[clang-format] Add VariableTemplates option (#121318)

Closes #120148.

Author: owenca

clang/docs/ClangFormatStyleOptions.rst

@@ -6798,6 +6798,15 @@ the configuration (without a prefix: ``Auto``).
 
 
 
+.. _VariableTemplates:
+
+**VariableTemplates** (``List of Strings``) :versionbadge:`clang-format 20` :ref:`¶ <VariableTemplates>`
+  A vector of non-keyword identifiers that should be interpreted as variable
+  template names.
+
+  A ``)`` after a variable template instantiation is **not** annotated as
+  the closing parenthesis of C-style cast operator.
+
 .. _VerilogBreakBetweenInstancePorts:
 
 **VerilogBreakBetweenInstancePorts** (``Boolean``) :versionbadge:`clang-format 17` :ref:`¶ <VerilogBreakBetweenInstancePorts>`

clang/docs/ReleaseNotes.rst

@@ -1125,6 +1125,7 @@ clang-format
 - Adds ``RemoveEmptyLinesInUnwrappedLines`` option.
 - Adds ``KeepFormFeed`` option and set it to ``true`` for ``GNU`` style.
 - Adds ``AllowShortNamespacesOnASingleLine`` option.
+- Adds ``VariableTemplates`` option.
 - Adds support for bash globstar in ``.clang-format-ignore``.
 
 libclang

clang/include/clang/Format/Format.h

@@ -5103,6 +5103,15 @@ struct FormatStyle {
   /// \version 3.7
   UseTabStyle UseTab;
 
+  /// A vector of non-keyword identifiers that should be interpreted as variable
+  /// template names.
+  ///
+  /// A ``)`` after a variable template instantiation is **not** annotated as
+  /// the closing parenthesis of C-style cast operator.
+  ///
+  /// \version 20
+  std::vector<std::string> VariableTemplates;
+
   /// For Verilog, put each port on its own line in module instantiations.
   /// \code
   ///    true:
@@ -5314,7 +5323,7 @@ struct FormatStyle {
            TableGenBreakInsideDAGArg == R.TableGenBreakInsideDAGArg &&
            TabWidth == R.TabWidth && TemplateNames == R.TemplateNames &&
            TypeNames == R.TypeNames && TypenameMacros == R.TypenameMacros &&
-           UseTab == R.UseTab &&
+           UseTab == R.UseTab && VariableTemplates == R.VariableTemplates &&
            VerilogBreakBetweenInstancePorts ==
                R.VerilogBreakBetweenInstancePorts &&
            WhitespaceSensitiveMacros == R.WhitespaceSensitiveMacros;

clang/lib/Format/Format.cpp

@@ -1166,6 +1166,7 @@ template <> struct MappingTraits<FormatStyle> {
     IO.mapOptional("TypeNames", Style.TypeNames);
     IO.mapOptional("TypenameMacros", Style.TypenameMacros);
     IO.mapOptional("UseTab", Style.UseTab);
+    IO.mapOptional("VariableTemplates", Style.VariableTemplates);
     IO.mapOptional("VerilogBreakBetweenInstancePorts",
                    Style.VerilogBreakBetweenInstancePorts);
     IO.mapOptional("WhitespaceSensitiveMacros",

clang/lib/Format/FormatToken.h

@@ -186,6 +186,7 @@ namespace format {
   TYPE(UnionLBrace)                                                            \
   TYPE(UnionRBrace)                                                            \
   TYPE(UntouchableMacroFunc)                                                   \
+  TYPE(VariableTemplate)                                                       \
   /* Like in 'assign x = 0, y = 1;' . */                                       \
   TYPE(VerilogAssignComma)                                                     \
   /* like in begin : block */                                                  \

clang/lib/Format/FormatTokenLexer.cpp

@@ -76,6 +76,8 @@ FormatTokenLexer::FormatTokenLexer(
     TemplateNames.insert(&IdentTable.get(TemplateName));
   for (const auto &TypeName : Style.TypeNames)
     TypeNames.insert(&IdentTable.get(TypeName));
+  for (const auto &VariableTemplate : Style.VariableTemplates)
+    VariableTemplates.insert(&IdentTable.get(VariableTemplate));
 }
 
 ArrayRef<FormatToken *> FormatTokenLexer::lex() {
@@ -1382,6 +1384,8 @@ FormatToken *FormatTokenLexer::getNextToken() {
         FormatTok->setFinalizedType(TT_TemplateName);
       else if (TypeNames.contains(Identifier))
         FormatTok->setFinalizedType(TT_TypeName);
+      else if (VariableTemplates.contains(Identifier))
+        FormatTok->setFinalizedType(TT_VariableTemplate);
     }
   }
 

clang/lib/Format/FormatTokenLexer.h

@@ -129,7 +129,8 @@ class FormatTokenLexer {
 
   llvm::SmallMapVector<IdentifierInfo *, TokenType, 8> Macros;
 
-  llvm::SmallPtrSet<IdentifierInfo *, 8> TemplateNames, TypeNames;
+  llvm::SmallPtrSet<IdentifierInfo *, 8> TemplateNames, TypeNames,
+      VariableTemplates;
 
   bool FormattingDisabled;
 

clang/lib/Format/TokenAnnotator.cpp

@@ -2792,6 +2792,16 @@ class AnnotatingParser {
       return true;
     }
 
+    auto IsNonVariableTemplate = [](const FormatToken &Tok) {
+      if (Tok.isNot(TT_TemplateCloser))
+        return false;
+      const auto *Less = Tok.MatchingParen;
+      if (!Less)
+        return false;
+      const auto *BeforeLess = Less->getPreviousNonComment();
+      return BeforeLess && BeforeLess->isNot(TT_VariableTemplate);
+    };
+
     // Heuristically try to determine whether the parentheses contain a type.
     auto IsQualifiedPointerOrReference = [](const FormatToken *T,
                                             const LangOptions &LangOpts) {
@@ -2825,10 +2835,11 @@ class AnnotatingParser {
       }
       return T && T->is(TT_PointerOrReference);
     };
-    bool ParensAreType =
-        BeforeRParen->isOneOf(TT_TemplateCloser, TT_TypeDeclarationParen) ||
-        BeforeRParen->isTypeName(LangOpts) ||
-        IsQualifiedPointerOrReference(BeforeRParen, LangOpts);
+
+    bool ParensAreType = IsNonVariableTemplate(*BeforeRParen) ||
+                         BeforeRParen->is(TT_TypeDeclarationParen) ||
+                         BeforeRParen->isTypeName(LangOpts) ||
+                         IsQualifiedPointerOrReference(BeforeRParen, LangOpts);
     bool ParensCouldEndDecl =
         AfterRParen->isOneOf(tok::equal, tok::semi, tok::l_brace, tok::greater);
     if (ParensAreType && !ParensCouldEndDecl)

clang/unittests/Format/TokenAnnotatorTest.cpp

@@ -3615,6 +3615,19 @@ TEST_F(TokenAnnotatorTest, TemplateInstantiation) {
   EXPECT_TOKEN(Tokens[18], tok::greater, TT_TemplateCloser);
 }
 
+TEST_F(TokenAnnotatorTest, VariableTemplate) {
+  auto Style = getLLVMStyle();
+  Style.VariableTemplates.push_back("a");
+
+  auto Tokens = annotate("auto t3 = (a<int>) + b;", Style);
+  ASSERT_EQ(Tokens.size(), 13u) << Tokens;
+  EXPECT_TOKEN(Tokens[4], tok::identifier, TT_VariableTemplate);
+  EXPECT_TOKEN(Tokens[5], tok::less, TT_TemplateOpener);
+  EXPECT_TOKEN(Tokens[7], tok::greater, TT_TemplateCloser);
+  EXPECT_TOKEN(Tokens[8], tok::r_paren, TT_Unknown); // Not TT_CastRParen
+  EXPECT_TOKEN(Tokens[9], tok::plus, TT_BinaryOperator);
+}
+
 TEST_F(TokenAnnotatorTest, SwitchInMacroArgument) {
   auto Tokens = annotate("FOOBAR(switch);\n"
                          "void f() {}");