[clangd] Lib to compute and represent selection under cursor.

Summary:
The primary problem this solves is to expose the codeAction selection to
AST-based refactorings in a way that makes it easy and efficient for them to
bind to the right parts of the AST.

It should also allow us to make XRefs based features (textDocument/definition)
more robust, more easily implement textDocument/typeDefinition etc.
As an example, template parameter references can be identified without special
handling.
There should be slight speedup too: we can prune most of the AST traversal
in most cases.

Elephant in the room: this is similar-but-different to Tooling/Refactoring/ASTSelection.
That captures a smaller set of AST nodes, has a slightly different way of
representing selections, and generally has mare features and does more work.
The overall shape is pretty similar, and yet I can't quite get to behave as I
expect.

Reviewers: ilya-biryukov, kadircet

Subscribers: mgorny, ioeric, MaskRay, jkorous, mgrang, arphaman

Tags: #clang

Differential Revision: https://reviews.llvm.org/D57562

git-svn-id: https://llvm.org/svn/llvm-project/clang-tools-extra/trunk@352874 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 670 insertions, 0 deletions

diff --git a/clangd/CMakeLists.txt b/clangd/CMakeLists.txt
index 24954f04..6db78920 100644
--- a/clangd/CMakeLists.txt
+++ b/clangd/CMakeLists.txt
@@ -46,6 +46,7 @@ add_clang_library(clangDaemon
   Protocol.cpp
   Quality.cpp
   RIFF.cpp
+  Selection.cpp
   SourceCode.cpp
   Threading.cpp
   Trace.cpp
diff --git a/clangd/Selection.cpp b/clangd/Selection.cpp
new file mode 100644
index 00000000..ada79891
--- /dev/null
+++ b/clangd/Selection.cpp
@@ -0,0 +1,301 @@
+//===--- Selection.h ------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "Selection.h"
+#include "ClangdUnit.h"
+#include "clang/AST/RecursiveASTVisitor.h"
+
+namespace clang {
+namespace clangd {
+namespace {
+using Node = SelectionTree::Node;
+using ast_type_traits::DynTypedNode;
+
+// We find the selection by visiting written nodes in the AST, looking for nodes
+// that intersect with the selected character range.
+//
+// While traversing, we maintain a parent stack. As nodes pop off the stack,
+// we decide whether to keep them or not. To be kept, they must either be
+// selected or contain some nodes that are.
+//
+// For simple cases (not inside macros) we prune subtrees that don't intersect.
+class SelectionVisitor : public RecursiveASTVisitor<SelectionVisitor> {
+public:
+  // Runs the visitor to gather selected nodes and their ancestors.
+  // If there is any selection, the root (TUDecl) is the first node.
+  static std::deque<Node> collect(ASTContext &AST, unsigned Begin,
+                                  unsigned End, FileID File) {
+    SelectionVisitor V(AST, Begin, End, File);
+    V.TraverseAST(AST);
+    assert(V.Stack.size() == 1 && "Unpaired push/pop?");
+    assert(V.Stack.top() == &V.Nodes.front());
+    if (V.Nodes.size() == 1) // TUDecl, but no nodes under it.
+      V.Nodes.clear();
+    return std::move(V.Nodes);
+  }
+
+  // We traverse all "well-behaved" nodes the same way:
+  //  - push the node onto the stack
+  //  - traverse its children recursively
+  //  - pop it from the stack
+  //  - hit testing: is intersection(node, selection) - union(children) empty?
+  //  - attach it to the tree if it or any children hit the selection
+  //
+  // Two categories of nodes are not "well-behaved":
+  //  - those without source range information, we don't record those
+  //  - those that can't be stored in DynTypedNode.
+  // We're missing some interesting things like Attr due to the latter.
+  bool TraverseDecl(Decl *X) {
+    if (isa<TranslationUnitDecl>(X))
+      return Base::TraverseDecl(X); // Already pushed by constructor.
+    return traverseNode(X, [&] { return Base::TraverseDecl(X); });
+  }
+  bool TraverseTypeLoc(TypeLoc X) {
+    return traverseNode(&X, [&] { return Base::TraverseTypeLoc(X); });
+  }
+  bool TraverseTypeNestedNameSpecifierLoc(NestedNameSpecifierLoc X) {
+    return traverseNode(
+        &X, [&] { return Base::TraverseNestedNameSpecifierLoc(X); });
+  }
+  bool TraverseConstructorInitializer(CXXCtorInitializer *X) {
+    return traverseNode(
+        X, [&] { return Base::TraverseConstructorInitializer(X); });
+  }
+  // Stmt is the same, but this form allows the data recursion optimization.
+  bool dataTraverseStmtPre(Stmt *X) {
+    if (!X || canSafelySkipNode(X->getSourceRange()))
+      return false;
+    push(DynTypedNode::create(*X));
+    return true;
+  }
+  bool dataTraverseStmtPost(Stmt *X) {
+    pop();
+    return true;
+  }
+  // Uninteresting parts of the AST that don't have locations within them.
+  bool TraverseNestedNameSpecifier(NestedNameSpecifier *) { return true; }
+  bool TraverseType(QualType) { return true; }
+
+private:
+  using Base = RecursiveASTVisitor<SelectionVisitor>;
+  SelectionVisitor(ASTContext &AST, unsigned SelBegin, unsigned SelEnd,
+                   FileID SelFile)
+      : SM(AST.getSourceManager()), LangOpts(AST.getLangOpts()),
+        SelBegin(SelBegin), SelEnd(SelEnd), SelFile(SelFile),
+        SelBeginTokenStart(SM.getFileOffset(Lexer::GetBeginningOfToken(
+            SM.getComposedLoc(SelFile, SelBegin), SM, LangOpts))) {
+    // Ensure we have a node for the TU decl, regardless of traversal scope.
+    Nodes.emplace_back();
+    Nodes.back().ASTNode = DynTypedNode::create(*AST.getTranslationUnitDecl());
+    Nodes.back().Parent = nullptr;
+    Nodes.back().Selected = SelectionTree::Unselected;
+    Stack.push(&Nodes.back());
+  }
+
+  // Generic case of TraverseFoo. Func should be the call to Base::TraverseFoo.
+  // Node is always a pointer so the generic code can handle any null checks.
+  template <typename T, typename Func>
+  bool traverseNode(T *Node, const Func &Body) {
+    if (Node == nullptr || canSafelySkipNode(Node->getSourceRange()))
+      return true;
+    push(DynTypedNode::create(*Node));
+    bool Ret = Body();
+    pop();
+    return Ret;
+  }
+
+  // An optimization for a common case: nodes outside macro expansions that
+  // don't intersect the selection may be recursively skipped.
+  bool canSafelySkipNode(SourceRange S) {
+    auto B = SM.getDecomposedLoc(S.getBegin());
+    auto E = SM.getDecomposedLoc(S.getEnd());
+    if (B.first != SelFile || E.first != SelFile)
+      return false;
+    return B.second >= SelEnd || E.second < SelBeginTokenStart;
+  }
+
+  // Pushes a node onto the ancestor stack. Pairs with pop().
+  void push(DynTypedNode Node) {
+    Nodes.emplace_back();
+    Nodes.back().ASTNode = std::move(Node);
+    Nodes.back().Parent = Stack.top();
+    Nodes.back().Selected = SelectionTree::Unselected;
+    Stack.push(&Nodes.back());
+  }
+
+  // Pops a node off the ancestor stack, and finalizes it. Pairs with push().
+  void pop() {
+    Node &N = *Stack.top();
+    N.Selected = computeSelection(N);
+    if (N.Selected || !N.Children.empty()) {
+      // Attach to the tree.
+      N.Parent->Children.push_back(&N);
+    } else {
+      // Neither N any children are selected, it doesn't belong in the tree.
+      assert(&N == &Nodes.back());
+      Nodes.pop_back();
+    }
+    Stack.pop();
+  }
+
+  // Perform hit-testing of a complete Node against the selection.
+  // This runs for every node in the AST, and must be fast in common cases.
+  // This is called from pop(), so we can take children into account.
+  SelectionTree::Selection computeSelection(const Node &N) {
+    SourceRange S = N.ASTNode.getSourceRange();
+    if (!S.isValid())
+      return SelectionTree::Unselected;
+    // getTopMacroCallerLoc() allows selection of constructs in macro args. e.g:
+    //   #define LOOP_FOREVER(Body) for(;;) { Body }
+    //   void IncrementLots(int &x) {
+    //     LOOP_FOREVER( ++x; )
+    //   }
+    // Selecting "++x" or "x" will do the right thing.
+    auto B = SM.getDecomposedLoc(SM.getTopMacroCallerLoc(S.getBegin()));
+    auto E = SM.getDecomposedLoc(SM.getTopMacroCallerLoc(S.getEnd()));
+    // Otherwise, nodes in macro expansions can't be selected.
+    if (B.first != SelFile || E.first != SelFile)
+      return SelectionTree::Unselected;
+    // Cheap test: is there any overlap at all between the selection and range?
+    // Note that E.second is the *start* of the last token, which is why we
+    // compare against the "rounded-down" SelBegin.
+    if (B.second >= SelEnd || E.second < SelBeginTokenStart)
+      return SelectionTree::Unselected;
+
+    // We hit something, need some more precise checks.
+    // Adjust [B, E) to be a half-open character range.
+    E.second += Lexer::MeasureTokenLength(S.getEnd(), SM, LangOpts);
+    // This node's own selected text is (this range ^ selection) - child ranges.
+    // If that's empty, then we've only collided with children.
+    if (nodesCoverRange(N.Children, std::max(SelBegin, B.second),
+                        std::min(SelEnd, E.second)))
+      return SelectionTree::Unselected; // Hit children only.
+    // Some of our own characters are covered, this is a true hit.
+    return (B.second >= SelBegin && E.second <= SelEnd)
+               ? SelectionTree::Complete
+               : SelectionTree::Partial;
+  }
+
+  // Is the range [Begin, End) entirely covered by the union of the Nodes?
+  // (The range is a parent node's extent, and the covering nodes are children).
+  bool nodesCoverRange(llvm::ArrayRef<const Node *> Nodes, unsigned Begin,
+                       unsigned End) {
+    if (Begin >= End)
+      return true;
+    if (Nodes.empty())
+      return false;
+
+    // Collect all the expansion ranges, as offsets.
+    SmallVector<std::pair<unsigned, unsigned>, 8> ChildRanges;
+    for (const Node *N : Nodes) {
+      CharSourceRange R = SM.getExpansionRange(N->ASTNode.getSourceRange());
+      auto B = SM.getDecomposedLoc(R.getBegin());
+      auto E = SM.getDecomposedLoc(R.getEnd());
+      if (B.first != SelFile || E.first != SelFile)
+        continue;
+      assert(R.isTokenRange());
+      // Try to cover up to the next token, spaces between children don't count.
+      if (auto Tok = Lexer::findNextToken(R.getEnd(), SM, LangOpts))
+        E.second = SM.getFileOffset(Tok->getLocation());
+      else
+        E.second += Lexer::MeasureTokenLength(R.getEnd(), SM, LangOpts);
+      ChildRanges.push_back({B.second, E.second});
+    }
+    llvm::sort(ChildRanges);
+
+    // Scan through the child ranges, removing as we go.
+    for (const auto R : ChildRanges) {
+      if (R.first > Begin)
+        return false;   // [Begin, R.first) is not covered.
+      Begin = R.second; // Eliminate [R.first, R.second).
+      if (Begin >= End)
+        return true; // Remaining range is empty.
+    }
+    return false; // Went through all children, trailing characters remain.
+  }
+
+  SourceManager &SM;
+  const LangOptions &LangOpts;
+  std::stack<Node *> Stack;
+  std::deque<Node> Nodes; // Stable pointers as we add more nodes.
+  // Half-open selection range.
+  unsigned SelBegin;
+  unsigned SelEnd;
+  FileID SelFile;
+  // If the selection start slices a token in half, the beginning of that token.
+  // This is useful for checking whether the end of a token range overlaps
+  // the selection: range.end < SelBeginTokenStart is equivalent to
+  // range.end + measureToken(range.end) < SelBegin (assuming range.end points
+  // to a token), and it saves a lex every time.
+  unsigned SelBeginTokenStart;
+};
+
+} // namespace
+
+void SelectionTree::print(llvm::raw_ostream &OS, const SelectionTree::Node &N,
+                          int Indent) const {
+  if (N.Selected)
+    OS.indent(Indent - 1) << (N.Selected == SelectionTree::Complete ? '*'
+                                                                    : '.');
+  else
+    OS.indent(Indent);
+  OS << N.ASTNode.getNodeKind().asStringRef() << " ";
+  N.ASTNode.print(OS, PrintPolicy);
+  OS << "\n";
+  for (const Node *Child : N.Children)
+    print(OS, *Child, Indent + 2);
+}
+
+// Decide which selection emulates a "point" query in between characters.
+static std::pair<unsigned, unsigned> pointBounds(unsigned Offset, FileID FID,
+                                                 ASTContext &AST) {
+  StringRef Buf = AST.getSourceManager().getBufferData(FID);
+  // Edge-cases where the choice is forced.
+  if (Buf.size() == 0)
+    return {0, 0};
+  if (Offset == 0)
+    return {0, 1};
+  if (Offset == Buf.size())
+    return {Offset - 1, Offset};
+  // We could choose either this byte or the previous. Usually we prefer the
+  // character on the right of the cursor (or under a block cursor).
+  // But if that's whitespace, we likely want the token on the left.
+  if (isWhitespace(Buf[Offset]) && !isWhitespace(Buf[Offset - 1]))
+    return {Offset - 1, Offset};
+  return {Offset, Offset + 1};
+}
+
+SelectionTree::SelectionTree(ASTContext &AST, unsigned Begin, unsigned End)
+    : PrintPolicy(AST.getLangOpts()) {
+  // No fundamental reason the selection needs to be in the main file,
+  // but that's all clangd has needed so far.
+  FileID FID = AST.getSourceManager().getMainFileID();
+  if (Begin == End)
+    std::tie(Begin, End) = pointBounds(Begin, FID, AST);
+  PrintPolicy.TerseOutput = true;
+
+  Nodes = SelectionVisitor::collect(AST, Begin, End, FID);
+  Root = Nodes.empty() ? nullptr : &Nodes.front();
+}
+
+SelectionTree::SelectionTree(ASTContext &AST, unsigned Offset)
+    : SelectionTree(AST, Offset, Offset) {}
+
+const Node *SelectionTree::commonAncestor() const {
+  if (!Root)
+    return nullptr;
+  for (const Node *Ancestor = Root;; Ancestor = Ancestor->Children.front()) {
+    if (Ancestor->Selected || Ancestor->Children.size() > 1)
+      return Ancestor;
+    // The tree only contains ancestors of the interesting nodes.
+    assert(!Ancestor->Children.empty() && "bad node in selection tree");
+  }
+}
+
+} // namespace clangd
+} // namespace clang
diff --git a/clangd/Selection.h b/clangd/Selection.h
new file mode 100644
index 00000000..c7cee64a
--- /dev/null
+++ b/clangd/Selection.h
@@ -0,0 +1,123 @@
+//===--- Selection.h - What's under the cursor? -------------------*-C++-*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+// Many features are triggered at locations/ranges and operate on AST nodes.
+// (e.g. go-to-definition or code tweaks).
+// At a high level, such features need to work out which node is the correct
+// target.
+//
+// There are a few levels of ambiguity here:
+//
+// Which tokens are included:
+//   int x = one + two;  // what should "go to definition" do?
+//            ^^^^^^
+//
+// Same token means multiple things:
+//   string("foo")       // class string, or a constructor?
+//   ^
+//
+// Which level of the AST is interesting?
+//   if (err) {          // reference to 'err', or operator bool(),
+//       ^               // or the if statement itself?
+//
+// Here we build and expose a data structure that allows features to resolve
+// these ambiguities in an appropriate way:
+//   - we determine which low-level nodes are partly or completely covered
+//     by the selection.
+//   - we expose a tree of the selected nodes and their lexical parents.
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CLANG_TOOLS_EXTRA_CLANGD_SELECTION_H
+#define LLVM_CLANG_TOOLS_EXTRA_CLANGD_SELECTION_H
+#include "clang/AST/ASTTypeTraits.h"
+#include "clang/AST/PrettyPrinter.h"
+#include "llvm/ADT/SmallVector.h"
+
+namespace clang {
+namespace clangd {
+class ParsedAST;
+
+// A selection can partially or completely cover several AST nodes.
+// The SelectionTree contains nodes that are covered, and their parents.
+// SelectionTree does not contain all AST nodes, rather only:
+//   Decl, Stmt, TypeLoc, NestedNamespaceSpecifierLoc, CXXCtorInitializer.
+// (These are the nodes with source ranges that fit in DynTypedNode).
+//
+// Usually commonAncestor() is the place to start:
+//  - it's the simplest answer to "what node is under the cursor"
+//  - the selected Expr (for example) can be found by walking up the parent
+//    chain and checking Node->ASTNode.
+//  - if you want to traverse the selected nodes, they are all under
+//    commonAncestor() in the tree.
+//
+// The SelectionTree owns the Node structures, but the ASTNode attributes
+// point back into the AST it was constructed with.
+class SelectionTree {
+public:
+  // Creates a selection tree at the given byte offset in the main file.
+  // This is approximately equivalent to a range of one character.
+  // (Usually, the character to the right of Offset, sometimes to the left).
+  SelectionTree(ASTContext &AST, unsigned Offset);
+  // Creates a selection tree for the given range in the main file.
+  // The range includes bytes [Start, End).
+  // If Start == End, uses the same heuristics as SelectionTree(AST, Start).
+  SelectionTree(ASTContext &AST, unsigned Start, unsigned End);
+
+  // Describes to what extent an AST node is covered by the selection.
+  enum Selection {
+    // The AST node owns no characters covered by the selection.
+    // Note that characters owned by children don't count:
+    //   if (x == 0) scream();
+    //       ^^^^^^
+    // The IfStmt would be Unselected because all the selected characters are
+    // associated with its children.
+    // (Invisible nodes like ImplicitCastExpr are always unselected).
+    Unselected,
+    // The AST node owns selected characters, but is not completely covered.
+    Partial,
+    // The AST node owns characters, and is covered by the selection.
+    Complete,
+  };
+  // An AST node that is implicated in the selection.
+  // (Either selected directly, or some descendant is selected).
+  struct Node {
+    // The parent within the selection tree. nullptr for TranslationUnitDecl.
+    Node *Parent;
+    // Direct children within the selection tree.
+    llvm::SmallVector<const Node *, 8> Children;
+    // The corresponding node from the full AST.
+    ast_type_traits::DynTypedNode ASTNode;
+    // The extent to which this node is covered by the selection.
+    Selection Selected;
+  };
+
+  // The most specific common ancestor of all the selected nodes.
+  // If there is no selection, this is nullptr.
+  const Node *commonAncestor() const;
+  // The selection node corresponding to TranslationUnitDecl.
+  // If there is no selection, this is nullptr.
+  const Node *root() const { return Root; }
+
+private:
+  std::deque<Node> Nodes; // Stable-pointer storage.
+  const Node *Root;
+  clang::PrintingPolicy PrintPolicy;
+
+  void print(llvm::raw_ostream &OS, const Node &N, int Indent) const;
+  friend llvm::raw_ostream &operator<<(llvm::raw_ostream &OS,
+                                       const SelectionTree &T) {
+    if (auto R = T.root())
+      T.print(OS, *R, 0);
+    else
+      OS << "(empty selection)\n";
+    return OS;
+  }
+};
+
+} // namespace clangd
+} // namespace clang
+#endif
diff --git a/unittests/clangd/CMakeLists.txt b/unittests/clangd/CMakeLists.txt
index aebab1fe..deae9ceb 100644
--- a/unittests/clangd/CMakeLists.txt
+++ b/unittests/clangd/CMakeLists.txt
@@ -34,6 +34,7 @@ add_extra_unittest(ClangdTests
   JSONTransportTests.cpp
   QualityTests.cpp
   RIFFTests.cpp
+  SelectionTests.cpp
   SerializationTests.cpp
   SourceCodeTests.cpp
   SymbolCollectorTests.cpp
diff --git a/unittests/clangd/SelectionTests.cpp b/unittests/clangd/SelectionTests.cpp
new file mode 100644
index 00000000..6cfee5d2
--- /dev/null
+++ b/unittests/clangd/SelectionTests.cpp
@@ -0,0 +1,244 @@
+//===-- RIFFTests.cpp - Binary container unit tests -----------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+#include "Annotations.h"
+#include "Selection.h"
+#include "SourceCode.h"
+#include "TestTU.h"
+#include "gmock/gmock.h"
+#include "gtest/gtest.h"
+
+namespace clang {
+namespace clangd {
+namespace {
+using ::testing::UnorderedElementsAreArray;
+
+SelectionTree makeSelectionTree(const StringRef MarkedCode, ParsedAST &AST) {
+  Annotations Test(MarkedCode);
+  switch (Test.points().size()) {
+  case 1: // Point selection.
+    return SelectionTree(AST.getASTContext(),
+                         cantFail(positionToOffset(Test.code(), Test.point())));
+  case 2: // Range selection.
+    return SelectionTree(
+        AST.getASTContext(),
+        cantFail(positionToOffset(Test.code(), Test.points()[0])),
+        cantFail(positionToOffset(Test.code(), Test.points()[1])));
+  default:
+    ADD_FAILURE() << "Expected 1-2 points for selection.\n" << MarkedCode;
+    return SelectionTree(AST.getASTContext(), 0u, 0u);
+  }
+}
+
+Range nodeRange(const SelectionTree::Node *N, ParsedAST &AST) {
+  if (!N)
+    return Range{};
+  SourceManager &SM = AST.getASTContext().getSourceManager();
+  StringRef Buffer = SM.getBufferData(SM.getMainFileID());
+  SourceRange SR = N->ASTNode.getSourceRange();
+  SR.setBegin(SM.getFileLoc(SR.getBegin()));
+  SR.setEnd(SM.getFileLoc(SR.getEnd()));
+  CharSourceRange R =
+      Lexer::getAsCharRange(SR, SM, AST.getASTContext().getLangOpts());
+  return Range{offsetToPosition(Buffer, SM.getFileOffset(R.getBegin())),
+               offsetToPosition(Buffer, SM.getFileOffset(R.getEnd()) + 1)};
+}
+
+std::string nodeKind(const SelectionTree::Node *N) {
+  if (!N)
+    return "<null>";
+  return N->ASTNode.getNodeKind().asStringRef().str();
+}
+
+std::vector<const SelectionTree::Node *> allNodes(const SelectionTree &T) {
+  std::vector<const SelectionTree::Node *> Result = {T.root()};
+  for (unsigned I = 0; I < Result.size(); ++I) {
+    const SelectionTree::Node *N = Result[I];
+    Result.insert(Result.end(), N->Children.begin(), N->Children.end());
+  }
+  return Result;
+}
+
+// Returns true if Common is a descendent of Root.
+// Verifies nothing is selected above Common.
+bool verifyCommonAncestor(const SelectionTree::Node *Root,
+                          const SelectionTree::Node *Common,
+                          StringRef MarkedCode) {
+  if (Root == Common)
+    return true;
+  if (Root->Selected)
+    ADD_FAILURE() << "Selected nodes outside common ancestor\n" << MarkedCode;
+  bool Seen = false;
+  for (const SelectionTree::Node *Child : Root->Children)
+    if (verifyCommonAncestor(Child, Common, MarkedCode)) {
+      if (Seen)
+        ADD_FAILURE() << "Saw common ancestor twice\n" << MarkedCode;
+      Seen = true;
+    }
+  return Seen;
+}
+
+TEST(SelectionTest, CommonAncestor) {
+  struct Case {
+    // Selection is between ^marks^.
+    // common ancestor marked with a [[range]].
+    const char *Code;
+    const char *CommonAncestorKind;
+  };
+  Case Cases[] = {
+      {
+          R"cpp(
+            struct AAA { struct BBB { static int ccc(); };};
+            int x = AAA::[[B^B^B]]::ccc();
+          )cpp",
+          "TypeLoc",
+      },
+      {
+          R"cpp(
+            struct AAA { struct BBB { static int ccc(); };};
+            int x = AAA::[[B^BB^]]::ccc();
+          )cpp",
+          "TypeLoc",
+      },
+      {
+          R"cpp(
+            struct AAA { struct BBB { static int ccc(); };};
+            int x = [[AAA::BBB::c^c^c]]();
+          )cpp",
+          "DeclRefExpr",
+      },
+      {
+          R"cpp(
+            struct AAA { struct BBB { static int ccc(); };};
+            int x = [[AAA::BBB::cc^c(^)]];
+          )cpp",
+          "CallExpr",
+      },
+
+      {
+          R"cpp(
+            void foo() { [[if (1^11) { return; } else {^ }]] }
+          )cpp",
+          "IfStmt",
+      },
+      {
+          R"cpp(
+            void foo();
+            #define CALL_FUNCTION(X) X()
+            void bar() { CALL_FUNCTION([[f^o^o]]); }
+          )cpp",
+          "DeclRefExpr",
+      },
+      {
+          R"cpp(
+            void foo();
+            #define CALL_FUNCTION(X) X()
+            void bar() { CALL_FUNC^TION([[fo^o]]); }
+          )cpp",
+          "DeclRefExpr",
+      },
+      {
+          R"cpp(
+            void foo();
+            #define CALL_FUNCTION(X) X()
+            void bar() [[{ C^ALL_FUNC^TION(foo); }]]
+          )cpp",
+          "CompoundStmt",
+      },
+      {
+          R"cpp(
+            void foo();
+            #define CALL_FUNCTION(X) X^()^
+            void bar() { CALL_FUNCTION(foo); }
+          )cpp",
+          nullptr,
+      },
+
+      // Point selections.
+      {"void foo() { [[^foo]](); }", "DeclRefExpr"},
+      {"void foo() { [[f^oo]](); }", "DeclRefExpr"},
+      {"void foo() { [[fo^o]](); }", "DeclRefExpr"},
+      {"void foo() { [[foo^()]]; }", "CallExpr"},
+      {"void foo() { [[foo^]] (); }", "DeclRefExpr"},
+      {"int bar; void foo() [[{ foo (); }]]^", "CompoundStmt"},
+      {"[[^void]] foo();", "TypeLoc"},
+      {"^", nullptr},
+      {"void foo() { [[foo^^]] (); }", "DeclRefExpr"},
+
+      // FIXME: Ideally we'd get a declstmt or the VarDecl itself here.
+      // This doesn't happen now; the RAV doesn't traverse a node containing ;.
+      {"int x = 42;^", nullptr},
+      {"int x = 42^;", nullptr},
+
+      // Node types that have caused problems in the past.
+      {"template <typename T> void foo() { [[^T]] t; }", "TypeLoc"},
+  };
+  for (const Case &C : Cases) {
+    Annotations Test(C.Code);
+    auto AST = TestTU::withCode(Test.code()).build();
+    auto T = makeSelectionTree(C.Code, AST);
+
+    if (Test.ranges().empty()) {
+      // If no [[range]] is marked in the example, there should be no selection.
+      EXPECT_FALSE(T.commonAncestor()) << C.Code << "\n" << T;
+      EXPECT_FALSE(T.root()) << C.Code << "\n" << T;
+    } else {
+      // If there is an expected selection, both common ancestor and root
+      // should exist with the appropriate node types in them.
+      EXPECT_EQ(C.CommonAncestorKind, nodeKind(T.commonAncestor()))
+          << C.Code << "\n"
+          << T;
+      EXPECT_EQ("TranslationUnitDecl", nodeKind(T.root())) << C.Code;
+      // Convert the reported common ancestor to a range and verify it.
+      EXPECT_EQ(nodeRange(T.commonAncestor(), AST), Test.range())
+          << C.Code << "\n"
+          << T;
+
+      // Check that common ancestor is reachable on exactly one path from root,
+      // and no nodes outside it are selected.
+      EXPECT_TRUE(verifyCommonAncestor(T.root(), T.commonAncestor(), C.Code))
+          << C.Code;
+    }
+  }
+}
+
+TEST(SelectionTest, Selected) {
+  // Selection with ^marks^.
+  // Partially selected nodes marked with a [[range]].
+  // Completely selected nodes marked with a $C[[range]].
+  const char *Cases[] = {
+      R"cpp( int abc, xyz = [[^ab^c]]; )cpp",
+      R"cpp( int abc, xyz = [[a^bc^]]; )cpp",
+      R"cpp( int abc, xyz = $C[[^abc^]]; )cpp",
+      R"cpp(
+        void foo() {
+          [[if ([[1^11]]) $C[[{
+            $C[[return]];
+          }]] else [[{^
+          }]]]]
+        }
+      )cpp",
+  };
+  for (const char *C : Cases) {
+    Annotations Test(C);
+    auto AST = TestTU::withCode(Test.code()).build();
+    auto T = makeSelectionTree(C, AST);
+
+    std::vector<Range> Complete, Partial;
+    for (const SelectionTree::Node *N : allNodes(T))
+      if (N->Selected == SelectionTree::Complete)
+        Complete.push_back(nodeRange(N, AST));
+      else if (N->Selected == SelectionTree::Partial)
+        Partial.push_back(nodeRange(N, AST));
+    EXPECT_THAT(Complete, UnorderedElementsAreArray(Test.ranges("C"))) << C;
+    EXPECT_THAT(Partial, UnorderedElementsAreArray(Test.ranges())) << C;
+  }
+}
+
+} // namespace
+} // namespace clangd
+} // namespace clang