
Conv = cast<UsingShadowDecl>(Conv)->getTargetDecl();
if (FunctionTemplateDecl *ConvTemp = dyn_cast<FunctionTemplateDecl>(Conv))
T = ConvTemp->getTemplatedDecl()->getResultType();
else
T = cast<CXXConversionDecl>(Conv)->getConversionType();
return Context.getCanonicalType(T);
}
/// Collect the visible conversions of a base class.
///
/// \param Base a base class of the class we're considering
/// \param InVirtual whether this base class is a virtual base (or a base
/// of a virtual base)
/// \param Access the access along the inheritance path to this base
/// \param ParentHiddenTypes the conversions provided by the inheritors
/// of this base
/// \param Output the set to which to add conversions from non-virtual bases
/// \param VOutput the set to which to add conversions from virtual bases
/// \param HiddenVBaseCs the set of conversions which were hidden in a
/// virtual base along some inheritance path
static void CollectVisibleConversions(ASTContext &Context,
CXXRecordDecl *Record,
bool InVirtual,
AccessSpecifier Access,
const llvm::SmallPtrSet<CanQualType, 8> &ParentHiddenTypes,
UnresolvedSetImpl &Output,
UnresolvedSetImpl &VOutput,
llvm::SmallPtrSet<NamedDecl*, 8> &HiddenVBaseCs) {
// The set of types which have conversions in this class or its
// subclasses. As an optimization, we don't copy the derived set
// unless it might change.
const llvm::SmallPtrSet<CanQualType, 8> *HiddenTypes = &ParentHiddenTypes;
llvm::SmallPtrSet<CanQualType, 8> HiddenTypesBuffer;
// Collect the direct conversions and figure out which conversions
// will be hidden in the subclasses.
UnresolvedSetImpl &Cs = *Record->getConversionFunctions();
if (!Cs.empty()) {
HiddenTypesBuffer = ParentHiddenTypes;
HiddenTypes = &HiddenTypesBuffer;
for (UnresolvedSetIterator I = Cs.begin(), E = Cs.end(); I != E; ++I) {
bool Hidden =
!HiddenTypesBuffer.insert(GetConversionType(Context, I.getDecl()));
// If this conversion is hidden and we're in a virtual base,
// remember that it's hidden along some inheritance path.
if (Hidden && InVirtual)
HiddenVBaseCs.insert(cast<NamedDecl>(I.getDecl()->getCanonicalDecl()));
// If this conversion isn't hidden, add it to the appropriate output.
else if (!Hidden) {
AccessSpecifier IAccess
= CXXRecordDecl::MergeAccess(Access, I.getAccess());
if (InVirtual)
VOutput.addDecl(I.getDecl(), IAccess);
else
Output.addDecl(I.getDecl(), IAccess);
}
}
}
// Collect information recursively from any base classes.
for (CXXRecordDecl::base_class_iterator
I = Record->bases_begin(), E = Record->bases_end(); I != E; ++I) {
const RecordType *RT = I->getType()->getAs<RecordType>();
if (!RT) continue;
AccessSpecifier BaseAccess
= CXXRecordDecl::MergeAccess(Access, I->getAccessSpecifier());
bool BaseInVirtual = InVirtual || I->isVirtual();
CXXRecordDecl *Base = cast<CXXRecordDecl>(RT->getDecl());
CollectVisibleConversions(Context, Base, BaseInVirtual, BaseAccess,
*HiddenTypes, Output, VOutput, HiddenVBaseCs);
}
}
/// Collect the visible conversions of a class.
///
/// This would be extremely straightforward if it weren't for virtual
/// bases. It might be worth special-casing that, really.
static void CollectVisibleConversions(ASTContext &Context,
CXXRecordDecl *Record,
UnresolvedSetImpl &Output) {
// The collection of all conversions in virtual bases that we've
// found. These will be added to the output as long as they don't
// appear in the hidden-conversions set.
UnresolvedSet<8> VBaseCs;
// The set of conversions in virtual bases that we've determined to
// be hidden.
llvm::SmallPtrSet<NamedDecl*, 8> HiddenVBaseCs;
// The set of types hidden by classes derived from this one.
llvm::SmallPtrSet<CanQualType, 8> HiddenTypes;
// Go ahead and collect the direct conversions and add them to the
// hidden-types set.
UnresolvedSetImpl &Cs = *Record->getConversionFunctions();
Output.append(Cs.begin(), Cs.end());
for (UnresolvedSetIterator I = Cs.begin(), E = Cs.end(); I != E; ++I)
HiddenTypes.insert(GetConversionType(Context, I.getDecl()));
// Recursively collect conversions from base classes.
for (CXXRecordDecl::base_class_iterator
I = Record->bases_begin(), E = Record->bases_end(); I != E; ++I) {
const RecordType *RT = I->getType()->getAs<RecordType>();
if (!RT) continue;
CollectVisibleConversions(Context, cast<CXXRecordDecl>(RT->getDecl()),
I->isVirtual(), I->getAccessSpecifier(),
HiddenTypes, Output, VBaseCs, HiddenVBaseCs);
}
// Add any unhidden conversions provided by virtual bases.
for (UnresolvedSetIterator I = VBaseCs.begin(), E = VBaseCs.end();
I != E; ++I) {
if (!HiddenVBaseCs.count(cast<NamedDecl>(I.getDecl()->getCanonicalDecl())))
Output.addDecl(I.getDecl(), I.getAccess());
}
}
/// getVisibleConversionFunctions - get all conversion functions visible
/// in current class; including conversion function templates.
const UnresolvedSetImpl *CXXRecordDecl::getVisibleConversionFunctions() {
// If root class, all conversions are visible.
if (bases_begin() == bases_end())
return &data().Conversions;
// If visible conversion list is already evaluated, return it.
if (data().ComputedVisibleConversions)
return &data().VisibleConversions;
CollectVisibleConversions(getASTContext(), this, data().VisibleConversions);
data().ComputedVisibleConversions = true;
return &data().VisibleConversions;
}
void CXXRecordDecl::removeConversion(const NamedDecl *ConvDecl) {
// This operation is O(N) but extremely rare. Sema only uses it to
// remove UsingShadowDecls in a class that were followed by a direct
// declaration, e.g.:
// class A : B {
// using B::operator int;
// operator int();
// };
// This is uncommon by itself and even more uncommon in conjunction
// with sufficiently large numbers of directly-declared conversions
// that asymptotic behavior matters.
UnresolvedSetImpl &Convs = *getConversionFunctions();
for (unsigned I = 0, E = Convs.size(); I != E; ++I) {
if (Convs[I].getDecl() == ConvDecl) {
Convs.erase(I);
assert(std::find(Convs.begin(), Convs.end(), ConvDecl) == Convs.end()
&& "conversion was found multiple times in unresolved set");
return;
}
}
llvm_unreachable("conversion not found in set!");
}
CXXRecordDecl *CXXRecordDecl::getInstantiatedFromMemberClass() const {
if (MemberSpecializationInfo *MSInfo = getMemberSpecializationInfo())
return cast<CXXRecordDecl>(MSInfo->getInstantiatedFrom());
return 0;
}
MemberSpecializationInfo *CXXRecordDecl::getMemberSpecializationInfo() const {
return TemplateOrInstantiation.dyn_cast<MemberSpecializationInfo *>();
}
void
CXXRecordDecl::setInstantiationOfMemberClass(CXXRecordDecl *RD,
TemplateSpecializationKind TSK) {
assert(TemplateOrInstantiation.isNull() &&
"Previous template or instantiation?");
assert(!isa<ClassTemplateSpecializationDecl>(this));
TemplateOrInstantiation
= new (getASTContext()) MemberSpecializationInfo(RD, TSK);
}
TemplateSpecializationKind CXXRecordDecl::getTemplateSpecializationKind() const{
if (const ClassTemplateSpecializationDecl *Spec
= dyn_cast<ClassTemplateSpecializationDecl>(this))
return Spec->getSpecializationKind();
if (MemberSpecializationInfo *MSInfo = getMemberSpecializationInfo())
return MSInfo->getTemplateSpecializationKind();
return TSK_Undeclared;
}
void
CXXRecordDecl::setTemplateSpecializationKind(TemplateSpecializationKind TSK) {
if (ClassTemplateSpecializationDecl *Spec
= dyn_cast<ClassTemplateSpecializationDecl>(this)) {
Spec->setSpecializationKind(TSK);
return;
}
if (MemberSpecializationInfo *MSInfo = getMemberSpecializationInfo()) {
MSInfo->setTemplateSpecializationKind(TSK);
return;
}
llvm_unreachable("Not a class template or member class specialization");
}
CXXDestructorDecl *CXXRecordDecl::getDestructor() const {
ASTContext &Context = getASTContext();
QualType ClassType = Context.getTypeDeclType(this);
DeclarationName Name
= Context.DeclarationNames.getCXXDestructorName(
Context.getCanonicalType(ClassType));
DeclContext::lookup_const_iterator I, E;
llvm::tie(I, E) = lookup(Name);
if (I == E)
return 0;
CXXDestructorDecl *Dtor = cast<CXXDestructorDecl>(*I);
return Dtor;
}
void CXXRecordDecl::completeDefinition() {
completeDefinition(0);
}
void CXXRecordDecl::completeDefinition(CXXFinalOverriderMap *FinalOverriders) {
RecordDecl::completeDefinition();
if (hasObjectMember() && getASTContext().getLangOptions().ObjCAutoRefCount) {
// Objective-C Automatic Reference Counting:
// If a class has a non-static data member of Objective-C pointer
// type (or array thereof), it is a non-POD type and its
// default constructor (if any), copy constructor, copy assignment
// operator, and destructor are non-trivial.
struct DefinitionData &Data = data();
Data.PlainOldData = false;
Data.HasTrivialDefaultConstructor = false;
Data.HasTrivialCopyConstructor = false;
Data.HasTrivialCopyAssignment = false;
Data.HasTrivialDestructor = false;
}
// If the class may be abstract (but hasn't been marked as such), check for
// any pure final overriders.
if (mayBeAbstract()) {
CXXFinalOverriderMap MyFinalOverriders;
if (!FinalOverriders) {
getFinalOverriders(MyFinalOverriders);
FinalOverriders = &MyFinalOverriders;
}
bool Done = false;
for (CXXFinalOverriderMap::iterator M = FinalOverriders->begin(),
MEnd = FinalOverriders->end();
M != MEnd && !Done; ++M) {
for (OverridingMethods::iterator SO = M->second.begin(),
SOEnd = M->second.end();
SO != SOEnd && !Done; ++SO) {
assert(SO->second.size() > 0 &&
"All virtual functions have overridding virtual functions");
// C++ [class.abstract]p4:
// A class is abstract if it contains or inherits at least one
// pure virtual function for which the final overrider is pure
// virtual.
if (SO->second.front().Method->isPure()) {
data().Abstract = true;
Done = true;
break;
}
}
}
}
// Set access bits correctly on the directly-declared conversions.
for (UnresolvedSetIterator I = data().Conversions.begin(),
E = data().Conversions.end();
I != E; ++I)
data().Conversions.setAccess(I, (*I)->getAccess());
}
bool CXXRecordDecl::mayBeAbstract() const {
if (data().Abstract || isInvalidDecl() || !data().Polymorphic ||
isDependentContext())
return false;
for (CXXRecordDecl::base_class_const_iterator B = bases_begin(),
BEnd = bases_end();
B != BEnd; ++B) {
CXXRecordDecl *BaseDecl
= cast<CXXRecordDecl>(B->getType()->getAs<RecordType>()->getDecl());
if (BaseDecl->isAbstract())
return true;
}
return false;
}
void CXXMethodDecl::anchor() { }
Change struct forward declarations and definitions to use unique RecordDecls, as opposed to creating a single RecordDecl and reusing it. This change effects both RecordDecls and CXXRecordDecls, but does not effect EnumDecls (yet). The motivation of this patch is as follows: - Capture more source information, necessary for refactoring/rewriting clients. - Pave the way to resolve ownership issues with RecordDecls with the forthcoming addition of DeclGroups. Current caveats: - Until DeclGroups are in place, we will leak RecordDecls not explicitly referenced by the AST. For example: typedef struct { ... } x; The RecordDecl for the struct will be leaked because the TypedefDecl doesn't refer to it. This will be solved with DeclGroups. - This patch also (temporarily) breaks CodeGen. More below. High-level changes: - As before, TagType still refers to a TagDecl, but it doesn't own it. When a struct/union/class is first referenced, a RecordType and RecordDecl are created for it, and the RecordType refers to that RecordDecl. Later, if a new RecordDecl is created, the pointer to a RecordDecl in RecordType is updated to point to the RecordDecl that defines the struct/union/class. - TagDecl and RecordDecl now how a method 'getDefinition()' to return the TagDecl*/RecordDecl* that refers to the TagDecl* that defines a particular enum/struct/class/union. This is useful from going from a RecordDecl* that defines a forward declaration to the RecordDecl* that provides the actual definition. Note that this also works for EnumDecls, except that in this case there is no distinction between forward declarations and definitions (yet). - Clients should no longer assume that 'isDefinition()' returns true from a RecordDecl if the corresponding struct/union/class has been defined. isDefinition() only returns true if a particular RecordDecl is the defining Decl. Use 'getDefinition()' instead to determine if a struct has been defined. - The main changes to Sema happen in ActOnTag. To make the changes more incremental, I split off the processing of enums and structs et al into two code paths. Enums use the original code path (which is in ActOnTag) and structs use the ActOnTagStruct. Eventually the two code paths will be merged, but the idea was to preserve the original logic both for comparison and not to change the logic for both enums and structs all at once. - There is NO CHAINING of RecordDecls for the same RecordType. All RecordDecls that correspond to the same type simply have a pointer to that type. If we need to figure out what are all the RecordDecls for a given type we can build a backmap. - The diff in CXXRecordDecl.[cpp,h] is actually very small; it just mimics the changes to RecordDecl. For some reason 'svn' marks the entire file as changed. Why is CodeGen broken: - Codegen assumes that there is an equivalence between RecordDecl* and RecordType*. This was true before because we only created one RecordDecl* for a given RecordType*, but it is no longer true. I believe this shouldn't be too hard to change, but the patch was big enough as it is. I have tested this patch on both the clang test suite, and by running the static analyzer over Postgresql and a large Apple-internal project (mix of Objective-C and C). llvm-svn: 55839
2008-09-06 01:16:31 +08:00
CXXMethodDecl *
CXXMethodDecl::Create(ASTContext &C, CXXRecordDecl *RD,
SourceLocation StartLoc,
const DeclarationNameInfo &NameInfo,
QualType T, TypeSourceInfo *TInfo,
bool isStatic, StorageClass SCAsWritten, bool isInline,
bool isConstexpr, SourceLocation EndLocation) {
return new (C) CXXMethodDecl(CXXMethod, RD, StartLoc, NameInfo, T, TInfo,
isStatic, SCAsWritten, isInline, isConstexpr,
EndLocation);
Change struct forward declarations and definitions to use unique RecordDecls, as opposed to creating a single RecordDecl and reusing it. This change effects both RecordDecls and CXXRecordDecls, but does not effect EnumDecls (yet). The motivation of this patch is as follows: - Capture more source information, necessary for refactoring/rewriting clients. - Pave the way to resolve ownership issues with RecordDecls with the forthcoming addition of DeclGroups. Current caveats: - Until DeclGroups are in place, we will leak RecordDecls not explicitly referenced by the AST. For example: typedef struct { ... } x; The RecordDecl for the struct will be leaked because the TypedefDecl doesn't refer to it. This will be solved with DeclGroups. - This patch also (temporarily) breaks CodeGen. More below. High-level changes: - As before, TagType still refers to a TagDecl, but it doesn't own it. When a struct/union/class is first referenced, a RecordType and RecordDecl are created for it, and the RecordType refers to that RecordDecl. Later, if a new RecordDecl is created, the pointer to a RecordDecl in RecordType is updated to point to the RecordDecl that defines the struct/union/class. - TagDecl and RecordDecl now how a method 'getDefinition()' to return the TagDecl*/RecordDecl* that refers to the TagDecl* that defines a particular enum/struct/class/union. This is useful from going from a RecordDecl* that defines a forward declaration to the RecordDecl* that provides the actual definition. Note that this also works for EnumDecls, except that in this case there is no distinction between forward declarations and definitions (yet). - Clients should no longer assume that 'isDefinition()' returns true from a RecordDecl if the corresponding struct/union/class has been defined. isDefinition() only returns true if a particular RecordDecl is the defining Decl. Use 'getDefinition()' instead to determine if a struct has been defined. - The main changes to Sema happen in ActOnTag. To make the changes more incremental, I split off the processing of enums and structs et al into two code paths. Enums use the original code path (which is in ActOnTag) and structs use the ActOnTagStruct. Eventually the two code paths will be merged, but the idea was to preserve the original logic both for comparison and not to change the logic for both enums and structs all at once. - There is NO CHAINING of RecordDecls for the same RecordType. All RecordDecls that correspond to the same type simply have a pointer to that type. If we need to figure out what are all the RecordDecls for a given type we can build a backmap. - The diff in CXXRecordDecl.[cpp,h] is actually very small; it just mimics the changes to RecordDecl. For some reason 'svn' marks the entire file as changed. Why is CodeGen broken: - Codegen assumes that there is an equivalence between RecordDecl* and RecordType*. This was true before because we only created one RecordDecl* for a given RecordType*, but it is no longer true. I believe this shouldn't be too hard to change, but the patch was big enough as it is. I have tested this patch on both the clang test suite, and by running the static analyzer over Postgresql and a large Apple-internal project (mix of Objective-C and C). llvm-svn: 55839
2008-09-06 01:16:31 +08:00
}
CXXMethodDecl *CXXMethodDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(CXXMethodDecl));
return new (Mem) CXXMethodDecl(CXXMethod, 0, SourceLocation(),
DeclarationNameInfo(), QualType(),
0, false, SC_None, false, false,
SourceLocation());
}
bool CXXMethodDecl::isUsualDeallocationFunction() const {
if (getOverloadedOperator() != OO_Delete &&
getOverloadedOperator() != OO_Array_Delete)
return false;
// C++ [basic.stc.dynamic.deallocation]p2:
// A template instance is never a usual deallocation function,
// regardless of its signature.
if (getPrimaryTemplate())
return false;
// C++ [basic.stc.dynamic.deallocation]p2:
// If a class T has a member deallocation function named operator delete
// with exactly one parameter, then that function is a usual (non-placement)
// deallocation function. [...]
if (getNumParams() == 1)
return true;
// C++ [basic.stc.dynamic.deallocation]p2:
// [...] If class T does not declare such an operator delete but does
// declare a member deallocation function named operator delete with
// exactly two parameters, the second of which has type std::size_t (18.1),
// then this function is a usual deallocation function.
ASTContext &Context = getASTContext();
if (getNumParams() != 2 ||
!Context.hasSameUnqualifiedType(getParamDecl(1)->getType(),
Context.getSizeType()))
return false;
// This function is a usual deallocation function if there are no
// single-parameter deallocation functions of the same kind.
for (DeclContext::lookup_const_result R = getDeclContext()->lookup(getDeclName());
R.first != R.second; ++R.first) {
if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(*R.first))
if (FD->getNumParams() == 1)
return false;
}
return true;
}
Complete reimplementation of the synthesis for implicitly-defined copy assignment operators. Previously, Sema provided type-checking and template instantiation for copy assignment operators, then CodeGen would synthesize the actual body of the copy constructor. Unfortunately, the two were not in sync, and CodeGen might pick a copy-assignment operator that is different from what Sema chose, leading to strange failures, e.g., link-time failures when CodeGen called a copy-assignment operator that was not instantiation, run-time failures when copy-assignment operators were overloaded for const/non-const references and the wrong one was picked, and run-time failures when by-value copy-assignment operators did not have their arguments properly copy-initialized. This implementation synthesizes the implicitly-defined copy assignment operator bodies in Sema, so that the resulting ASTs encode exactly what CodeGen needs to do; there is no longer any special code in CodeGen to synthesize copy-assignment operators. The synthesis of the body is relatively simple, and we generate one of three different kinds of copy statements for each base or member: - For a class subobject, call the appropriate copy-assignment operator, after overload resolution has determined what that is. - For an array of scalar types or an array of class types that have trivial copy assignment operators, construct a call to __builtin_memcpy. - For an array of class types with non-trivial copy assignment operators, synthesize a (possibly nested!) for loop whose inner statement calls the copy constructor. - For a scalar type, use built-in assignment. This patch fixes at least a few tests cases in Boost.Spirit that were failing because CodeGen picked the wrong copy-assignment operator (leading to link-time failures), and I suspect a number of undiagnosed problems will also go away with this change. Some of the diagnostics we had previously have gotten worse with this change, since we're going through generic code for our type-checking. I will improve this in a subsequent patch. llvm-svn: 102853
2010-05-02 04:49:11 +08:00
bool CXXMethodDecl::isCopyAssignmentOperator() const {
// C++0x [class.copy]p17:
Complete reimplementation of the synthesis for implicitly-defined copy assignment operators. Previously, Sema provided type-checking and template instantiation for copy assignment operators, then CodeGen would synthesize the actual body of the copy constructor. Unfortunately, the two were not in sync, and CodeGen might pick a copy-assignment operator that is different from what Sema chose, leading to strange failures, e.g., link-time failures when CodeGen called a copy-assignment operator that was not instantiation, run-time failures when copy-assignment operators were overloaded for const/non-const references and the wrong one was picked, and run-time failures when by-value copy-assignment operators did not have their arguments properly copy-initialized. This implementation synthesizes the implicitly-defined copy assignment operator bodies in Sema, so that the resulting ASTs encode exactly what CodeGen needs to do; there is no longer any special code in CodeGen to synthesize copy-assignment operators. The synthesis of the body is relatively simple, and we generate one of three different kinds of copy statements for each base or member: - For a class subobject, call the appropriate copy-assignment operator, after overload resolution has determined what that is. - For an array of scalar types or an array of class types that have trivial copy assignment operators, construct a call to __builtin_memcpy. - For an array of class types with non-trivial copy assignment operators, synthesize a (possibly nested!) for loop whose inner statement calls the copy constructor. - For a scalar type, use built-in assignment. This patch fixes at least a few tests cases in Boost.Spirit that were failing because CodeGen picked the wrong copy-assignment operator (leading to link-time failures), and I suspect a number of undiagnosed problems will also go away with this change. Some of the diagnostics we had previously have gotten worse with this change, since we're going through generic code for our type-checking. I will improve this in a subsequent patch. llvm-svn: 102853
2010-05-02 04:49:11 +08:00
// A user-declared copy assignment operator X::operator= is a non-static
// non-template member function of class X with exactly one parameter of
// type X, X&, const X&, volatile X& or const volatile X&.
if (/*operator=*/getOverloadedOperator() != OO_Equal ||
/*non-static*/ isStatic() ||
/*non-template*/getPrimaryTemplate() || getDescribedFunctionTemplate())
Complete reimplementation of the synthesis for implicitly-defined copy assignment operators. Previously, Sema provided type-checking and template instantiation for copy assignment operators, then CodeGen would synthesize the actual body of the copy constructor. Unfortunately, the two were not in sync, and CodeGen might pick a copy-assignment operator that is different from what Sema chose, leading to strange failures, e.g., link-time failures when CodeGen called a copy-assignment operator that was not instantiation, run-time failures when copy-assignment operators were overloaded for const/non-const references and the wrong one was picked, and run-time failures when by-value copy-assignment operators did not have their arguments properly copy-initialized. This implementation synthesizes the implicitly-defined copy assignment operator bodies in Sema, so that the resulting ASTs encode exactly what CodeGen needs to do; there is no longer any special code in CodeGen to synthesize copy-assignment operators. The synthesis of the body is relatively simple, and we generate one of three different kinds of copy statements for each base or member: - For a class subobject, call the appropriate copy-assignment operator, after overload resolution has determined what that is. - For an array of scalar types or an array of class types that have trivial copy assignment operators, construct a call to __builtin_memcpy. - For an array of class types with non-trivial copy assignment operators, synthesize a (possibly nested!) for loop whose inner statement calls the copy constructor. - For a scalar type, use built-in assignment. This patch fixes at least a few tests cases in Boost.Spirit that were failing because CodeGen picked the wrong copy-assignment operator (leading to link-time failures), and I suspect a number of undiagnosed problems will also go away with this change. Some of the diagnostics we had previously have gotten worse with this change, since we're going through generic code for our type-checking. I will improve this in a subsequent patch. llvm-svn: 102853
2010-05-02 04:49:11 +08:00
return false;
QualType ParamType = getParamDecl(0)->getType();
if (const LValueReferenceType *Ref = ParamType->getAs<LValueReferenceType>())
ParamType = Ref->getPointeeType();
ASTContext &Context = getASTContext();
QualType ClassType
= Context.getCanonicalType(Context.getTypeDeclType(getParent()));
return Context.hasSameUnqualifiedType(ClassType, ParamType);
}
bool CXXMethodDecl::isMoveAssignmentOperator() const {
// C++0x [class.copy]p19:
// A user-declared move assignment operator X::operator= is a non-static
// non-template member function of class X with exactly one parameter of type
// X&&, const X&&, volatile X&&, or const volatile X&&.
if (getOverloadedOperator() != OO_Equal || isStatic() ||
getPrimaryTemplate() || getDescribedFunctionTemplate())
return false;
QualType ParamType = getParamDecl(0)->getType();
if (!isa<RValueReferenceType>(ParamType))
return false;
ParamType = ParamType->getPointeeType();
ASTContext &Context = getASTContext();
QualType ClassType
= Context.getCanonicalType(Context.getTypeDeclType(getParent()));
return Context.hasSameUnqualifiedType(ClassType, ParamType);
}
void CXXMethodDecl::addOverriddenMethod(const CXXMethodDecl *MD) {
assert(MD->isCanonicalDecl() && "Method is not canonical!");
assert(!MD->getParent()->isDependentContext() &&
"Can't add an overridden method to a class template!");
getASTContext().addOverriddenMethod(this, MD);
}
CXXMethodDecl::method_iterator CXXMethodDecl::begin_overridden_methods() const {
return getASTContext().overridden_methods_begin(this);
}
CXXMethodDecl::method_iterator CXXMethodDecl::end_overridden_methods() const {
return getASTContext().overridden_methods_end(this);
}
unsigned CXXMethodDecl::size_overridden_methods() const {
return getASTContext().overridden_methods_size(this);
}
Change struct forward declarations and definitions to use unique RecordDecls, as opposed to creating a single RecordDecl and reusing it. This change effects both RecordDecls and CXXRecordDecls, but does not effect EnumDecls (yet). The motivation of this patch is as follows: - Capture more source information, necessary for refactoring/rewriting clients. - Pave the way to resolve ownership issues with RecordDecls with the forthcoming addition of DeclGroups. Current caveats: - Until DeclGroups are in place, we will leak RecordDecls not explicitly referenced by the AST. For example: typedef struct { ... } x; The RecordDecl for the struct will be leaked because the TypedefDecl doesn't refer to it. This will be solved with DeclGroups. - This patch also (temporarily) breaks CodeGen. More below. High-level changes: - As before, TagType still refers to a TagDecl, but it doesn't own it. When a struct/union/class is first referenced, a RecordType and RecordDecl are created for it, and the RecordType refers to that RecordDecl. Later, if a new RecordDecl is created, the pointer to a RecordDecl in RecordType is updated to point to the RecordDecl that defines the struct/union/class. - TagDecl and RecordDecl now how a method 'getDefinition()' to return the TagDecl*/RecordDecl* that refers to the TagDecl* that defines a particular enum/struct/class/union. This is useful from going from a RecordDecl* that defines a forward declaration to the RecordDecl* that provides the actual definition. Note that this also works for EnumDecls, except that in this case there is no distinction between forward declarations and definitions (yet). - Clients should no longer assume that 'isDefinition()' returns true from a RecordDecl if the corresponding struct/union/class has been defined. isDefinition() only returns true if a particular RecordDecl is the defining Decl. Use 'getDefinition()' instead to determine if a struct has been defined. - The main changes to Sema happen in ActOnTag. To make the changes more incremental, I split off the processing of enums and structs et al into two code paths. Enums use the original code path (which is in ActOnTag) and structs use the ActOnTagStruct. Eventually the two code paths will be merged, but the idea was to preserve the original logic both for comparison and not to change the logic for both enums and structs all at once. - There is NO CHAINING of RecordDecls for the same RecordType. All RecordDecls that correspond to the same type simply have a pointer to that type. If we need to figure out what are all the RecordDecls for a given type we can build a backmap. - The diff in CXXRecordDecl.[cpp,h] is actually very small; it just mimics the changes to RecordDecl. For some reason 'svn' marks the entire file as changed. Why is CodeGen broken: - Codegen assumes that there is an equivalence between RecordDecl* and RecordType*. This was true before because we only created one RecordDecl* for a given RecordType*, but it is no longer true. I believe this shouldn't be too hard to change, but the patch was big enough as it is. I have tested this patch on both the clang test suite, and by running the static analyzer over Postgresql and a large Apple-internal project (mix of Objective-C and C). llvm-svn: 55839
2008-09-06 01:16:31 +08:00
QualType CXXMethodDecl::getThisType(ASTContext &C) const {
// C++ 9.3.2p1: The type of this in a member function of a class X is X*.
// If the member function is declared const, the type of this is const X*,
// if the member function is declared volatile, the type of this is
// volatile X*, and if the member function is declared const volatile,
// the type of this is const volatile X*.
Change struct forward declarations and definitions to use unique RecordDecls, as opposed to creating a single RecordDecl and reusing it. This change effects both RecordDecls and CXXRecordDecls, but does not effect EnumDecls (yet). The motivation of this patch is as follows: - Capture more source information, necessary for refactoring/rewriting clients. - Pave the way to resolve ownership issues with RecordDecls with the forthcoming addition of DeclGroups. Current caveats: - Until DeclGroups are in place, we will leak RecordDecls not explicitly referenced by the AST. For example: typedef struct { ... } x; The RecordDecl for the struct will be leaked because the TypedefDecl doesn't refer to it. This will be solved with DeclGroups. - This patch also (temporarily) breaks CodeGen. More below. High-level changes: - As before, TagType still refers to a TagDecl, but it doesn't own it. When a struct/union/class is first referenced, a RecordType and RecordDecl are created for it, and the RecordType refers to that RecordDecl. Later, if a new RecordDecl is created, the pointer to a RecordDecl in RecordType is updated to point to the RecordDecl that defines the struct/union/class. - TagDecl and RecordDecl now how a method 'getDefinition()' to return the TagDecl*/RecordDecl* that refers to the TagDecl* that defines a particular enum/struct/class/union. This is useful from going from a RecordDecl* that defines a forward declaration to the RecordDecl* that provides the actual definition. Note that this also works for EnumDecls, except that in this case there is no distinction between forward declarations and definitions (yet). - Clients should no longer assume that 'isDefinition()' returns true from a RecordDecl if the corresponding struct/union/class has been defined. isDefinition() only returns true if a particular RecordDecl is the defining Decl. Use 'getDefinition()' instead to determine if a struct has been defined. - The main changes to Sema happen in ActOnTag. To make the changes more incremental, I split off the processing of enums and structs et al into two code paths. Enums use the original code path (which is in ActOnTag) and structs use the ActOnTagStruct. Eventually the two code paths will be merged, but the idea was to preserve the original logic both for comparison and not to change the logic for both enums and structs all at once. - There is NO CHAINING of RecordDecls for the same RecordType. All RecordDecls that correspond to the same type simply have a pointer to that type. If we need to figure out what are all the RecordDecls for a given type we can build a backmap. - The diff in CXXRecordDecl.[cpp,h] is actually very small; it just mimics the changes to RecordDecl. For some reason 'svn' marks the entire file as changed. Why is CodeGen broken: - Codegen assumes that there is an equivalence between RecordDecl* and RecordType*. This was true before because we only created one RecordDecl* for a given RecordType*, but it is no longer true. I believe this shouldn't be too hard to change, but the patch was big enough as it is. I have tested this patch on both the clang test suite, and by running the static analyzer over Postgresql and a large Apple-internal project (mix of Objective-C and C). llvm-svn: 55839
2008-09-06 01:16:31 +08:00
assert(isInstance() && "No 'this' for static methods!");
QualType ClassTy = C.getTypeDeclType(getParent());
ClassTy = C.getQualifiedType(ClassTy,
Qualifiers::fromCVRMask(getTypeQualifiers()));
return C.getPointerType(ClassTy);
Change struct forward declarations and definitions to use unique RecordDecls, as opposed to creating a single RecordDecl and reusing it. This change effects both RecordDecls and CXXRecordDecls, but does not effect EnumDecls (yet). The motivation of this patch is as follows: - Capture more source information, necessary for refactoring/rewriting clients. - Pave the way to resolve ownership issues with RecordDecls with the forthcoming addition of DeclGroups. Current caveats: - Until DeclGroups are in place, we will leak RecordDecls not explicitly referenced by the AST. For example: typedef struct { ... } x; The RecordDecl for the struct will be leaked because the TypedefDecl doesn't refer to it. This will be solved with DeclGroups. - This patch also (temporarily) breaks CodeGen. More below. High-level changes: - As before, TagType still refers to a TagDecl, but it doesn't own it. When a struct/union/class is first referenced, a RecordType and RecordDecl are created for it, and the RecordType refers to that RecordDecl. Later, if a new RecordDecl is created, the pointer to a RecordDecl in RecordType is updated to point to the RecordDecl that defines the struct/union/class. - TagDecl and RecordDecl now how a method 'getDefinition()' to return the TagDecl*/RecordDecl* that refers to the TagDecl* that defines a particular enum/struct/class/union. This is useful from going from a RecordDecl* that defines a forward declaration to the RecordDecl* that provides the actual definition. Note that this also works for EnumDecls, except that in this case there is no distinction between forward declarations and definitions (yet). - Clients should no longer assume that 'isDefinition()' returns true from a RecordDecl if the corresponding struct/union/class has been defined. isDefinition() only returns true if a particular RecordDecl is the defining Decl. Use 'getDefinition()' instead to determine if a struct has been defined. - The main changes to Sema happen in ActOnTag. To make the changes more incremental, I split off the processing of enums and structs et al into two code paths. Enums use the original code path (which is in ActOnTag) and structs use the ActOnTagStruct. Eventually the two code paths will be merged, but the idea was to preserve the original logic both for comparison and not to change the logic for both enums and structs all at once. - There is NO CHAINING of RecordDecls for the same RecordType. All RecordDecls that correspond to the same type simply have a pointer to that type. If we need to figure out what are all the RecordDecls for a given type we can build a backmap. - The diff in CXXRecordDecl.[cpp,h] is actually very small; it just mimics the changes to RecordDecl. For some reason 'svn' marks the entire file as changed. Why is CodeGen broken: - Codegen assumes that there is an equivalence between RecordDecl* and RecordType*. This was true before because we only created one RecordDecl* for a given RecordType*, but it is no longer true. I believe this shouldn't be too hard to change, but the patch was big enough as it is. I have tested this patch on both the clang test suite, and by running the static analyzer over Postgresql and a large Apple-internal project (mix of Objective-C and C). llvm-svn: 55839
2008-09-06 01:16:31 +08:00
}
bool CXXMethodDecl::hasInlineBody() const {
// If this function is a template instantiation, look at the template from
// which it was instantiated.
const FunctionDecl *CheckFn = getTemplateInstantiationPattern();
if (!CheckFn)
CheckFn = this;
const FunctionDecl *fn;
return CheckFn->hasBody(fn) && !fn->isOutOfLine();
}
CXXCtorInitializer::CXXCtorInitializer(ASTContext &Context,
TypeSourceInfo *TInfo, bool IsVirtual,
SourceLocation L, Expr *Init,
SourceLocation R,
SourceLocation EllipsisLoc)
: Initializee(TInfo), MemberOrEllipsisLocation(EllipsisLoc), Init(Init),
LParenLoc(L), RParenLoc(R), IsDelegating(false), IsVirtual(IsVirtual),
IsWritten(false), SourceOrderOrNumArrayIndices(0)
{
}
CXXCtorInitializer::CXXCtorInitializer(ASTContext &Context,
FieldDecl *Member,
SourceLocation MemberLoc,
SourceLocation L, Expr *Init,
SourceLocation R)
: Initializee(Member), MemberOrEllipsisLocation(MemberLoc), Init(Init),
LParenLoc(L), RParenLoc(R), IsDelegating(false), IsVirtual(false),
IsWritten(false), SourceOrderOrNumArrayIndices(0)
{
}
CXXCtorInitializer::CXXCtorInitializer(ASTContext &Context,
IndirectFieldDecl *Member,
SourceLocation MemberLoc,
SourceLocation L, Expr *Init,
SourceLocation R)
: Initializee(Member), MemberOrEllipsisLocation(MemberLoc), Init(Init),
LParenLoc(L), RParenLoc(R), IsDelegating(false), IsVirtual(false),
IsWritten(false), SourceOrderOrNumArrayIndices(0)
{
}
CXXCtorInitializer::CXXCtorInitializer(ASTContext &Context,
TypeSourceInfo *TInfo,
SourceLocation L, Expr *Init,
SourceLocation R)
: Initializee(TInfo), MemberOrEllipsisLocation(), Init(Init),
LParenLoc(L), RParenLoc(R), IsDelegating(true), IsVirtual(false),
IsWritten(false), SourceOrderOrNumArrayIndices(0)
{
}
CXXCtorInitializer::CXXCtorInitializer(ASTContext &Context,
FieldDecl *Member,
SourceLocation MemberLoc,
SourceLocation L, Expr *Init,
SourceLocation R,
VarDecl **Indices,
unsigned NumIndices)
: Initializee(Member), MemberOrEllipsisLocation(MemberLoc), Init(Init),
LParenLoc(L), RParenLoc(R), IsVirtual(false),
IsWritten(false), SourceOrderOrNumArrayIndices(NumIndices)
Reimplement code generation for copying fields in the implicitly-generated copy constructor. Previously, Sema would perform some checking and instantiation to determine which copy constructors, etc., would be called, then CodeGen would attempt to figure out which copy constructor to call... but would get it wrong, or poke at an uninstantiated default argument, or fail in other ways. The new scheme is similar to what we now do for the implicit copy-assignment operator, where Sema performs all of the semantic analysis and builds specific ASTs that look similar to the ASTs we'd get from explicitly writing the copy constructor, so that CodeGen need only do a direct translation. However, it's not quite that simple because one cannot explicit write elementwise copy-construction of an array. So, I've extended CXXBaseOrMemberInitializer to contain a list of indexing variables used to copy-construct the elements. For example, if we have: struct A { A(const A&); }; struct B { A array[2][3]; }; then we generate an implicit copy assignment operator for B that looks something like this: B::B(const B &other) : array[i0][i1](other.array[i0][i1]) { } CodeGen will loop over the invented variables i0 and i1 to visit all elements in the array, so that each element in the destination array will be copy-constructed from the corresponding element in the source array. Of course, if we're dealing with arrays of scalars or class types with trivial copy-assignment operators, we just generate a memcpy rather than a loop. Fixes PR6928, PR5989, and PR6887. Boost.Regex now compiles and passes all of its regression tests. Conspicuously missing from this patch is handling for the exceptional case, where we need to destruct those objects that we have constructed. I'll address that case separately. llvm-svn: 103079
2010-05-05 13:51:00 +08:00
{
VarDecl **MyIndices = reinterpret_cast<VarDecl **> (this + 1);
memcpy(MyIndices, Indices, NumIndices * sizeof(VarDecl *));
}
CXXCtorInitializer *CXXCtorInitializer::Create(ASTContext &Context,
FieldDecl *Member,
SourceLocation MemberLoc,
SourceLocation L, Expr *Init,
SourceLocation R,
VarDecl **Indices,
unsigned NumIndices) {
void *Mem = Context.Allocate(sizeof(CXXCtorInitializer) +
Reimplement code generation for copying fields in the implicitly-generated copy constructor. Previously, Sema would perform some checking and instantiation to determine which copy constructors, etc., would be called, then CodeGen would attempt to figure out which copy constructor to call... but would get it wrong, or poke at an uninstantiated default argument, or fail in other ways. The new scheme is similar to what we now do for the implicit copy-assignment operator, where Sema performs all of the semantic analysis and builds specific ASTs that look similar to the ASTs we'd get from explicitly writing the copy constructor, so that CodeGen need only do a direct translation. However, it's not quite that simple because one cannot explicit write elementwise copy-construction of an array. So, I've extended CXXBaseOrMemberInitializer to contain a list of indexing variables used to copy-construct the elements. For example, if we have: struct A { A(const A&); }; struct B { A array[2][3]; }; then we generate an implicit copy assignment operator for B that looks something like this: B::B(const B &other) : array[i0][i1](other.array[i0][i1]) { } CodeGen will loop over the invented variables i0 and i1 to visit all elements in the array, so that each element in the destination array will be copy-constructed from the corresponding element in the source array. Of course, if we're dealing with arrays of scalars or class types with trivial copy-assignment operators, we just generate a memcpy rather than a loop. Fixes PR6928, PR5989, and PR6887. Boost.Regex now compiles and passes all of its regression tests. Conspicuously missing from this patch is handling for the exceptional case, where we need to destruct those objects that we have constructed. I'll address that case separately. llvm-svn: 103079
2010-05-05 13:51:00 +08:00
sizeof(VarDecl *) * NumIndices,
llvm::alignOf<CXXCtorInitializer>());
return new (Mem) CXXCtorInitializer(Context, Member, MemberLoc, L, Init, R,
Indices, NumIndices);
Reimplement code generation for copying fields in the implicitly-generated copy constructor. Previously, Sema would perform some checking and instantiation to determine which copy constructors, etc., would be called, then CodeGen would attempt to figure out which copy constructor to call... but would get it wrong, or poke at an uninstantiated default argument, or fail in other ways. The new scheme is similar to what we now do for the implicit copy-assignment operator, where Sema performs all of the semantic analysis and builds specific ASTs that look similar to the ASTs we'd get from explicitly writing the copy constructor, so that CodeGen need only do a direct translation. However, it's not quite that simple because one cannot explicit write elementwise copy-construction of an array. So, I've extended CXXBaseOrMemberInitializer to contain a list of indexing variables used to copy-construct the elements. For example, if we have: struct A { A(const A&); }; struct B { A array[2][3]; }; then we generate an implicit copy assignment operator for B that looks something like this: B::B(const B &other) : array[i0][i1](other.array[i0][i1]) { } CodeGen will loop over the invented variables i0 and i1 to visit all elements in the array, so that each element in the destination array will be copy-constructed from the corresponding element in the source array. Of course, if we're dealing with arrays of scalars or class types with trivial copy-assignment operators, we just generate a memcpy rather than a loop. Fixes PR6928, PR5989, and PR6887. Boost.Regex now compiles and passes all of its regression tests. Conspicuously missing from this patch is handling for the exceptional case, where we need to destruct those objects that we have constructed. I'll address that case separately. llvm-svn: 103079
2010-05-05 13:51:00 +08:00
}
TypeLoc CXXCtorInitializer::getBaseClassLoc() const {
if (isBaseInitializer())
return Initializee.get<TypeSourceInfo*>()->getTypeLoc();
else
return TypeLoc();
}
const Type *CXXCtorInitializer::getBaseClass() const {
if (isBaseInitializer())
return Initializee.get<TypeSourceInfo*>()->getType().getTypePtr();
else
return 0;
}
SourceLocation CXXCtorInitializer::getSourceLocation() const {
if (isAnyMemberInitializer())
return getMemberLocation();
if (isInClassMemberInitializer())
return getAnyMember()->getLocation();
if (TypeSourceInfo *TSInfo = Initializee.get<TypeSourceInfo*>())
return TSInfo->getTypeLoc().getLocalSourceRange().getBegin();
return SourceLocation();
}
SourceRange CXXCtorInitializer::getSourceRange() const {
if (isInClassMemberInitializer()) {
FieldDecl *D = getAnyMember();
if (Expr *I = D->getInClassInitializer())
return I->getSourceRange();
return SourceRange();
}
return SourceRange(getSourceLocation(), getRParenLoc());
}
void CXXConstructorDecl::anchor() { }
CXXConstructorDecl *
CXXConstructorDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(CXXConstructorDecl));
return new (Mem) CXXConstructorDecl(0, SourceLocation(),DeclarationNameInfo(),
QualType(), 0, false, false, false,false);
}
CXXConstructorDecl *
CXXConstructorDecl::Create(ASTContext &C, CXXRecordDecl *RD,
SourceLocation StartLoc,
const DeclarationNameInfo &NameInfo,
QualType T, TypeSourceInfo *TInfo,
bool isExplicit, bool isInline,
bool isImplicitlyDeclared, bool isConstexpr) {
assert(NameInfo.getName().getNameKind()
== DeclarationName::CXXConstructorName &&
"Name must refer to a constructor");
return new (C) CXXConstructorDecl(RD, StartLoc, NameInfo, T, TInfo,
isExplicit, isInline, isImplicitlyDeclared,
isConstexpr);
}
CXXConstructorDecl *CXXConstructorDecl::getTargetConstructor() const {
assert(isDelegatingConstructor() && "Not a delegating constructor!");
Expr *E = (*init_begin())->getInit()->IgnoreImplicit();
if (CXXConstructExpr *Construct = dyn_cast<CXXConstructExpr>(E))
return Construct->getConstructor();
return 0;
}
bool CXXConstructorDecl::isDefaultConstructor() const {
// C++ [class.ctor]p5:
// A default constructor for a class X is a constructor of class
// X that can be called without an argument.
return (getNumParams() == 0) ||
(getNumParams() > 0 && getParamDecl(0)->hasDefaultArg());
}
bool
CXXConstructorDecl::isCopyConstructor(unsigned &TypeQuals) const {
return isCopyOrMoveConstructor(TypeQuals) &&
getParamDecl(0)->getType()->isLValueReferenceType();
}
bool CXXConstructorDecl::isMoveConstructor(unsigned &TypeQuals) const {
return isCopyOrMoveConstructor(TypeQuals) &&
getParamDecl(0)->getType()->isRValueReferenceType();
}
/// \brief Determine whether this is a copy or move constructor.
bool CXXConstructorDecl::isCopyOrMoveConstructor(unsigned &TypeQuals) const {
// C++ [class.copy]p2:
// A non-template constructor for class X is a copy constructor
// if its first parameter is of type X&, const X&, volatile X& or
// const volatile X&, and either there are no other parameters
// or else all other parameters have default arguments (8.3.6).
// C++0x [class.copy]p3:
// A non-template constructor for class X is a move constructor if its
// first parameter is of type X&&, const X&&, volatile X&&, or
// const volatile X&&, and either there are no other parameters or else
// all other parameters have default arguments.
if ((getNumParams() < 1) ||
(getNumParams() > 1 && !getParamDecl(1)->hasDefaultArg()) ||
(getPrimaryTemplate() != 0) ||
(getDescribedFunctionTemplate() != 0))
return false;
const ParmVarDecl *Param = getParamDecl(0);
// Do we have a reference type?
const ReferenceType *ParamRefType = Param->getType()->getAs<ReferenceType>();
if (!ParamRefType)
return false;
// Is it a reference to our class type?
ASTContext &Context = getASTContext();
CanQualType PointeeType
= Context.getCanonicalType(ParamRefType->getPointeeType());
CanQualType ClassTy
= Context.getCanonicalType(Context.getTagDeclType(getParent()));
if (PointeeType.getUnqualifiedType() != ClassTy)
return false;
// FIXME: other qualifiers?
// We have a copy or move constructor.
TypeQuals = PointeeType.getCVRQualifiers();
return true;
}
2009-08-29 00:57:08 +08:00
bool CXXConstructorDecl::isConvertingConstructor(bool AllowExplicit) const {
// C++ [class.conv.ctor]p1:
// A constructor declared without the function-specifier explicit
// that can be called with a single parameter specifies a
// conversion from the type of its first parameter to the type of
// its class. Such a constructor is called a converting
// constructor.
2009-08-29 00:57:08 +08:00
if (isExplicit() && !AllowExplicit)
return false;
return (getNumParams() == 0 &&
getType()->getAs<FunctionProtoType>()->isVariadic()) ||
(getNumParams() == 1) ||
(getNumParams() > 1 && getParamDecl(1)->hasDefaultArg());
}
bool CXXConstructorDecl::isSpecializationCopyingObject() const {
if ((getNumParams() < 1) ||
(getNumParams() > 1 && !getParamDecl(1)->hasDefaultArg()) ||
(getPrimaryTemplate() == 0) ||
(getDescribedFunctionTemplate() != 0))
return false;
const ParmVarDecl *Param = getParamDecl(0);
ASTContext &Context = getASTContext();
CanQualType ParamType = Context.getCanonicalType(Param->getType());
// Is it the same as our our class type?
CanQualType ClassTy
= Context.getCanonicalType(Context.getTagDeclType(getParent()));
if (ParamType.getUnqualifiedType() != ClassTy)
return false;
return true;
}
const CXXConstructorDecl *CXXConstructorDecl::getInheritedConstructor() const {
// Hack: we store the inherited constructor in the overridden method table
method_iterator It = begin_overridden_methods();
if (It == end_overridden_methods())
return 0;
return cast<CXXConstructorDecl>(*It);
}
void
CXXConstructorDecl::setInheritedConstructor(const CXXConstructorDecl *BaseCtor){
// Hack: we store the inherited constructor in the overridden method table
assert(size_overridden_methods() == 0 && "Base ctor already set.");
addOverriddenMethod(BaseCtor);
}
void CXXDestructorDecl::anchor() { }
CXXDestructorDecl *
CXXDestructorDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(CXXDestructorDecl));
return new (Mem) CXXDestructorDecl(0, SourceLocation(), DeclarationNameInfo(),
QualType(), 0, false, false);
}
CXXDestructorDecl *
CXXDestructorDecl::Create(ASTContext &C, CXXRecordDecl *RD,
SourceLocation StartLoc,
const DeclarationNameInfo &NameInfo,
QualType T, TypeSourceInfo *TInfo,
bool isInline, bool isImplicitlyDeclared) {
assert(NameInfo.getName().getNameKind()
== DeclarationName::CXXDestructorName &&
"Name must refer to a destructor");
return new (C) CXXDestructorDecl(RD, StartLoc, NameInfo, T, TInfo, isInline,
isImplicitlyDeclared);
}
void CXXConversionDecl::anchor() { }
CXXConversionDecl *
CXXConversionDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(CXXConversionDecl));
return new (Mem) CXXConversionDecl(0, SourceLocation(), DeclarationNameInfo(),
QualType(), 0, false, false, false,
SourceLocation());
}
CXXConversionDecl *
CXXConversionDecl::Create(ASTContext &C, CXXRecordDecl *RD,
SourceLocation StartLoc,
const DeclarationNameInfo &NameInfo,
QualType T, TypeSourceInfo *TInfo,
bool isInline, bool isExplicit,
bool isConstexpr, SourceLocation EndLocation) {
assert(NameInfo.getName().getNameKind()
== DeclarationName::CXXConversionFunctionName &&
"Name must refer to a conversion function");
return new (C) CXXConversionDecl(RD, StartLoc, NameInfo, T, TInfo,
isInline, isExplicit, isConstexpr,
EndLocation);
}
void LinkageSpecDecl::anchor() { }
LinkageSpecDecl *LinkageSpecDecl::Create(ASTContext &C,
DeclContext *DC,
SourceLocation ExternLoc,
SourceLocation LangLoc,
LanguageIDs Lang,
SourceLocation RBraceLoc) {
return new (C) LinkageSpecDecl(DC, ExternLoc, LangLoc, Lang, RBraceLoc);
}
LinkageSpecDecl *LinkageSpecDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(LinkageSpecDecl));
return new (Mem) LinkageSpecDecl(0, SourceLocation(), SourceLocation(),
lang_c, SourceLocation());
}
void UsingDirectiveDecl::anchor() { }
UsingDirectiveDecl *UsingDirectiveDecl::Create(ASTContext &C, DeclContext *DC,
SourceLocation L,
SourceLocation NamespaceLoc,
NestedNameSpecifierLoc QualifierLoc,
SourceLocation IdentLoc,
NamedDecl *Used,
DeclContext *CommonAncestor) {
if (NamespaceDecl *NS = dyn_cast_or_null<NamespaceDecl>(Used))
Used = NS->getOriginalNamespace();
return new (C) UsingDirectiveDecl(DC, L, NamespaceLoc, QualifierLoc,
IdentLoc, Used, CommonAncestor);
}
UsingDirectiveDecl *
UsingDirectiveDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(UsingDirectiveDecl));
return new (Mem) UsingDirectiveDecl(0, SourceLocation(), SourceLocation(),
NestedNameSpecifierLoc(),
SourceLocation(), 0, 0);
}
NamespaceDecl *UsingDirectiveDecl::getNominatedNamespace() {
if (NamespaceAliasDecl *NA =
dyn_cast_or_null<NamespaceAliasDecl>(NominatedNamespace))
return NA->getNamespace();
return cast_or_null<NamespaceDecl>(NominatedNamespace);
}
void NamespaceDecl::anchor() { }
NamespaceDecl::NamespaceDecl(DeclContext *DC, bool Inline,
SourceLocation StartLoc,
SourceLocation IdLoc, IdentifierInfo *Id,
NamespaceDecl *PrevDecl)
: NamedDecl(Namespace, DC, IdLoc, Id), DeclContext(Namespace),
LocStart(StartLoc), RBraceLoc(), AnonOrFirstNamespaceAndInline(0, Inline)
{
setPreviousDeclaration(PrevDecl);
if (PrevDecl)
AnonOrFirstNamespaceAndInline.setPointer(PrevDecl->getOriginalNamespace());
}
NamespaceDecl *NamespaceDecl::Create(ASTContext &C, DeclContext *DC,
bool Inline, SourceLocation StartLoc,
SourceLocation IdLoc, IdentifierInfo *Id,
NamespaceDecl *PrevDecl) {
return new (C) NamespaceDecl(DC, Inline, StartLoc, IdLoc, Id, PrevDecl);
}
NamespaceDecl *NamespaceDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(NamespaceDecl));
return new (Mem) NamespaceDecl(0, false, SourceLocation(), SourceLocation(),
0, 0);
}
void NamespaceAliasDecl::anchor() { }
NamespaceAliasDecl *NamespaceAliasDecl::Create(ASTContext &C, DeclContext *DC,
SourceLocation UsingLoc,
SourceLocation AliasLoc,
IdentifierInfo *Alias,
NestedNameSpecifierLoc QualifierLoc,
SourceLocation IdentLoc,
NamedDecl *Namespace) {
if (NamespaceDecl *NS = dyn_cast_or_null<NamespaceDecl>(Namespace))
Namespace = NS->getOriginalNamespace();
return new (C) NamespaceAliasDecl(DC, UsingLoc, AliasLoc, Alias,
QualifierLoc, IdentLoc, Namespace);
}
NamespaceAliasDecl *
NamespaceAliasDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(NamespaceAliasDecl));
return new (Mem) NamespaceAliasDecl(0, SourceLocation(), SourceLocation(), 0,
NestedNameSpecifierLoc(),
SourceLocation(), 0);
}
void UsingShadowDecl::anchor() { }
UsingShadowDecl *
UsingShadowDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(UsingShadowDecl));
return new (Mem) UsingShadowDecl(0, SourceLocation(), 0, 0);
}
UsingDecl *UsingShadowDecl::getUsingDecl() const {
const UsingShadowDecl *Shadow = this;
while (const UsingShadowDecl *NextShadow =
dyn_cast<UsingShadowDecl>(Shadow->UsingOrNextShadow))
Shadow = NextShadow;
return cast<UsingDecl>(Shadow->UsingOrNextShadow);
}
void UsingDecl::anchor() { }
void UsingDecl::addShadowDecl(UsingShadowDecl *S) {
assert(std::find(shadow_begin(), shadow_end(), S) == shadow_end() &&
"declaration already in set");
assert(S->getUsingDecl() == this);
if (FirstUsingShadow.getPointer())
S->UsingOrNextShadow = FirstUsingShadow.getPointer();
FirstUsingShadow.setPointer(S);
}
void UsingDecl::removeShadowDecl(UsingShadowDecl *S) {
assert(std::find(shadow_begin(), shadow_end(), S) != shadow_end() &&
"declaration not in set");
assert(S->getUsingDecl() == this);
// Remove S from the shadow decl chain. This is O(n) but hopefully rare.
if (FirstUsingShadow.getPointer() == S) {
FirstUsingShadow.setPointer(
dyn_cast<UsingShadowDecl>(S->UsingOrNextShadow));
S->UsingOrNextShadow = this;
return;
}
UsingShadowDecl *Prev = FirstUsingShadow.getPointer();
while (Prev->UsingOrNextShadow != S)
Prev = cast<UsingShadowDecl>(Prev->UsingOrNextShadow);
Prev->UsingOrNextShadow = S->UsingOrNextShadow;
S->UsingOrNextShadow = this;
}
UsingDecl *UsingDecl::Create(ASTContext &C, DeclContext *DC, SourceLocation UL,
NestedNameSpecifierLoc QualifierLoc,
const DeclarationNameInfo &NameInfo,
bool IsTypeNameArg) {
return new (C) UsingDecl(DC, UL, QualifierLoc, NameInfo, IsTypeNameArg);
}
UsingDecl *UsingDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(UsingDecl));
return new (Mem) UsingDecl(0, SourceLocation(), NestedNameSpecifierLoc(),
DeclarationNameInfo(), false);
}
void UnresolvedUsingValueDecl::anchor() { }
UnresolvedUsingValueDecl *
UnresolvedUsingValueDecl::Create(ASTContext &C, DeclContext *DC,
SourceLocation UsingLoc,
NestedNameSpecifierLoc QualifierLoc,
const DeclarationNameInfo &NameInfo) {
return new (C) UnresolvedUsingValueDecl(DC, C.DependentTy, UsingLoc,
QualifierLoc, NameInfo);
}
UnresolvedUsingValueDecl *
UnresolvedUsingValueDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(UnresolvedUsingValueDecl));
return new (Mem) UnresolvedUsingValueDecl(0, QualType(), SourceLocation(),
NestedNameSpecifierLoc(),
DeclarationNameInfo());
}
void UnresolvedUsingTypenameDecl::anchor() { }
UnresolvedUsingTypenameDecl *
UnresolvedUsingTypenameDecl::Create(ASTContext &C, DeclContext *DC,
SourceLocation UsingLoc,
SourceLocation TypenameLoc,
NestedNameSpecifierLoc QualifierLoc,
SourceLocation TargetNameLoc,
DeclarationName TargetName) {
return new (C) UnresolvedUsingTypenameDecl(DC, UsingLoc, TypenameLoc,
QualifierLoc, TargetNameLoc,
TargetName.getAsIdentifierInfo());
}
UnresolvedUsingTypenameDecl *
UnresolvedUsingTypenameDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID,
sizeof(UnresolvedUsingTypenameDecl));
return new (Mem) UnresolvedUsingTypenameDecl(0, SourceLocation(),
SourceLocation(),
NestedNameSpecifierLoc(),
SourceLocation(),
0);
}
void StaticAssertDecl::anchor() { }
StaticAssertDecl *StaticAssertDecl::Create(ASTContext &C, DeclContext *DC,
SourceLocation StaticAssertLoc,
Expr *AssertExpr,
StringLiteral *Message,
SourceLocation RParenLoc) {
return new (C) StaticAssertDecl(DC, StaticAssertLoc, AssertExpr, Message,
RParenLoc);
}
StaticAssertDecl *StaticAssertDecl::CreateDeserialized(ASTContext &C,
unsigned ID) {
void *Mem = AllocateDeserializedDecl(C, ID, sizeof(StaticAssertDecl));
return new (Mem) StaticAssertDecl(0, SourceLocation(), 0, 0,SourceLocation());
}
static const char *getAccessName(AccessSpecifier AS) {
switch (AS) {
default:
case AS_none:
llvm_unreachable("Invalid access specifier!");
case AS_public:
return "public";
case AS_private:
return "private";
case AS_protected:
return "protected";
}
}
const DiagnosticBuilder &clang::operator<<(const DiagnosticBuilder &DB,
AccessSpecifier AS) {
return DB << getAccessName(AS);
}