include-what-you-use/iwyu_ast_util.h

849 lines
35 KiB
C
Raw Normal View History

//===--- iwyu_ast_util.h - clang-AST utilities for include-what-you-use ---===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
// Utilities that make it easier to work with Clang's AST.
#ifndef INCLUDE_WHAT_YOU_USE_IWYU_AST_UTIL_H_
#define INCLUDE_WHAT_YOU_USE_IWYU_AST_UTIL_H_
#include <map> // for map
#include <set> // for set
#include <string> // for string
2019-12-26 12:18:47 +00:00
#include "iwyu_port.h" // for CHECK_
#include "iwyu_use_flags.h"
#include "llvm/Support/Casting.h"
#include "clang/AST/DeclBase.h"
#include "clang/AST/NestedNameSpecifier.h"
#include "clang/AST/Stmt.h"
#include "clang/AST/TemplateBase.h"
#include "clang/AST/Type.h"
#include "clang/AST/TypeLoc.h"
#include "clang/Basic/SourceLocation.h"
namespace clang {
class CXXConstructExpr;
class CXXConstructorDecl;
class CXXDeleteExpr;
class CXXDestructorDecl;
class CXXMethodDecl;
class CXXRecordDecl;
class CallExpr;
class CastExpr;
class ClassTemplateDecl;
class Expr;
class FunctionDecl;
class NamedDecl;
class TagDecl;
class TemplateDecl;
class TemplateName;
class TranslationUnitDecl;
class TypeDecl;
class ValueDecl;
struct ASTTemplateArgumentListInfo;
} // namespace clang
namespace include_what_you_use {
using std::map;
using std::set;
using std::string;
//------------------------------------------------------------
// ASTNode and friends.
// ASTNode represents a single node of the AST tree. An AST node may be
// a statement, declaration, type, template-name, etc. ASTNode keeps
// track of its parent node, as we do AST traversal, allowing queries
// on the "context" of a node.
//
// We also store some state that's useful for iwyu. For instance,
// we store whether a node is in a 'forward-declarable' context
// (such as a function parameter), meaning all types seen below
// that node are legal to fowrard-declare according to c++.
class ASTNode {
public:
// In each case, the caller owns the object, and must guarantee it
// lives for at least as long as the ASTNode object does.
ASTNode(const clang::Decl* decl)
: kind_(kDeclKind), as_decl_(decl),
parent_(nullptr), in_fwd_decl_context_(false) { }
ASTNode(const clang::Stmt* stmt)
: kind_(kStmtKind), as_stmt_(stmt),
parent_(nullptr), in_fwd_decl_context_(false) { }
ASTNode(const clang::Type* type)
: kind_(kTypeKind), as_type_(type),
parent_(nullptr), in_fwd_decl_context_(false) { }
ASTNode(const clang::TypeLoc* typeloc)
: kind_(kTypelocKind), as_typeloc_(typeloc),
parent_(nullptr), in_fwd_decl_context_(false) { }
ASTNode(const clang::NestedNameSpecifier* nns)
: kind_(kNNSKind), as_nns_(nns),
parent_(nullptr), in_fwd_decl_context_(false) { }
ASTNode(const clang::NestedNameSpecifierLoc* nnsloc)
: kind_(kNNSLocKind), as_nnsloc_(nnsloc),
parent_(nullptr), in_fwd_decl_context_(false) { }
ASTNode(const clang::TemplateName* template_name)
: kind_(kTemplateNameKind), as_template_name_(template_name),
parent_(nullptr), in_fwd_decl_context_(false) { }
ASTNode(const clang::TemplateArgument* template_arg)
: kind_(kTemplateArgumentKind), as_template_arg_(template_arg),
parent_(nullptr), in_fwd_decl_context_(false) { }
ASTNode(const clang::TemplateArgumentLoc* template_argloc)
: kind_(kTemplateArgumentLocKind), as_template_argloc_(template_argloc),
parent_(nullptr), in_fwd_decl_context_(false) { }
// A 'forward-declare' context means some parent of us can be
// forward-declared, which means we can be too. e.g. in
// MyClass<Foo>* x, Foo is fwd-declarable because MyClass<Foo> is.
bool in_forward_declare_context() const {
return in_fwd_decl_context_;
}
void set_in_forward_declare_context(bool b) { in_fwd_decl_context_ = b; }
const ASTNode* parent() const {
return parent_;
}
void SetParent(const ASTNode* parent) {
parent_ = parent;
if (parent) // We inherit this from parent.
set_in_forward_declare_context(parent->in_forward_declare_context());
}
// The number of nodes above this node in the AST tree.
int depth() const {
int depth = 0;
for (const ASTNode* node = this; node != nullptr; node = node->parent_)
depth++;
return depth - 1; // don't count "this"
}
// If this node knows its location, returns it. If not, and it's
// likely its location is very close (say, within a few lines) of
// its parent, ask its parent. Unfortunately, there's nothing which
// tells us whether a parent's location is very close to its child.
// We assume that they always are (empirically this is true)
// *except* for the case the parent is in a macro: then it often
// happens that the parent belongs at the spelling location, while
// the child is a macro arg and hence belongs in the instantiation
// location. Those could be far away, even in different files. For
// example: '#define NEW_FUNC(cls) void Func(cls* x) {}'. Func is
// at the spelling loc, but its child Type 'cls' is at the
// instantiation loc. In that case, or if *no* ancestor of the
// current node knows its location, returns an invalid SourceLocation.
clang::SourceLocation GetLocation() const;
// Returns true if this node points to the exact same
// decl/typeloc/etc as the one you pass in. For Decl/Stmt/Type, the
// pointer is canonical (every instance of type X has the same
// clang::Type*). But for most, the value is canonical (each type
// has the same QualType but not QualType*). The IsA<> checks are
// needed to avoid false matches when target_node is nullptr.
bool ContentIs(const clang::Decl* target_node) const {
return IsA<clang::Decl>() && GetAs<clang::Decl>() == target_node;
}
bool ContentIs(const clang::Stmt* target_node) const {
return IsA<clang::Stmt>() && GetAs<clang::Stmt>() == target_node;
}
bool ContentIs(const clang::Type* target_node) const {
return IsA<clang::Type>() && GetAs<clang::Type>() == target_node;
}
bool ContentIs(const clang::TypeLoc* target_node) const {
if (!IsA<clang::TypeLoc>())
return false;
const clang::TypeLoc* type_loc = GetAs<clang::TypeLoc>();
if (type_loc == nullptr || target_node == nullptr)
return type_loc == target_node;
return *type_loc == *target_node;
}
// We don't define ContentIs() for other kinds of AST nodes
// (e.g. TemplateName) as it's non-trivial (Clang doesn't define
// equality comparison functions for them) and we don't need that
// yet.
// Returns true if the current node or any ancestor of it contains
// the exact same thing as ptr. One use of this is to check for
// recursion.
template<typename T> bool StackContainsContent(const T* ptr) const {
for (const ASTNode* node = this; node != nullptr; node = node->parent_) {
if (node->ContentIs(ptr))
return true;
}
return false;
}
// Generation 0 == you, 1 == parent, etc.
template<typename To> const To* GetAncestorAs(int generation) const {
const ASTNode* target_node = this;
for (; generation > 0; --generation) {
if (target_node->parent_ == nullptr)
return nullptr;
target_node = target_node->parent_;
}
// DynCast needs a dummy argument of type To* to help its resolution.
const To* dummy = nullptr;
return target_node->DynCast<To>(dummy);
}
// Convenience methods.
template<typename To> bool AncestorIsA(int generation) const {
return GetAncestorAs<To>(generation) != nullptr;
}
// Returns true if this node or any of its ancestors contains a T*.
template<typename T> bool HasAncestorOfType() const {
for (const ASTNode* node = this; node != nullptr; node = node->parent_) {
if (node->IsA<T>())
return true;
}
return false;
}
template<typename To> const To* GetParentAs() const {
return GetAncestorAs<To>(1);
}
template<typename To> bool ParentIsA() const {
return AncestorIsA<To>(1);
}
template<typename To> const To* GetAs() const {
return GetAncestorAs<To>(0);
}
template<typename To> bool IsA() const {
return AncestorIsA<To>(0);
}
private:
enum NodeKind {
kDeclKind, kStmtKind, kTypeKind, kTypelocKind, kNNSKind, kNNSLocKind,
kTemplateNameKind, kTemplateArgumentKind, kTemplateArgumentLocKind
};
// Returns a casted pointer if this object actually is of the given
// type (or a subclass of the given type), and nullptr otherwise. We
// have to use overloading on To's kind_, in these helper
// methods, in order to get llvm's dyn_cast to compile -- it gets
// upset (at compile time, sadly) if from-type and to-type aren't in
// the same type hierarchy. So To must be specified both in the
// template arg and in the method parameter.
template<typename To> const To* DynCast(const clang::Decl*) const {
if (kind_ != kDeclKind)
return nullptr;
return ::llvm::dyn_cast<To>(as_decl_);
}
template<typename To> const To* DynCast(const clang::Stmt*) const {
if (kind_ != kStmtKind)
return nullptr;
return ::llvm::dyn_cast<To>(as_stmt_);
}
template<typename To> const To* DynCast(const clang::Type*) const {
// We also will cast ourselves to a type if we're a typeloc.
// This simplifies a lot of code lower down that doesn't care
// to distinguish. For code that *does* care to distinguish,
// it should check for typelocs first:
// if (node.IsA<FooTypeLoc>()) ... else if (node.IsA<FooType>()) ...
if (kind_ == kTypelocKind)
return ::llvm::dyn_cast<To>(as_typeloc_->getTypePtr());
if (kind_ != kTypeKind)
return nullptr;
return ::llvm::dyn_cast<To>(as_type_);
}
template<typename To> const To* DynCast(const clang::TypeLoc*) const {
if (kind_ != kTypelocKind)
return nullptr;
return ::llvm::dyn_cast<To>(as_typeloc_);
}
template<typename To> const To* DynCast(
const clang::NestedNameSpecifier*) const {
// Like Type, this will cast to NNS if we're an NNSLoc. For code
// that cares to distinguish, it should check for nnslocs first.
if (kind_ == kNNSLocKind)
return as_nnsloc_->getNestedNameSpecifier();
if (kind_ != kNNSKind)
return nullptr;
return as_nns_;
}
template<typename To> const To* DynCast(
const clang::NestedNameSpecifierLoc*) const {
if (kind_ != kNNSLocKind)
return nullptr;
return as_nnsloc_;
}
template<typename To> const To* DynCast(const clang::TemplateName*) const {
if (kind_ != kTemplateNameKind)
return nullptr;
return as_template_name_;
}
template<typename To> const To* DynCast(
const clang::TemplateArgument*) const {
// We also will cast ourselves to a templateargument if we're a
// templateargumentloc. This simplifies a lot of code lower down
// that doesn't care to distinguish. For code that *does* care to
// distinguish, it should check for typelocs first.
if (kind_ == kTemplateArgumentLocKind)
return &as_template_argloc_->getArgument();
if (kind_ != kTemplateArgumentKind)
return nullptr;
return as_template_arg_;
}
template<typename To> const To* DynCast(
const clang::TemplateArgumentLoc*) const {
if (kind_ != kTemplateArgumentLocKind)
return nullptr;
return as_template_argloc_;
}
// We also allow casting to void*
template<typename Ignored> const void* DynCast(const void*) const {
switch (kind_) { // this is just to avoid aliasing violations.
case kDeclKind:
return as_decl_;
case kStmtKind:
return as_stmt_;
case kTypeKind:
return as_type_;
case kTypelocKind:
return as_typeloc_;
case kNNSKind:
return as_nns_;
case kNNSLocKind:
return as_nnsloc_;
case kTemplateNameKind:
return as_template_name_;
case kTemplateArgumentKind:
return as_template_arg_;
case kTemplateArgumentLocKind:
return as_template_argloc_;
}
CHECK_UNREACHABLE_("Unknown kind");
}
// If this node is of a type that knows its location, sets loc and
// returns true, otherwise returns false and leaves loc unchanged.
bool FillLocationIfKnown(clang::SourceLocation* loc) const;
const NodeKind kind_;
union {
const clang::Decl* as_decl_;
const clang::Stmt* as_stmt_;
const clang::Type* as_type_;
const clang::TypeLoc* as_typeloc_;
const clang::NestedNameSpecifier* as_nns_;
const clang::NestedNameSpecifierLoc* as_nnsloc_;
const clang::TemplateName* as_template_name_;
const clang::TemplateArgument* as_template_arg_;
const clang::TemplateArgumentLoc* as_template_argloc_;
};
const ASTNode* parent_;
bool in_fwd_decl_context_;
};
// --- Helper classes for ASTNode.
// An object of this type modifies a given variable in the constructor
// and restores its original value in the destructor.
template<typename T> class ValueSaver {
public:
ValueSaver(T* p, const T& newval) : ptr_(p), oldval_(*ptr_) {
*ptr_ = newval;
}
// The one-arg version just uses the current value as newval.
explicit ValueSaver(T* p) : ptr_(p), oldval_(*ptr_) { }
~ValueSaver() { *ptr_ = oldval_; }
private:
T* const ptr_;
const T oldval_;
};
// An object of this type updates current_ast_node_ to be the given
// node, and sets the given node's parent to be the old
// current_ast_node_. It then undoes this work in its destructor.
// The caller owns both old_current_node and new_current_node, and
// must make sure each of them lives at least as long as this object.
class CurrentASTNodeUpdater {
public:
CurrentASTNodeUpdater(ASTNode** old_current_node,
ASTNode* new_current_node)
: old_current_node_value_(*old_current_node),
node_saver_(old_current_node, new_current_node) {
new_current_node->SetParent(old_current_node_value_);
}
private:
ASTNode* const old_current_node_value_;
const ValueSaver<ASTNode*> node_saver_;
};
// --- Utilities for ASTNode.
// See if a given ast_node is a 'real' ElaboratedType(Loc). (An
// elaboration is 'class Foo myvar' instead of just 'Foo myvar'.)
// We avoid 'fake' elaborations that are caused because clang also
// uses ElaboratedType for namespaces ('ns::Foo myvar').
bool IsElaboratedTypeSpecifier(const ASTNode* ast_node);
// Walk up to parents of the given node so long as each parent is an
// elaborated type node.
// Can expand from a node representing 'X' to e.g. 'struct X' or 'mylib::X'.
const ASTNode* MostElaboratedAncestor(const ASTNode* ast_node);
// See if a given ast_node is a qualified name part of an ElaboratedType
// node (e.g. 'class ns::Foo x', 'class ::Global x' or 'class Outer::Inner x'.)
bool IsQualifiedNameNode(const ASTNode* ast_node);
// Return true if the given ast_node is inside a C++ method body. Do
// this by walking up the AST tree until you find a CXXMethodDecl,
// then see if the node just before you reached it is the body. We
// also check if the node is in an initializer (either explicitly or
// implicitly), or the implicit (non-body) code of a destructor.
bool IsNodeInsideCXXMethodBody(const ASTNode* ast_node);
// Return UseFlags for the current node.
// These flags provide context around the use to help later IWYU analysis,
UseFlags ComputeUseFlags(const ASTNode* ast_node);
// Return true if we're a nested tag type as written, that is, we're a
// class or enum decl inside another class decl. The parent class may be
// templated, but we should not be. (We could extend the function to
// handle that case, but there's been no need yet.)
bool IsNestedTagAsWritten(const ASTNode* ast_node);
// Is ast_node the 'D' in the following:
// template<template <typename A> class T = D> class C { ... }
// ('D' might be something like 'vector').
bool IsDefaultTemplateTemplateArg(const ASTNode* ast_node);
// Returns true if this node is a ConstructExpr that is being used to
// construct a field in a class (that is, it's part of a constructor
// initializer list).
bool IsCXXConstructExprInInitializer(const ASTNode* ast_node);
// Returns true if this node is a CXXConstructExpr in CXXNewExpr.
bool IsCXXConstructExprInNewExpr(const ASTNode* ast_node);
// If ASTNode is of a kind that has a qualifier (something that
// comes before the ::), return that, else return nullptr.
const clang::NestedNameSpecifier* GetQualifier(const ASTNode* ast_node);
// Returns the decl-context of the deepest decl in the ast-chain.
const clang::DeclContext* GetDeclContext(const ASTNode* ast_node);
//------------------------------------------------------------
// Helper functions for working with raw Clang AST nodes.
// --- Printers.
string PrintableLoc(clang::SourceLocation loc);
string PrintableDecl(const clang::Decl* decl, bool terse=true);
string PrintableStmt(const clang::Stmt* stmt);
string PrintableType(const clang::Type* type);
string PrintableTypeLoc(const clang::TypeLoc& typeloc);
string PrintableNestedNameSpecifier(const clang::NestedNameSpecifier* nns);
string PrintableTemplateName(const clang::TemplateName& tpl_name);
string PrintableTemplateArgument(const clang::TemplateArgument& arg);
string PrintableTemplateArgumentLoc(const clang::TemplateArgumentLoc& arg);
string PrintableASTNode(const ASTNode* node);
// These print to stderr. They're useful for debugging (e.g. inside gdb).
void PrintASTNode(const ASTNode* node);
void PrintStmt(const clang::Stmt* stmt);
// Written name means name without unwritten scopes. Unwritten scopes are
// anonymous and inline namespaces. Always consider using
// GetWrittenQualifiedNameAsString instead of
// NamedDecl::getQualifiedNameAsString.
string GetWrittenQualifiedNameAsString(const clang::NamedDecl* named_decl);
// --- Type conversion utilities.
namespace internal {
// For implementing DynCastFrom() -- don't use directly.
template <typename T>
class DynCastPtr {
public:
explicit DynCastPtr(T* ptr) : ptr_(ptr) { }
template <typename U> operator U*() const {
return ::llvm::dyn_cast_or_null<U>(ptr_);
}
private:
T* ptr_;
};
} // namespace internal
// dyn_cast<> and dyn_cast_or_null<> require the user to write the
// type of the target pointer, which is redundant when the result is
// immediately assigned to a newly declared pointer variable of the
// target type (the typical case). DynCastFrom() lets us omit the
// target type, e.g.
//
// if (const CXXConstructExpr* expr = DynCastFrom(source_expr)) ...
//
// instead of
//
// if (const CXXConstructExpr* expr =
// dyn_cast_or_null<CXXConstructExpr>(source_expr)) ...
//
// For readability, DynCastFrom() should only be used as the
// initializer of a variable declaration, where the target type is
// obvious.
template <typename T>
internal::DynCastPtr<T> DynCastFrom(T* ptr) {
return internal::DynCastPtr<T>(ptr);
}
// --- Utilities for Decl.
// Returns true if the decl is for a templatized function.
bool IsTemplatizedFunctionDecl(const clang::FunctionDecl* decl);
// Returns true if the given class has at least one implicit
// conversion constructor.
bool HasImplicitConversionCtor(const clang::CXXRecordDecl* cxx_class);
// Returns true if the given method is an override with covariant return type
// compared to its base.
bool HasCovariantReturnType(const clang::CXXMethodDecl* method_decl);
// If this decl is a (possibly templatized) tag decl, return the decl
// that defines the class, if present. Otherwise return nullptr.
const clang::TagDecl* GetTagDefinition(const clang::Decl* decl);
// Given a class, returns a SourceRange that encompasses the beginning
// of the class declaration (including template<> prefix, etc) to the
// class name. Used to determine where forward-declares are.
clang::SourceRange GetSourceRangeOfClassDecl(const clang::Decl* decl);
// Collect all late-parsed function templates in a translation unit.
set<clang::FunctionDecl*> GetLateParsedFunctionDecls(
clang::TranslationUnitDecl* decl);
A major revamp of the way we handle template arguments. The crux of the change is that the InstantiatedTemplateVisitor no longer takes a set of types-of-interest, but instead takes a map called the resugar_map. The resugar map is a tool to deal with the fact that clang canonicalizes all substituted template types (the "T"'s in a written template), so if you say typedef int MyTypedef; template<class T> void MyFunc() { T foo; } MyFunc<MyTypedef>(); clang will say the body is 'int foo;', not 'MyTypedef foo;'. This is difficult for include-what-you-use. There's one entry in the resugar map for every template parameter. Each entry has the form <canonical-type, type-as-written>, to make it easy to map from the canonical type back to the type-as-written. When the type-as-written has component types (e.g. both Foo* and vector<Foo> have a component type of Foo), there is also an entry for each component type. Both template classes and template functions have a complicating factor. For template classes, it's default template args, which are *not* written by the caller and usually want to attribute to the function-author. We store these in the resugar map with a value of NULL, to indicate they are default arguments that have no as-written form. For template functions, likewise some or all template arguments may be omitted by the caller, in which case the compiler derives them from the function arguments. We do something similar, looking for plausible mappings between types-as-written in template functions (or their components) and the clang-derived template arguments. This part could still use improvement. To really do it right, we'd need to refactor SemaTemplateDeduction to take an argument to not canonicalize deduced template arguments. In InstantiatedTemplateVisitor, we use the resugar-map to beef up CanIgnoreType() and ReportTypeUse(). We ignore types that are not in the resugar map (and thus do not correspond to template arguments as typed). When we do use a type, we resugar it before reporting its use. The net result is that we should see much lower incidence of clang reporting a weird dependency because of a type that the template-caller has never even heard of. We also use the fact we can tell an argument is a default template argument to decide if the template-author or the template-caller is responsible for the type. We say the template-caller is *unless* the author intends-to-provide the type, based on #includes. This handles the case when hash<T> uses the default implementation (in stl_hash, which stl_hashtable.h #includes) vs when it uses a user-provided implementation (which stl_hashtable.h obviously doesn't #include). Implementation-wise, we needed to beef up HandleFunctionCall to pass around the calling Expr, which holds the template arguments as written in some cases. We also needed to update the cache to handle the new data structures. Administrative note: wan reviewed this but had to bow out before finishing all the back-and-forth, and dsturtevant reviewed it but didn't feel qualified to judge entirely, so take the "R=" below with a grain of salt. This may require more work in the future. R=wan,dsturtevant DELTA=1173 (852 added, 138 deleted, 183 changed) Revision created by MOE tool push_codebase. MOE_MIGRATION=1077
2011-03-26 22:16:53 +00:00
// One can't have partial template specialization or default template
// args for function templates, but they're complicated in their own
// way: they can have deduced template arguments (deduced from the
// function arguments). When a templated function call does not
// specify the template arguments explicitly, but instead derives them
// from the function arguments, clang canonicalizes (desugars) the
// template args. For
// template<class T> void MyFunc(T t) { ... }
// MyFunc(typedef_type)
// clang will say this is a call to MyFunc<canonical_type>(). Also:
// MyFunc(my_int_vector)
// clang will say this is MyFunc<vector<int, alloc<int>>(), with
A major revamp of the way we handle template arguments. The crux of the change is that the InstantiatedTemplateVisitor no longer takes a set of types-of-interest, but instead takes a map called the resugar_map. The resugar map is a tool to deal with the fact that clang canonicalizes all substituted template types (the "T"'s in a written template), so if you say typedef int MyTypedef; template<class T> void MyFunc() { T foo; } MyFunc<MyTypedef>(); clang will say the body is 'int foo;', not 'MyTypedef foo;'. This is difficult for include-what-you-use. There's one entry in the resugar map for every template parameter. Each entry has the form <canonical-type, type-as-written>, to make it easy to map from the canonical type back to the type-as-written. When the type-as-written has component types (e.g. both Foo* and vector<Foo> have a component type of Foo), there is also an entry for each component type. Both template classes and template functions have a complicating factor. For template classes, it's default template args, which are *not* written by the caller and usually want to attribute to the function-author. We store these in the resugar map with a value of NULL, to indicate they are default arguments that have no as-written form. For template functions, likewise some or all template arguments may be omitted by the caller, in which case the compiler derives them from the function arguments. We do something similar, looking for plausible mappings between types-as-written in template functions (or their components) and the clang-derived template arguments. This part could still use improvement. To really do it right, we'd need to refactor SemaTemplateDeduction to take an argument to not canonicalize deduced template arguments. In InstantiatedTemplateVisitor, we use the resugar-map to beef up CanIgnoreType() and ReportTypeUse(). We ignore types that are not in the resugar map (and thus do not correspond to template arguments as typed). When we do use a type, we resugar it before reporting its use. The net result is that we should see much lower incidence of clang reporting a weird dependency because of a type that the template-caller has never even heard of. We also use the fact we can tell an argument is a default template argument to decide if the template-author or the template-caller is responsible for the type. We say the template-caller is *unless* the author intends-to-provide the type, based on #includes. This handles the case when hash<T> uses the default implementation (in stl_hash, which stl_hashtable.h #includes) vs when it uses a user-provided implementation (which stl_hashtable.h obviously doesn't #include). Implementation-wise, we needed to beef up HandleFunctionCall to pass around the calling Expr, which holds the template arguments as written in some cases. We also needed to update the cache to handle the new data structures. Administrative note: wan reviewed this but had to bow out before finishing all the back-and-forth, and dsturtevant reviewed it but didn't feel qualified to judge entirely, so take the "R=" below with a grain of salt. This may require more work in the future. R=wan,dsturtevant DELTA=1173 (852 added, 138 deleted, 183 changed) Revision created by MOE tool push_codebase. MOE_MIGRATION=1077
2011-03-26 22:16:53 +00:00
// no indication that alloc<int> is actually a default parameter.
// Equally bad:
// template<class T> void OtherFunc(MyClass<T> t) { ... }
// typedef MyClass<Foo> FooClass;
// OtherFunc(my_foo_class);
// clang will see T as MyClass<Foo> even though my_foo_class hides the
// use of Foo through the typedef.
// This routine attempts to solve all these problems by looking at
// the type-as-written for the actual arguments (and return value) to
// try to reverse engineer the derived-argument matching that was
// done. (It's easy in the rare cases the template args are
// explicitly specified.) It returns a map from the unsugared
// (canonical) types of each template argument to its sugared
// (as-written) type. For now we ignore non-type template args.
// We also include mappings for component types: if we have an entry
// 'vector<TypedefType>' -> 'vector<Foo>', we also add an entry
// 'TypedefType' -> 'Foo'.
A major revamp of the way we handle template arguments. The crux of the change is that the InstantiatedTemplateVisitor no longer takes a set of types-of-interest, but instead takes a map called the resugar_map. The resugar map is a tool to deal with the fact that clang canonicalizes all substituted template types (the "T"'s in a written template), so if you say typedef int MyTypedef; template<class T> void MyFunc() { T foo; } MyFunc<MyTypedef>(); clang will say the body is 'int foo;', not 'MyTypedef foo;'. This is difficult for include-what-you-use. There's one entry in the resugar map for every template parameter. Each entry has the form <canonical-type, type-as-written>, to make it easy to map from the canonical type back to the type-as-written. When the type-as-written has component types (e.g. both Foo* and vector<Foo> have a component type of Foo), there is also an entry for each component type. Both template classes and template functions have a complicating factor. For template classes, it's default template args, which are *not* written by the caller and usually want to attribute to the function-author. We store these in the resugar map with a value of NULL, to indicate they are default arguments that have no as-written form. For template functions, likewise some or all template arguments may be omitted by the caller, in which case the compiler derives them from the function arguments. We do something similar, looking for plausible mappings between types-as-written in template functions (or their components) and the clang-derived template arguments. This part could still use improvement. To really do it right, we'd need to refactor SemaTemplateDeduction to take an argument to not canonicalize deduced template arguments. In InstantiatedTemplateVisitor, we use the resugar-map to beef up CanIgnoreType() and ReportTypeUse(). We ignore types that are not in the resugar map (and thus do not correspond to template arguments as typed). When we do use a type, we resugar it before reporting its use. The net result is that we should see much lower incidence of clang reporting a weird dependency because of a type that the template-caller has never even heard of. We also use the fact we can tell an argument is a default template argument to decide if the template-author or the template-caller is responsible for the type. We say the template-caller is *unless* the author intends-to-provide the type, based on #includes. This handles the case when hash<T> uses the default implementation (in stl_hash, which stl_hashtable.h #includes) vs when it uses a user-provided implementation (which stl_hashtable.h obviously doesn't #include). Implementation-wise, we needed to beef up HandleFunctionCall to pass around the calling Expr, which holds the template arguments as written in some cases. We also needed to update the cache to handle the new data structures. Administrative note: wan reviewed this but had to bow out before finishing all the back-and-forth, and dsturtevant reviewed it but didn't feel qualified to judge entirely, so take the "R=" below with a grain of salt. This may require more work in the future. R=wan,dsturtevant DELTA=1173 (852 added, 138 deleted, 183 changed) Revision created by MOE tool push_codebase. MOE_MIGRATION=1077
2011-03-26 22:16:53 +00:00
// NOTE: This routine is far from perfect. To really do this right,
// we'd need to refactor SemaTemplateDeduction to take an argument to
// not canonicalize deduced template arguments.
// calling_expr should be a CallExpr, CXXConstructExpr, or DeclRefExpr.
map<const clang::Type*, const clang::Type*> GetTplTypeResugarMapForFunction(
const clang::FunctionDecl* decl, const clang::Expr* calling_expr);
// If class_decl is instantiated from a class template,
// returns the decl for that template; otherwise returns class_decl.
// As an example, consider this code:
// template<class T> class Foo { ... }; // template decl
// template<> class Foo<int> { ... }; // explicit specialization
// template class Foo<char>; // note: no body specified
// Foo<int> v1;
// Foo<float> v2;
// Foo<char> v3;
// The types Foo<int>, Foo<float>, and Foo<char> all have a
// corresponding decl. In the case of Foo<int>, the decl is the
// explicit specialization. This is *not* a decl that this routine
// deals with; the routine will just return its argument in this case.
// But in the case of Foo<float>, the decl is an implicit
// instantiation of Foo<T>, and this routine will return the decl for
// Foo<T>. Foo<char> is a rarer corner case: an explicit
// instantiation definition (it just causes code for that template
// case to be generated). It is treated the same as Foo<float>: we
// return the template decl, which provides the actual class body.
// We try to return a decl that's also a definition, when possible.
const clang::NamedDecl* GetInstantiatedFromDecl(
const clang::CXXRecordDecl* class_decl);
// For an implicitly instantiated templated c++ class -- that is, a
// class like vector<int> that isn't explicitly written in the source
// code but instead derived from vector<T> -- returns the
// class-as-written (vector<T>, in this case). For an implicitly
// instantiated template function -- Fn<int> when all that is written
// in the source code is Fn<T> -- returns the function-as-written.
// (In each case, prefers the definition of the class or function
// as-written, rather than a forward-declaration.) Otherwise, returns
// the original input.
const clang::NamedDecl* GetDefinitionAsWritten(const clang::NamedDecl* decl);
// Returns true if this decl is part of a friend decl.
bool IsFriendDecl(const clang::Decl* decl);
// Returns true if this decl is an explicit template instantiation declaration
// or definition.
bool IsExplicitInstantiation(const clang::Decl* decl);
// Returns true if this decl is nested inside an inline namespace.
bool IsInInlineNamespace(const clang::Decl* decl);
// Returns true if a named decl looks like a forward-declaration of a
// class (rather than a definition, a friend declaration, or an 'in
// place' declaration like 'struct Foo' in 'void MyFunc(struct Foo*);'
// Always returns false for enums.
bool IsForwardDecl(const clang::NamedDecl* decl);
// Returns true if this decl is defined inside another class/struct.
// Unlike IsNestedTagAsWritten(), which works on an ASTNode, this
// function considers decl to be nested even if it's not syntactically
// written inside its outer class (that is, 'class Foo::Bar {...}' is
// considered nested, even though it's not written inside Foo).
bool IsNestedClass(const clang::TagDecl* decl);
bool HasDefaultTemplateParameters(const clang::TemplateDecl* decl);
// For any decl that inherits from clang::Redeclarable *except* for
// classes and class templates -- enums, typedefs, functions, vars --
// returns all the declarations of decl. For any other decl, the
// output is just the input decl. Output decls are guaranteed to be
// of the same type as the input Decl. Because iwyu fundamentally
// treats classes different from other redeclarable types, it has
// its own separate function. (If that proves to be annoying, we
// can merge them.)
set<const clang::NamedDecl*> GetNonTagRedecls(const clang::NamedDecl* decl);
// Given a class, returns a set of all declarations of that class
// (forward-declarations and, if present, the definition). This
// accepts both TagDecls and ClassTemplateDecls -- the return Decls
// are guaranteed to be of the same type as the input Decl. Returns
// the empty set if the input is not a TagDecl or ClassTemplateDecl.
// Otherwise, always returns at least one element (since the input
// decl is its own redecl).
set<const clang::NamedDecl*> GetTagRedecls(const clang::NamedDecl* decl);
// Returns the redecl of decl that occurs first in the translation
// unit (that is, is the first one you'd see if you did 'cc -E').
// Returns nullptr if the input is not a TagDecl or ClassTemplateDecl.
const clang::NamedDecl* GetFirstRedecl(const clang::NamedDecl* decl);
// Given a class or class template, returns the declaration of that
// class that specifies the values of the default template arguments.
// If there are no default template arguments, returns nullptr.
const clang::ClassTemplateDecl* GetClassRedeclSpecifyingDefaultTplArgs(
const clang::ClassTemplateDecl* decl);
// Picks one redecl from GetTagRedecls() arbitrarily.
// This is used to recover from the clang bug that mixes friend decls
// with 'real' redecls (http://llvm.org/bugs/show_bug.cgi?id=8669);
// this function returns a 'real' redecl. If the input decl is a
// friend decl, returns an arbitrary non-friend redecl of it;
// otherwise returns decl itself.
// TODO(csilvers): remove once PR 8669 is fixed.
const clang::NamedDecl* GetNonfriendClassRedecl(const clang::NamedDecl* decl);
// Returns true if the innermost DeclContext for each decl is the
// same, and it's a class (or struct).
bool DeclsAreInSameClass(const clang::Decl* decl1, const clang::Decl* decl2);
// Returns true if the given decl/name is a builtin function
bool IsBuiltinFunction(const clang::NamedDecl* decl);
// --- Utilities for Type.
const clang::Type* GetTypeOf(const clang::Expr* expr);
// Returns the type of the constructed class.
const clang::Type* GetTypeOf(const clang::CXXConstructExpr* expr);
// Returns the type of the given variable, function, or enum declaration.
const clang::Type* GetTypeOf(const clang::ValueDecl* decl);
// ...or class, struct, union, enum, typedef, or template type.
const clang::Type* GetTypeOf(const clang::TypeDecl* decl);
A major revamp of the way we handle template arguments. The crux of the change is that the InstantiatedTemplateVisitor no longer takes a set of types-of-interest, but instead takes a map called the resugar_map. The resugar map is a tool to deal with the fact that clang canonicalizes all substituted template types (the "T"'s in a written template), so if you say typedef int MyTypedef; template<class T> void MyFunc() { T foo; } MyFunc<MyTypedef>(); clang will say the body is 'int foo;', not 'MyTypedef foo;'. This is difficult for include-what-you-use. There's one entry in the resugar map for every template parameter. Each entry has the form <canonical-type, type-as-written>, to make it easy to map from the canonical type back to the type-as-written. When the type-as-written has component types (e.g. both Foo* and vector<Foo> have a component type of Foo), there is also an entry for each component type. Both template classes and template functions have a complicating factor. For template classes, it's default template args, which are *not* written by the caller and usually want to attribute to the function-author. We store these in the resugar map with a value of NULL, to indicate they are default arguments that have no as-written form. For template functions, likewise some or all template arguments may be omitted by the caller, in which case the compiler derives them from the function arguments. We do something similar, looking for plausible mappings between types-as-written in template functions (or their components) and the clang-derived template arguments. This part could still use improvement. To really do it right, we'd need to refactor SemaTemplateDeduction to take an argument to not canonicalize deduced template arguments. In InstantiatedTemplateVisitor, we use the resugar-map to beef up CanIgnoreType() and ReportTypeUse(). We ignore types that are not in the resugar map (and thus do not correspond to template arguments as typed). When we do use a type, we resugar it before reporting its use. The net result is that we should see much lower incidence of clang reporting a weird dependency because of a type that the template-caller has never even heard of. We also use the fact we can tell an argument is a default template argument to decide if the template-author or the template-caller is responsible for the type. We say the template-caller is *unless* the author intends-to-provide the type, based on #includes. This handles the case when hash<T> uses the default implementation (in stl_hash, which stl_hashtable.h #includes) vs when it uses a user-provided implementation (which stl_hashtable.h obviously doesn't #include). Implementation-wise, we needed to beef up HandleFunctionCall to pass around the calling Expr, which holds the template arguments as written in some cases. We also needed to update the cache to handle the new data structures. Administrative note: wan reviewed this but had to bow out before finishing all the back-and-forth, and dsturtevant reviewed it but didn't feel qualified to judge entirely, so take the "R=" below with a grain of salt. This may require more work in the future. R=wan,dsturtevant DELTA=1173 (852 added, 138 deleted, 183 changed) Revision created by MOE tool push_codebase. MOE_MIGRATION=1077
2011-03-26 22:16:53 +00:00
// Template parameters are always reduced to the canonical type.
const clang::Type* GetCanonicalType(const clang::Type* type);
// Use Desugar to walk down the AST skipping type sugar nodes until a non-sugar
// node is found, much like Type::getUnqualifiedDesugaredType.
// IWYU has a slightly more liberal notion of sugar than Clang does:
// typedefs, using types and template specializations are not considered sugar,
// because they need to be respected in IWYU analysis.
const clang::Type* Desugar(const clang::Type* type);
iwyu was egregiously wrong in how it handled template arguments using the 'precomputed cache'. In such situations, it totally ignored the currently active resugar_map, replacing it with one of its own. That worked fine for types outside of templates, but not fine for types inside (such as a 'hash_map<T>' inside a templated class). I "fixed" this. "Fixed" is in quotes because this turned up a whole slew of other problems I don't even attempt to resolve here (though I spent a few hours trying). One is that it's possible to have a type like hash_map that has some arguments that are dependent and some that aren't; in theory, for these types, we can correctly attribute the use to the template author or template instantiator depending on which type it is. But I can't figure out how to get clang to do any meaningful analysis of incomplete (dependent) types, so I've punted on that for now. The second thing wrong is I jumped through all sorts of hoops to handle default template arguments correctly, so if a class has a hash_map<T> and you instantiate T with string, you're also made responsible for hash<string>. This *should* work, but clang is giving hash<string> a type I don't expect (RecordType, not TemplateSpecializationType), and I don't know how to deal with that -- I don't know how to extract the 'string' part of this RecordType. Ugh. I punt on this now, as well. Even in this incomplete form, it's enough to resolve a P1 bug, so I figure it's worth putting in. R=dsturtevant DELTA=141 (97 added, 7 deleted, 37 changed) Revision created by MOE tool push_codebase. MOE_MIGRATION=3147
2011-08-31 12:42:23 +01:00
// A 'component' of a type is a type beneath it in the AST tree.
// So 'Foo*' has component 'Foo', as does 'vector<Foo>', while
// vector<pair<Foo, Bar>> has components pair<Foo,Bar>, Foo, and Bar.
iwyu was egregiously wrong in how it handled template arguments using the 'precomputed cache'. In such situations, it totally ignored the currently active resugar_map, replacing it with one of its own. That worked fine for types outside of templates, but not fine for types inside (such as a 'hash_map<T>' inside a templated class). I "fixed" this. "Fixed" is in quotes because this turned up a whole slew of other problems I don't even attempt to resolve here (though I spent a few hours trying). One is that it's possible to have a type like hash_map that has some arguments that are dependent and some that aren't; in theory, for these types, we can correctly attribute the use to the template author or template instantiator depending on which type it is. But I can't figure out how to get clang to do any meaningful analysis of incomplete (dependent) types, so I've punted on that for now. The second thing wrong is I jumped through all sorts of hoops to handle default template arguments correctly, so if a class has a hash_map<T> and you instantiate T with string, you're also made responsible for hash<string>. This *should* work, but clang is giving hash<string> a type I don't expect (RecordType, not TemplateSpecializationType), and I don't know how to deal with that -- I don't know how to extract the 'string' part of this RecordType. Ugh. I punt on this now, as well. Even in this incomplete form, it's enough to resolve a P1 bug, so I figure it's worth putting in. R=dsturtevant DELTA=141 (97 added, 7 deleted, 37 changed) Revision created by MOE tool push_codebase. MOE_MIGRATION=3147
2011-08-31 12:42:23 +01:00
set<const clang::Type*> GetComponentsOfType(const clang::Type* type);
// Returns true if the type has any template arguments.
bool IsTemplatizedType(const clang::Type* type);
// Returns true if the type is a RecordType or a TemplateSpecializationType.
bool IsClassType(const clang::Type* type);
// Returns true if any type involved (recursively examining template
// arguments) satisfies the given predicate.
bool InvolvesTypeForWhich(const clang::Type* type,
std::function<bool(const clang::Type*)> pred);
// Returns true if type is a pointer type (pointer or reference,
// looking through elaborations like 'class Foo*' (vs 'Foo*'),
// but *not* following typedefs (which is why we can't just use
// type->isPointerType()).
// TODO(csilvers): what about array-type?
bool IsPointerOrReferenceAsWritten(const clang::Type* type);
const clang::Type* RemoveReferenceAsWritten(const clang::Type* type);
// Gets rid of all the pointers and references to get to the 'base'
// type. Also removes all elaborations (like 'class' keyword). We
// can't just use the default desugar() routine, because we *don't*
// want to look through typedefs.
const clang::Type* RemovePointersAndReferencesAsWritten(
const clang::Type* type);
// Remove one layer of pointers (or references) from type. We go
// through typedefs and the like, but only if we have to in order to
// figure out the dereferenced type, which is why we don't just use
// desugar(). Returns nullptr if not a pointer.
const clang::Type* RemovePointerFromType(const clang::Type* type);
// This follows typedefs/etc to remove pointers, if necessary.
const clang::Type* RemovePointersAndReferences(const clang::Type* type);
// To the best of our ability, maps a type to a declaration of that
// type *that is written in the source code*. For most types, of
// course, declarations are always written, but for implicit template
// specializations they typically aren't. For instance, if the type
// is vector<int>, there is a decl for 'class vector<int>', but it's
// not written anywhere; what is written is generic code, for
// 'class vector<T>'. 'class vector<int>' is an implicit declaration,
// constructed at need. This routine does not return that implicit
// declaration.
// If the type is a substituted template parameter, we get a
// decl for the substituted type. That is, for this code:
// template<class T> void MyFunc() { T foo; }
// MyFunc<MyClass>();
// If we're evaluating MyFunc<MyClass>() and see the type that's in
// the function body, this function returns a decl for MyClass.
// If the type is built-in, or otherwise doesn't have a decl,
// this function returns nullptr.
const clang::NamedDecl* TypeToDeclAsWritten(const clang::Type* type);
// This is similar to TypeToDeclAsWritten, but in this case we are less
// interested in how the type was written; we want the Decl which we can
// explore the contents of, for example to determine which of its template
// arguments are used in a manner that constitutes a full use.
//
// The difference arises particularly for type aliases, where
// TypeToDeclAsWritten returns the Decl for the alias, whereas
// TypeToDeclForContent returns the underlying aliased Decl.
const clang::NamedDecl* TypeToDeclForContent(const clang::Type* type);
// Returns true if it's possible to implicitly convert a value of a
// different type to 'type' via an implicit constructor.
bool HasImplicitConversionConstructor(const clang::Type* type);
A major revamp of the way we handle template arguments. The crux of the change is that the InstantiatedTemplateVisitor no longer takes a set of types-of-interest, but instead takes a map called the resugar_map. The resugar map is a tool to deal with the fact that clang canonicalizes all substituted template types (the "T"'s in a written template), so if you say typedef int MyTypedef; template<class T> void MyFunc() { T foo; } MyFunc<MyTypedef>(); clang will say the body is 'int foo;', not 'MyTypedef foo;'. This is difficult for include-what-you-use. There's one entry in the resugar map for every template parameter. Each entry has the form <canonical-type, type-as-written>, to make it easy to map from the canonical type back to the type-as-written. When the type-as-written has component types (e.g. both Foo* and vector<Foo> have a component type of Foo), there is also an entry for each component type. Both template classes and template functions have a complicating factor. For template classes, it's default template args, which are *not* written by the caller and usually want to attribute to the function-author. We store these in the resugar map with a value of NULL, to indicate they are default arguments that have no as-written form. For template functions, likewise some or all template arguments may be omitted by the caller, in which case the compiler derives them from the function arguments. We do something similar, looking for plausible mappings between types-as-written in template functions (or their components) and the clang-derived template arguments. This part could still use improvement. To really do it right, we'd need to refactor SemaTemplateDeduction to take an argument to not canonicalize deduced template arguments. In InstantiatedTemplateVisitor, we use the resugar-map to beef up CanIgnoreType() and ReportTypeUse(). We ignore types that are not in the resugar map (and thus do not correspond to template arguments as typed). When we do use a type, we resugar it before reporting its use. The net result is that we should see much lower incidence of clang reporting a weird dependency because of a type that the template-caller has never even heard of. We also use the fact we can tell an argument is a default template argument to decide if the template-author or the template-caller is responsible for the type. We say the template-caller is *unless* the author intends-to-provide the type, based on #includes. This handles the case when hash<T> uses the default implementation (in stl_hash, which stl_hashtable.h #includes) vs when it uses a user-provided implementation (which stl_hashtable.h obviously doesn't #include). Implementation-wise, we needed to beef up HandleFunctionCall to pass around the calling Expr, which holds the template arguments as written in some cases. We also needed to update the cache to handle the new data structures. Administrative note: wan reviewed this but had to bow out before finishing all the back-and-forth, and dsturtevant reviewed it but didn't feel qualified to judge entirely, so take the "R=" below with a grain of salt. This may require more work in the future. R=wan,dsturtevant DELTA=1173 (852 added, 138 deleted, 183 changed) Revision created by MOE tool push_codebase. MOE_MIGRATION=1077
2011-03-26 22:16:53 +00:00
// clang desugars template arguments: follows typedefs, etc. We
// want the unsugared type, so this function provides a map from
// the desugared type back to the original type-as-written, as
// determined from the class's template arguments. For default
// template arguments that are not specified by the caller, we
// map the type to nullptr, to indicate there's no inherent sugaring.
// We also include mappings for component types: if we have an entry
// 'vector<TypedefType>' -> 'vector<Foo>', we also add an entry
// 'TypedefType' -> 'Foo'.
// For ease of calling, this accept any type, but will return an empty
A major revamp of the way we handle template arguments. The crux of the change is that the InstantiatedTemplateVisitor no longer takes a set of types-of-interest, but instead takes a map called the resugar_map. The resugar map is a tool to deal with the fact that clang canonicalizes all substituted template types (the "T"'s in a written template), so if you say typedef int MyTypedef; template<class T> void MyFunc() { T foo; } MyFunc<MyTypedef>(); clang will say the body is 'int foo;', not 'MyTypedef foo;'. This is difficult for include-what-you-use. There's one entry in the resugar map for every template parameter. Each entry has the form <canonical-type, type-as-written>, to make it easy to map from the canonical type back to the type-as-written. When the type-as-written has component types (e.g. both Foo* and vector<Foo> have a component type of Foo), there is also an entry for each component type. Both template classes and template functions have a complicating factor. For template classes, it's default template args, which are *not* written by the caller and usually want to attribute to the function-author. We store these in the resugar map with a value of NULL, to indicate they are default arguments that have no as-written form. For template functions, likewise some or all template arguments may be omitted by the caller, in which case the compiler derives them from the function arguments. We do something similar, looking for plausible mappings between types-as-written in template functions (or their components) and the clang-derived template arguments. This part could still use improvement. To really do it right, we'd need to refactor SemaTemplateDeduction to take an argument to not canonicalize deduced template arguments. In InstantiatedTemplateVisitor, we use the resugar-map to beef up CanIgnoreType() and ReportTypeUse(). We ignore types that are not in the resugar map (and thus do not correspond to template arguments as typed). When we do use a type, we resugar it before reporting its use. The net result is that we should see much lower incidence of clang reporting a weird dependency because of a type that the template-caller has never even heard of. We also use the fact we can tell an argument is a default template argument to decide if the template-author or the template-caller is responsible for the type. We say the template-caller is *unless* the author intends-to-provide the type, based on #includes. This handles the case when hash<T> uses the default implementation (in stl_hash, which stl_hashtable.h #includes) vs when it uses a user-provided implementation (which stl_hashtable.h obviously doesn't #include). Implementation-wise, we needed to beef up HandleFunctionCall to pass around the calling Expr, which holds the template arguments as written in some cases. We also needed to update the cache to handle the new data structures. Administrative note: wan reviewed this but had to bow out before finishing all the back-and-forth, and dsturtevant reviewed it but didn't feel qualified to judge entirely, so take the "R=" below with a grain of salt. This may require more work in the future. R=wan,dsturtevant DELTA=1173 (852 added, 138 deleted, 183 changed) Revision created by MOE tool push_codebase. MOE_MIGRATION=1077
2011-03-26 22:16:53 +00:00
// map for any input that's not a template specialization type.
map<const clang::Type*, const clang::Type*> GetTplTypeResugarMapForClass(
const clang::Type* type);
// Like GetTplTypeResugarMapForClass, but if a type has
// components (for instance, 'Foo*' and 'vector<Foo>' both
// have a component Foo), we don't include the components
// in the result-map.
map<const clang::Type*, const clang::Type*>
GetTplTypeResugarMapForClassNoComponentTypes(const clang::Type* type);
// Returns true if, for the given enumeration type, opaque (i.e. forward,
// in fact) declarations are allowed. It means that the enumeration should be
// either scoped or unscoped with explicitly stated underlying type,
// according to the standard.
bool CanBeOpaqueDeclared(const clang::EnumType* type);
// --- Utilities for Stmt.
// Returns true if the given expr is '&<something>'.
bool IsAddressOf(const clang::Expr* expr);
// If this function call comes from a class method -- either a normal
// one or a static one -- returns the type of the class. Otherwise,
// returns nullptr. Note that static class methods do *not* have a
// CXXMemberCallExpr type, which is why we take a CallExpr.
const clang::Type* TypeOfParentIfMethod(const clang::CallExpr* expr);
// Given a function call, return the first argument that's a class
// (possibly a template specialization). Note we ignore pointers to a
// class. This is used with 'free' overloaded operators ('ostream&
// operator<<(ostream& a, int x)' to figure out what class the
// operator 'logically' belongs to. This is a heuristic (the operator
// may "belong" to more than one argument, for instance), but covers
// all the common cases. Returns nullptr if no class-type argument is
// found.
const clang::Expr* GetFirstClassArgument(clang::CallExpr* expr);
// Returns nullptr if we're deleting an argument that has no destructor.
const clang::CXXDestructorDecl* GetDestructorForDeleteExpr(
const clang::CXXDeleteExpr* expr);
// Returns nullptr if the constructor has no corresponding destructor.
const clang::CXXDestructorDecl* GetSiblingDestructorFor(
const clang::CXXConstructorDecl* ctor);
const clang::CXXDestructorDecl* GetSiblingDestructorFor(
const clang::CXXConstructExpr* ctor_expr);
// Figuring out the function type is non-trivial because the callee
// may be a function pointer. This code is based on clang's Expr.cpp.
// Should never return nullptr.
const clang::FunctionType* GetCalleeFunctionType(clang::CallExpr* expr);
A major revamp of the way we handle template arguments. The crux of the change is that the InstantiatedTemplateVisitor no longer takes a set of types-of-interest, but instead takes a map called the resugar_map. The resugar map is a tool to deal with the fact that clang canonicalizes all substituted template types (the "T"'s in a written template), so if you say typedef int MyTypedef; template<class T> void MyFunc() { T foo; } MyFunc<MyTypedef>(); clang will say the body is 'int foo;', not 'MyTypedef foo;'. This is difficult for include-what-you-use. There's one entry in the resugar map for every template parameter. Each entry has the form <canonical-type, type-as-written>, to make it easy to map from the canonical type back to the type-as-written. When the type-as-written has component types (e.g. both Foo* and vector<Foo> have a component type of Foo), there is also an entry for each component type. Both template classes and template functions have a complicating factor. For template classes, it's default template args, which are *not* written by the caller and usually want to attribute to the function-author. We store these in the resugar map with a value of NULL, to indicate they are default arguments that have no as-written form. For template functions, likewise some or all template arguments may be omitted by the caller, in which case the compiler derives them from the function arguments. We do something similar, looking for plausible mappings between types-as-written in template functions (or their components) and the clang-derived template arguments. This part could still use improvement. To really do it right, we'd need to refactor SemaTemplateDeduction to take an argument to not canonicalize deduced template arguments. In InstantiatedTemplateVisitor, we use the resugar-map to beef up CanIgnoreType() and ReportTypeUse(). We ignore types that are not in the resugar map (and thus do not correspond to template arguments as typed). When we do use a type, we resugar it before reporting its use. The net result is that we should see much lower incidence of clang reporting a weird dependency because of a type that the template-caller has never even heard of. We also use the fact we can tell an argument is a default template argument to decide if the template-author or the template-caller is responsible for the type. We say the template-caller is *unless* the author intends-to-provide the type, based on #includes. This handles the case when hash<T> uses the default implementation (in stl_hash, which stl_hashtable.h #includes) vs when it uses a user-provided implementation (which stl_hashtable.h obviously doesn't #include). Implementation-wise, we needed to beef up HandleFunctionCall to pass around the calling Expr, which holds the template arguments as written in some cases. We also needed to update the cache to handle the new data structures. Administrative note: wan reviewed this but had to bow out before finishing all the back-and-forth, and dsturtevant reviewed it but didn't feel qualified to judge entirely, so take the "R=" below with a grain of salt. This may require more work in the future. R=wan,dsturtevant DELTA=1173 (852 added, 138 deleted, 183 changed) Revision created by MOE tool push_codebase. MOE_MIGRATION=1077
2011-03-26 22:16:53 +00:00
// Returns the list of explicit template args for all exprs that support
// such a concept (declrefexpr, memberexpr), and empty list if none is present.
clang::TemplateArgumentListInfo GetExplicitTplArgs(const clang::Expr* expr);
A major revamp of the way we handle template arguments. The crux of the change is that the InstantiatedTemplateVisitor no longer takes a set of types-of-interest, but instead takes a map called the resugar_map. The resugar map is a tool to deal with the fact that clang canonicalizes all substituted template types (the "T"'s in a written template), so if you say typedef int MyTypedef; template<class T> void MyFunc() { T foo; } MyFunc<MyTypedef>(); clang will say the body is 'int foo;', not 'MyTypedef foo;'. This is difficult for include-what-you-use. There's one entry in the resugar map for every template parameter. Each entry has the form <canonical-type, type-as-written>, to make it easy to map from the canonical type back to the type-as-written. When the type-as-written has component types (e.g. both Foo* and vector<Foo> have a component type of Foo), there is also an entry for each component type. Both template classes and template functions have a complicating factor. For template classes, it's default template args, which are *not* written by the caller and usually want to attribute to the function-author. We store these in the resugar map with a value of NULL, to indicate they are default arguments that have no as-written form. For template functions, likewise some or all template arguments may be omitted by the caller, in which case the compiler derives them from the function arguments. We do something similar, looking for plausible mappings between types-as-written in template functions (or their components) and the clang-derived template arguments. This part could still use improvement. To really do it right, we'd need to refactor SemaTemplateDeduction to take an argument to not canonicalize deduced template arguments. In InstantiatedTemplateVisitor, we use the resugar-map to beef up CanIgnoreType() and ReportTypeUse(). We ignore types that are not in the resugar map (and thus do not correspond to template arguments as typed). When we do use a type, we resugar it before reporting its use. The net result is that we should see much lower incidence of clang reporting a weird dependency because of a type that the template-caller has never even heard of. We also use the fact we can tell an argument is a default template argument to decide if the template-author or the template-caller is responsible for the type. We say the template-caller is *unless* the author intends-to-provide the type, based on #includes. This handles the case when hash<T> uses the default implementation (in stl_hash, which stl_hashtable.h #includes) vs when it uses a user-provided implementation (which stl_hashtable.h obviously doesn't #include). Implementation-wise, we needed to beef up HandleFunctionCall to pass around the calling Expr, which holds the template arguments as written in some cases. We also needed to update the cache to handle the new data structures. Administrative note: wan reviewed this but had to bow out before finishing all the back-and-forth, and dsturtevant reviewed it but didn't feel qualified to judge entirely, so take the "R=" below with a grain of salt. This may require more work in the future. R=wan,dsturtevant DELTA=1173 (852 added, 138 deleted, 183 changed) Revision created by MOE tool push_codebase. MOE_MIGRATION=1077
2011-03-26 22:16:53 +00:00
// Return the kind- or class-name for various AST node types.
std::string GetKindName(const clang::Decl* decl);
std::string GetKindName(const clang::Stmt* stmt);
std::string GetKindName(const clang::Type* type);
std::string GetKindName(const clang::TypeLoc typeloc);
// Returns true if decl is entirely inside a function, which implies it's only
// visible from said function.
bool IsDeclaredInsideFunction(const clang::Decl* decl);
} // namespace include_what_you_use
#endif // INCLUDE_WHAT_YOU_USE_IWYU_AST_UTIL_H_