[clang-format] Add Options to break inside the TableGen DAGArg. (llvm#83149)

hnakamura5 · web-flow · commit 0c423af59c97 · 2024-03-19T00:13:59.000+09:00
Add two options to control the line break inside TableGen DAGArg.
- TableGenBreakInsideDAGArg
- TableGenBreakingDAGArgOperators
diff --git a/clang/docs/ClangFormatStyleOptions.rst b/clang/docs/ClangFormatStyleOptions.rst
@@ -6158,6 +6158,70 @@ the configuration (without a prefix: ``Auto``).
 **TabWidth** (``Unsigned``) :versionbadge:`clang-format 3.7` :ref:`¶ <TabWidth>`
   The number of columns used for tab stops.
 
+.. _TableGenBreakInsideDAGArg:
+
+**TableGenBreakInsideDAGArg** (``DAGArgStyle``) :versionbadge:`clang-format 19` :ref:`¶ <TableGenBreakInsideDAGArg>`
+  The styles of the line break inside the DAGArg in TableGen.
+
+  Possible values:
+
+  * ``DAS_DontBreak`` (in configuration: ``DontBreak``)
+    Never break inside DAGArg.
+
+    .. code-block:: c++
+
+      let DAGArgIns = (ins i32:$src1, i32:$src2);
+
+  * ``DAS_BreakElements`` (in configuration: ``BreakElements``)
+    Break inside DAGArg after each list element but for the last.
+    This aligns to the first element.
+
+    .. code-block:: c++
+
+      let DAGArgIns = (ins i32:$src1,
+                           i32:$src2);
+
+  * ``DAS_BreakAll`` (in configuration: ``BreakAll``)
+    Break inside DAGArg after the operator and the all elements.
+
+    .. code-block:: c++
+
+      let DAGArgIns = (ins
+          i32:$src1,
+          i32:$src2
+      );
+
+
+
+.. _TableGenBreakingDAGArgOperators:
+
+**TableGenBreakingDAGArgOperators** (``List of Strings``) :versionbadge:`clang-format 19` :ref:`¶ <TableGenBreakingDAGArgOperators>`
+  Works only when TableGenBreakInsideDAGArg is not DontBreak.
+  The string list needs to consist of identifiers in TableGen.
+  If any identifier is specified, this limits the line breaks by
+  TableGenBreakInsideDAGArg option only on DAGArg values beginning with
+  the specified identifiers.
+
+  For example the configuration,
+
+  .. code-block:: c++
+
+    TableGenBreakInsideDAGArg: BreakAll
+    TableGenBreakingDAGArgOperators: ['ins', 'outs']
+
+  makes the line break only occurs inside DAGArgs beginning with the
+  specified identifiers 'ins' and 'outs'.
+
+
+  .. code-block:: c++
+
+    let DAGArgIns = (ins
+        i32:$src1,
+        i32:$src2
+    );
+    let DAGArgOtherID = (other i32:$other1, i32:$other2);
+    let DAGArgBang = (!cast<SomeType>("Some") i32:$src1, i32:$src2)
+
 .. _TypeNames:
 
 **TypeNames** (``List of Strings``) :versionbadge:`clang-format 17` :ref:`¶ <TypeNames>`
diff --git a/clang/include/clang/Format/Format.h b/clang/include/clang/Format/Format.h
@@ -4728,6 +4728,60 @@ struct FormatStyle {
   /// \version 8
   std::vector<std::string> StatementMacros;
 
+  /// Works only when TableGenBreakInsideDAGArg is not DontBreak.
+  /// The string list needs to consist of identifiers in TableGen.
+  /// If any identifier is specified, this limits the line breaks by
+  /// TableGenBreakInsideDAGArg option only on DAGArg values beginning with
+  /// the specified identifiers.
+  ///
+  /// For example the configuration,
+  /// \code
+  ///   TableGenBreakInsideDAGArg: BreakAll
+  ///   TableGenBreakingDAGArgOperators: ['ins', 'outs']
+  /// \endcode
+  ///
+  /// makes the line break only occurs inside DAGArgs beginning with the
+  /// specified identifiers 'ins' and 'outs'.
+  ///
+  /// \code
+  ///   let DAGArgIns = (ins
+  ///       i32:$src1,
+  ///       i32:$src2
+  ///   );
+  ///   let DAGArgOtherID = (other i32:$other1, i32:$other2);
+  ///   let DAGArgBang = (!cast<SomeType>("Some") i32:$src1, i32:$src2)
+  /// \endcode
+  /// \version 19
+  std::vector<std::string> TableGenBreakingDAGArgOperators;
+
+  /// Different ways to control the format inside TableGen DAGArg.
+  enum DAGArgStyle : int8_t {
+    /// Never break inside DAGArg.
+    /// \code
+    ///   let DAGArgIns = (ins i32:$src1, i32:$src2);
+    /// \endcode
+    DAS_DontBreak,
+    /// Break inside DAGArg after each list element but for the last.
+    /// This aligns to the first element.
+    /// \code
+    ///   let DAGArgIns = (ins i32:$src1,
+    ///                        i32:$src2);
+    /// \endcode
+    DAS_BreakElements,
+    /// Break inside DAGArg after the operator and the all elements.
+    /// \code
+    ///   let DAGArgIns = (ins
+    ///       i32:$src1,
+    ///       i32:$src2
+    ///   );
+    /// \endcode
+    DAS_BreakAll,
+  };
+
+  /// The styles of the line break inside the DAGArg in TableGen.
+  /// \version 19
+  DAGArgStyle TableGenBreakInsideDAGArg;
+
   /// The number of columns used for tab stops.
   /// \version 3.7
   unsigned TabWidth;
@@ -4980,9 +5034,12 @@ struct FormatStyle {
            SpacesInSquareBrackets == R.SpacesInSquareBrackets &&
            Standard == R.Standard &&
            StatementAttributeLikeMacros == R.StatementAttributeLikeMacros &&
-           StatementMacros == R.StatementMacros && TabWidth == R.TabWidth &&
-           TypeNames == R.TypeNames && TypenameMacros == R.TypenameMacros &&
-           UseTab == R.UseTab &&
+           StatementMacros == R.StatementMacros &&
+           TableGenBreakingDAGArgOperators ==
+               R.TableGenBreakingDAGArgOperators &&
+           TableGenBreakInsideDAGArg == R.TableGenBreakInsideDAGArg &&
+           TabWidth == R.TabWidth && TypeNames == R.TypeNames &&
+           TypenameMacros == R.TypenameMacros && UseTab == R.UseTab &&
            VerilogBreakBetweenInstancePorts ==
                R.VerilogBreakBetweenInstancePorts &&
            WhitespaceSensitiveMacros == R.WhitespaceSensitiveMacros;
diff --git a/clang/lib/Format/ContinuationIndenter.cpp b/clang/lib/Format/ContinuationIndenter.cpp
@@ -815,6 +815,7 @@ void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
       !CurrentState.IsCSharpGenericTypeConstraint && Previous.opensScope() &&
       Previous.isNot(TT_ObjCMethodExpr) && Previous.isNot(TT_RequiresClause) &&
       Previous.isNot(TT_TableGenDAGArgOpener) &&
+      Previous.isNot(TT_TableGenDAGArgOpenerToBreak) &&
       !(Current.MacroParent && Previous.MacroParent) &&
       (Current.isNot(TT_LineComment) ||
        Previous.isOneOf(BK_BracedInit, TT_VerilogMultiLineListLParen))) {
@@ -1437,7 +1438,9 @@ unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) {
       Style.BreakInheritanceList == FormatStyle::BILS_AfterColon) {
     return CurrentState.Indent;
   }
-  if (Previous.is(tok::r_paren) && !Current.isBinaryOperator() &&
+  if (Previous.is(tok::r_paren) &&
+      Previous.isNot(TT_TableGenDAGArgOperatorToBreak) &&
+      !Current.isBinaryOperator() &&
       !Current.isOneOf(tok::colon, tok::comment)) {
     return ContinuationIndent;
   }
@@ -1698,7 +1701,8 @@ void ContinuationIndenter::moveStatePastFakeLParens(LineState &State,
         (Style.AlignAfterOpenBracket != FormatStyle::BAS_DontAlign ||
          PrecedenceLevel != prec::Comma || Current.NestingLevel == 0) &&
         (!Style.isTableGen() ||
-         (Previous && Previous->is(TT_TableGenDAGArgListComma)))) {
+         (Previous && Previous->isOneOf(TT_TableGenDAGArgListComma,
+                                        TT_TableGenDAGArgListCommaToBreak)))) {
       NewParenState.Indent = std::max(
           std::max(State.Column, NewParenState.Indent), CurrentState.LastSpace);
     }
@@ -1835,6 +1839,17 @@ void ContinuationIndenter::moveStatePastScopeOpener(LineState &State,
         Style.ContinuationIndentWidth +
         std::max(CurrentState.LastSpace, CurrentState.StartOfFunctionCall);
 
+    if (Style.isTableGen() && Current.is(TT_TableGenDAGArgOpenerToBreak) &&
+        Style.TableGenBreakInsideDAGArg == FormatStyle::DAS_BreakElements) {
+      // For the case the next token is a TableGen DAGArg operator identifier
+      // that is not marked to have a line break after it.
+      // In this case the option DAS_BreakElements requires to align the
+      // DAGArg elements to the operator.
+      const FormatToken *Next = Current.Next;
+      if (Next && Next->is(TT_TableGenDAGArgOperatorID))
+        NewIndent = State.Column + Next->TokenText.size() + 2;
+    }
+
     // Ensure that different different brackets force relative alignment, e.g.:
     // void SomeFunction(vector<  // break
     //                       int> v);
diff --git a/clang/lib/Format/Format.cpp b/clang/lib/Format/Format.cpp
@@ -279,6 +279,14 @@ struct ScalarEnumerationTraits<FormatStyle::BreakTemplateDeclarationsStyle> {
   }
 };
 
+template <> struct ScalarEnumerationTraits<FormatStyle::DAGArgStyle> {
+  static void enumeration(IO &IO, FormatStyle::DAGArgStyle &Value) {
+    IO.enumCase(Value, "DontBreak", FormatStyle::DAS_DontBreak);
+    IO.enumCase(Value, "BreakElements", FormatStyle::DAS_BreakElements);
+    IO.enumCase(Value, "BreakAll", FormatStyle::DAS_BreakAll);
+  }
+};
+
 template <>
 struct ScalarEnumerationTraits<FormatStyle::DefinitionReturnTypeBreakingStyle> {
   static void
@@ -1092,6 +1100,10 @@ template <> struct MappingTraits<FormatStyle> {
     IO.mapOptional("StatementAttributeLikeMacros",
                    Style.StatementAttributeLikeMacros);
     IO.mapOptional("StatementMacros", Style.StatementMacros);
+    IO.mapOptional("TableGenBreakingDAGArgOperators",
+                   Style.TableGenBreakingDAGArgOperators);
+    IO.mapOptional("TableGenBreakInsideDAGArg",
+                   Style.TableGenBreakInsideDAGArg);
     IO.mapOptional("TabWidth", Style.TabWidth);
     IO.mapOptional("TypeNames", Style.TypeNames);
     IO.mapOptional("TypenameMacros", Style.TypenameMacros);
@@ -1552,6 +1564,8 @@ FormatStyle getLLVMStyle(FormatStyle::LanguageKind Language) {
   LLVMStyle.StatementAttributeLikeMacros.push_back("Q_EMIT");
   LLVMStyle.StatementMacros.push_back("Q_UNUSED");
   LLVMStyle.StatementMacros.push_back("QT_REQUIRE_VERSION");
+  LLVMStyle.TableGenBreakingDAGArgOperators = {};
+  LLVMStyle.TableGenBreakInsideDAGArg = FormatStyle::DAS_DontBreak;
   LLVMStyle.TabWidth = 8;
   LLVMStyle.UseTab = FormatStyle::UT_Never;
   LLVMStyle.VerilogBreakBetweenInstancePorts = true;
diff --git a/clang/lib/Format/FormatToken.h b/clang/lib/Format/FormatToken.h
@@ -156,7 +156,11 @@ extern bool IsCpp;
   TYPE(TableGenDAGArgCloser)                                                   \
   TYPE(TableGenDAGArgListColon)                                                \
   TYPE(TableGenDAGArgListComma)                                                \
+  TYPE(TableGenDAGArgListCommaToBreak)                                         \
   TYPE(TableGenDAGArgOpener)                                                   \
+  TYPE(TableGenDAGArgOpenerToBreak)                                            \
+  TYPE(TableGenDAGArgOperatorID)                                               \
+  TYPE(TableGenDAGArgOperatorToBreak)                                          \
   TYPE(TableGenListCloser)                                                     \
   TYPE(TableGenListOpener)                                                     \
   TYPE(TableGenMultiLineString)                                                \
diff --git a/clang/lib/Format/TokenAnnotator.cpp b/clang/lib/Format/TokenAnnotator.cpp
@@ -986,16 +986,59 @@ class AnnotatingParser {
     return false;
   }
 
+  // Judge if the token is a operator ID to insert line break in DAGArg.
+  // That is, TableGenBreakingDAGArgOperators is empty (by the definition of the
+  // option) or the token is in the list.
+  bool isTableGenDAGArgBreakingOperator(const FormatToken &Tok) {
+    auto &Opes = Style.TableGenBreakingDAGArgOperators;
+    // If the list is empty, all operators are breaking operators.
+    if (Opes.empty())
+      return true;
+    // Otherwise, the operator is limited to normal identifiers.
+    if (Tok.isNot(tok::identifier) ||
+        Tok.isOneOf(TT_TableGenBangOperator, TT_TableGenCondOperator)) {
+      return false;
+    }
+    // The case next is colon, it is not a operator of identifier.
+    if (!Tok.Next || Tok.Next->is(tok::colon))
+      return false;
+    return std::find(Opes.begin(), Opes.end(), Tok.TokenText.str()) !=
+           Opes.end();
+  }
+
   // SimpleValue6 ::=  "(" DagArg [DagArgList] ")"
   // This parses SimpleValue 6's inside part of "(" ")"
   bool parseTableGenDAGArgAndList(FormatToken *Opener) {
+    FormatToken *FirstTok = CurrentToken;
     if (!parseTableGenDAGArg())
       return false;
+    bool BreakInside = false;
+    if (Style.TableGenBreakInsideDAGArg != FormatStyle::DAS_DontBreak) {
+      // Specialized detection for DAGArgOperator, that determines the way of
+      // line break for this DAGArg elements.
+      if (isTableGenDAGArgBreakingOperator(*FirstTok)) {
+        // Special case for identifier DAGArg operator.
+        BreakInside = true;
+        Opener->setType(TT_TableGenDAGArgOpenerToBreak);
+        if (FirstTok->isOneOf(TT_TableGenBangOperator,
+                              TT_TableGenCondOperator)) {
+          // Special case for bang/cond operators. Set the whole operator as
+          // the DAGArg operator. Always break after it.
+          CurrentToken->Previous->setType(TT_TableGenDAGArgOperatorToBreak);
+        } else if (FirstTok->is(tok::identifier)) {
+          if (Style.TableGenBreakInsideDAGArg == FormatStyle::DAS_BreakAll)
+            FirstTok->setType(TT_TableGenDAGArgOperatorToBreak);
+          else
+            FirstTok->setType(TT_TableGenDAGArgOperatorID);
+        }
+      }
+    }
     // Parse the [DagArgList] part
     bool FirstDAGArgListElm = true;
     while (CurrentToken) {
       if (!FirstDAGArgListElm && CurrentToken->is(tok::comma)) {
-        CurrentToken->setType(TT_TableGenDAGArgListComma);
+        CurrentToken->setType(BreakInside ? TT_TableGenDAGArgListCommaToBreak
+                                          : TT_TableGenDAGArgListComma);
         skipToNextNonComment();
       }
       if (CurrentToken && CurrentToken->is(tok::r_paren)) {
@@ -5086,6 +5129,11 @@ bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line,
     }
     if (Right.is(TT_TableGenCondOperatorColon))
       return false;
+    if (Left.isOneOf(TT_TableGenDAGArgOperatorID,
+                     TT_TableGenDAGArgOperatorToBreak) &&
+        Right.isNot(TT_TableGenDAGArgCloser)) {
+      return true;
+    }
     // Do not insert bang operators and consequent openers.
     if (Right.isOneOf(tok::l_paren, tok::less) &&
         Left.isOneOf(TT_TableGenBangOperator, TT_TableGenCondOperator)) {
@@ -5462,6 +5510,18 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
     //       case2:0);
     if (Left.is(TT_TableGenCondOperatorComma))
       return true;
+    if (Left.is(TT_TableGenDAGArgOperatorToBreak) &&
+        Right.isNot(TT_TableGenDAGArgCloser)) {
+      return true;
+    }
+    if (Left.is(TT_TableGenDAGArgListCommaToBreak))
+      return true;
+    if (Right.is(TT_TableGenDAGArgCloser) && Right.MatchingParen &&
+        Right.MatchingParen->is(TT_TableGenDAGArgOpenerToBreak) &&
+        &Left != Right.MatchingParen->Next) {
+      // Check to avoid empty DAGArg such as (ins).
+      return Style.TableGenBreakInsideDAGArg == FormatStyle::DAS_BreakAll;
+    }
   }
 
   if (Line.startsWith(tok::kw_asm) && Right.is(TT_InlineASMColon) &&
@@ -5876,6 +5936,8 @@ bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
     // Avoid to break around paste operator.
     if (Left.is(tok::hash) || Right.is(tok::hash))
       return false;
+    if (Left.isOneOf(TT_TableGenBangOperator, TT_TableGenCondOperator))
+      return false;
   }
 
   if (Left.is(tok::at))
diff --git a/clang/unittests/Format/FormatTestTableGen.cpp b/clang/unittests/Format/FormatTestTableGen.cpp
diff --git a/clang/unittests/Format/TokenAnnotatorTest.cpp b/clang/unittests/Format/TokenAnnotatorTest.cpp
