Speed up check by using a recursive visitor.
Summary:
Use a recursive visitor instead of forEachDescendant() matcher.
The latter requires several layers of virtual function calls for each node and
it is more expensive than the visitor.
Benchmark results show improvement of ~6% walltime in clang-tidy.
Reviewers: alexfh
Subscribers: cfe-commits
Differential Revision: http://reviews.llvm.org/D20597
llvm-svn: 270714
diff --git a/clang-tools-extra/clang-tidy/readability/FunctionSizeCheck.cpp b/clang-tools-extra/clang-tidy/readability/FunctionSizeCheck.cpp
index dc411fc..d02972d 100644
--- a/clang-tools-extra/clang-tidy/readability/FunctionSizeCheck.cpp
+++ b/clang-tools-extra/clang-tidy/readability/FunctionSizeCheck.cpp
@@ -8,6 +8,7 @@
//===----------------------------------------------------------------------===//
#include "FunctionSizeCheck.h"
+#include "clang/AST/RecursiveASTVisitor.h"
#include "clang/ASTMatchers/ASTMatchFinder.h"
using namespace clang::ast_matchers;
@@ -16,6 +17,57 @@
namespace tidy {
namespace readability {
+class FunctionASTVisitor : public RecursiveASTVisitor<FunctionASTVisitor> {
+ using Base = RecursiveASTVisitor<FunctionASTVisitor>;
+
+public:
+ bool TraverseStmt(Stmt *Node) {
+ if (!Node)
+ return Base::TraverseStmt(Node);
+
+ if (TrackedParent.back() && !isa<CompoundStmt>(Node))
+ ++Info.Statements;
+
+ switch (Node->getStmtClass()) {
+ case Stmt::IfStmtClass:
+ case Stmt::WhileStmtClass:
+ case Stmt::DoStmtClass:
+ case Stmt::CXXForRangeStmtClass:
+ case Stmt::ForStmtClass:
+ case Stmt::SwitchStmtClass:
+ ++Info.Branches;
+ // fallthrough
+ case Stmt::CompoundStmtClass:
+ TrackedParent.push_back(true);
+ break;
+ default:
+ TrackedParent.push_back(false);
+ break;
+ }
+
+ Base::TraverseStmt(Node);
+
+ TrackedParent.pop_back();
+ return true;
+ }
+
+ bool TraverseDecl(Decl *Node) {
+ TrackedParent.push_back(false);
+ Base::TraverseDecl(Node);
+ TrackedParent.pop_back();
+ return true;
+ }
+
+ struct FunctionInfo {
+ FunctionInfo() : Lines(0), Statements(0), Branches(0) {}
+ unsigned Lines;
+ unsigned Statements;
+ unsigned Branches;
+ };
+ FunctionInfo Info;
+ std::vector<bool> TrackedParent;
+};
+
FunctionSizeCheck::FunctionSizeCheck(StringRef Name, ClangTidyContext *Context)
: ClangTidyCheck(Name, Context),
LineThreshold(Options.get("LineThreshold", -1U)),
@@ -29,76 +81,52 @@
}
void FunctionSizeCheck::registerMatchers(MatchFinder *Finder) {
- Finder->addMatcher(
- functionDecl(
- unless(isInstantiated()),
- forEachDescendant(
- stmt(unless(compoundStmt()),
- hasParent(stmt(anyOf(compoundStmt(), ifStmt(),
- anyOf(whileStmt(), doStmt(),
- cxxForRangeStmt(), forStmt())))))
- .bind("stmt"))).bind("func"),
- this);
+ Finder->addMatcher(functionDecl(unless(isInstantiated())).bind("func"), this);
}
void FunctionSizeCheck::check(const MatchFinder::MatchResult &Result) {
const auto *Func = Result.Nodes.getNodeAs<FunctionDecl>("func");
- FunctionInfo &FI = FunctionInfos[Func];
+ FunctionASTVisitor Visitor;
+ Visitor.TraverseDecl(const_cast<FunctionDecl *>(Func));
+ auto &FI = Visitor.Info;
+
+ if (FI.Statements == 0)
+ return;
// Count the lines including whitespace and comments. Really simple.
- if (!FI.Lines) {
- if (const Stmt *Body = Func->getBody()) {
- SourceManager *SM = Result.SourceManager;
- if (SM->isWrittenInSameFile(Body->getLocStart(), Body->getLocEnd())) {
- FI.Lines = SM->getSpellingLineNumber(Body->getLocEnd()) -
- SM->getSpellingLineNumber(Body->getLocStart());
- }
+ if (const Stmt *Body = Func->getBody()) {
+ SourceManager *SM = Result.SourceManager;
+ if (SM->isWrittenInSameFile(Body->getLocStart(), Body->getLocEnd())) {
+ FI.Lines = SM->getSpellingLineNumber(Body->getLocEnd()) -
+ SM->getSpellingLineNumber(Body->getLocStart());
}
}
- const auto *Statement = Result.Nodes.getNodeAs<Stmt>("stmt");
- ++FI.Statements;
-
- // TODO: switch cases, gotos
- if (isa<IfStmt>(Statement) || isa<WhileStmt>(Statement) ||
- isa<ForStmt>(Statement) || isa<SwitchStmt>(Statement) ||
- isa<DoStmt>(Statement) || isa<CXXForRangeStmt>(Statement))
- ++FI.Branches;
-}
-
-void FunctionSizeCheck::onEndOfTranslationUnit() {
- // If we're above the limit emit a warning.
- for (const auto &P : FunctionInfos) {
- const FunctionInfo &FI = P.second;
- if (FI.Lines > LineThreshold || FI.Statements > StatementThreshold ||
- FI.Branches > BranchThreshold) {
- diag(P.first->getLocation(),
- "function %0 exceeds recommended size/complexity thresholds")
- << P.first;
- }
-
- if (FI.Lines > LineThreshold) {
- diag(P.first->getLocation(),
- "%0 lines including whitespace and comments (threshold %1)",
- DiagnosticIDs::Note)
- << FI.Lines << LineThreshold;
- }
-
- if (FI.Statements > StatementThreshold) {
- diag(P.first->getLocation(), "%0 statements (threshold %1)",
- DiagnosticIDs::Note)
- << FI.Statements << StatementThreshold;
- }
-
- if (FI.Branches > BranchThreshold) {
- diag(P.first->getLocation(), "%0 branches (threshold %1)",
- DiagnosticIDs::Note)
- << FI.Branches << BranchThreshold;
- }
+ if (FI.Lines > LineThreshold || FI.Statements > StatementThreshold ||
+ FI.Branches > BranchThreshold) {
+ diag(Func->getLocation(),
+ "function %0 exceeds recommended size/complexity thresholds")
+ << Func;
}
- FunctionInfos.clear();
+ if (FI.Lines > LineThreshold) {
+ diag(Func->getLocation(),
+ "%0 lines including whitespace and comments (threshold %1)",
+ DiagnosticIDs::Note)
+ << FI.Lines << LineThreshold;
+ }
+
+ if (FI.Statements > StatementThreshold) {
+ diag(Func->getLocation(), "%0 statements (threshold %1)",
+ DiagnosticIDs::Note)
+ << FI.Statements << StatementThreshold;
+ }
+
+ if (FI.Branches > BranchThreshold) {
+ diag(Func->getLocation(), "%0 branches (threshold %1)", DiagnosticIDs::Note)
+ << FI.Branches << BranchThreshold;
+ }
}
} // namespace readability