[Clang] Add support for the C _Defer TS (#162848)

This implements WG14 N3734 (https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3734.pdf),
aka `_Defer`; it is currently only supported in C if `-fdefer-ts` is passed.

Author: Sirraide

clang-tools-extra/clang-tidy/bugprone/BranchCloneCheck.cpp

@@ -237,6 +237,12 @@ static bool isIdenticalStmt(const ASTContext &Ctx, const Stmt *Stmt1,
       return false;
     return true;
   }
+  case Stmt::DeferStmtClass: {
+    const auto *DefStmt1 = cast<DeferStmt>(Stmt1);
+    const auto *DefStmt2 = cast<DeferStmt>(Stmt2);
+    return isIdenticalStmt(Ctx, DefStmt1->getBody(), DefStmt2->getBody(),
+                           IgnoreSideEffects);
+  }
   case Stmt::CompoundStmtClass: {
     const auto *CompStmt1 = cast<CompoundStmt>(Stmt1);
     const auto *CompStmt2 = cast<CompoundStmt>(Stmt2);

clang/docs/ReleaseNotes.rst

@@ -208,6 +208,11 @@ Resolutions to C++ Defect Reports
 C Language Changes
 ------------------
 
+- Implemented the ``defer`` draft Technical Specification
+  (`WG14 N3734 <https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3734.pdf>`_); it is enabled in C mode by
+  passing ``-fdefer-ts``. Note, the details of this feature are subject to change given that the Technical
+  Specification is not yet ratified.
+
 C2y Feature Support
 ^^^^^^^^^^^^^^^^^^^
 - No longer triggering ``-Wstatic-in-inline`` in C2y mode; use of a static

clang/include/clang/AST/RecursiveASTVisitor.h

@@ -2561,6 +2561,7 @@ DEF_TRAVERSE_STMT(DefaultStmt, {})
 DEF_TRAVERSE_STMT(DoStmt, {})
 DEF_TRAVERSE_STMT(ForStmt, {})
 DEF_TRAVERSE_STMT(GotoStmt, {})
+DEF_TRAVERSE_STMT(DeferStmt, {})
 DEF_TRAVERSE_STMT(IfStmt, {})
 DEF_TRAVERSE_STMT(IndirectGotoStmt, {})
 DEF_TRAVERSE_STMT(LabelStmt, {})

clang/include/clang/AST/Stmt.h

@@ -317,6 +317,16 @@ class alignas(void *) Stmt {
     SourceLocation KeywordLoc;
   };
 
+  class DeferStmtBitfields {
+    friend class DeferStmt;
+
+    LLVM_PREFERRED_TYPE(StmtBitfields)
+    unsigned : NumStmtBits;
+
+    /// The location of the "defer".
+    SourceLocation DeferLoc;
+  };
+
   //===--- Expression bitfields classes ---===//
 
   class ExprBitfields {
@@ -1318,6 +1328,7 @@ class alignas(void *) Stmt {
     LoopControlStmtBitfields LoopControlStmtBits;
     ReturnStmtBitfields ReturnStmtBits;
     SwitchCaseBitfields SwitchCaseBits;
+    DeferStmtBitfields DeferStmtBits;
 
     // Expressions
     ExprBitfields ExprBits;
@@ -3211,6 +3222,47 @@ class ReturnStmt final
   }
 };
 
+/// DeferStmt - This represents a deferred statement.
+class DeferStmt : public Stmt {
+  friend class ASTStmtReader;
+
+  /// The deferred statement.
+  Stmt *Body;
+
+  DeferStmt(EmptyShell Empty);
+  DeferStmt(SourceLocation DeferLoc, Stmt *Body);
+
+public:
+  static DeferStmt *CreateEmpty(ASTContext &Context, EmptyShell Empty);
+  static DeferStmt *Create(ASTContext &Context, SourceLocation DeferLoc,
+                           Stmt *Body);
+
+  SourceLocation getDeferLoc() const { return DeferStmtBits.DeferLoc; }
+  void setDeferLoc(SourceLocation DeferLoc) {
+    DeferStmtBits.DeferLoc = DeferLoc;
+  }
+
+  Stmt *getBody() { return Body; }
+  const Stmt *getBody() const { return Body; }
+  void setBody(Stmt *S) {
+    assert(S && "defer body must not be null");
+    Body = S;
+  }
+
+  SourceLocation getBeginLoc() const { return getDeferLoc(); }
+  SourceLocation getEndLoc() const { return Body->getEndLoc(); }
+
+  child_range children() { return child_range(&Body, &Body + 1); }
+
+  const_child_range children() const {
+    return const_child_range(&Body, &Body + 1);
+  }
+
+  static bool classof(const Stmt *S) {
+    return S->getStmtClass() == DeferStmtClass;
+  }
+};
+
 /// AsmStmt is the base class for GCCAsmStmt and MSAsmStmt.
 class AsmStmt : public Stmt {
 protected:

clang/include/clang/Basic/DiagnosticParseKinds.td

@@ -350,6 +350,8 @@ def err_address_of_label_outside_fn : Error<
   "use of address-of-label extension outside of a function body">;
 def err_asm_operand_wide_string_literal : Error<
   "cannot use %select{unicode|wide}0 string literal in 'asm'">;
+def err_defer_ts_labeled_stmt : Error<
+  "substatement of defer must not be a label">;
 
 def err_asm_expected_string : Error<
   "expected string literal %select{or parenthesized constant expression |}0in 'asm'">;

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -6848,6 +6848,7 @@ def note_protected_by_objc_weak_init : Note<
   "jump bypasses initialization of __weak variable">;
 def note_protected_by_non_trivial_c_struct_init : Note<
   "jump bypasses initialization of variable of non-trivial C struct type">;
+def note_protected_by_defer_stmt : Note<"jump bypasses defer statement">;
 def note_enters_block_captures_cxx_obj : Note<
   "jump enters lifetime of block which captures a destructible C++ object">;
 def note_enters_block_captures_strong : Note<
@@ -6861,6 +6862,7 @@ def note_enters_compound_literal_scope : Note<
   "jump enters lifetime of a compound literal that is non-trivial to destruct">;
 def note_enters_statement_expression : Note<
   "jump enters a statement expression">;
+def note_enters_defer_stmt : Note<"jump enters a defer statement">;
 
 def note_exits_cleanup : Note<
   "jump exits scope of variable with __attribute__((cleanup))">;
@@ -6906,6 +6908,16 @@ def note_exits_block_captures_non_trivial_c_struct : Note<
   "to destroy">;
 def note_exits_compound_literal_scope : Note<
   "jump exits lifetime of a compound literal that is non-trivial to destruct">;
+def note_exits_defer_stmt : Note<"jump exits a defer statement">;
+def err_jump_out_of_defer_stmt : Error<
+  "cannot %enum_select<DeferJumpKind>{"
+  "%Break{break out of a}|"
+  "%Continue{continue loop outside of enclosing}|"
+  "%Return{return from a}|"
+  "%SEHLeave{__leave a}"
+  "}0 defer statement">;
+def err_defer_invalid_sjlj : Error<
+  "cannot use %0 inside a defer statement">;
 
 def err_func_returning_qualified_void : ExtWarn<
   "function cannot return qualified void type %0">,
@@ -11020,6 +11032,8 @@ def err_switch_explicit_conversion : Error<
 def err_switch_incomplete_class_type : Error<
   "switch condition has incomplete class type %0">;
 
+// TODO: It ought to be possible to refactor these to be a single warning that
+// uses %enum_select.
 def warn_empty_if_body : Warning<
   "if statement has empty body">, InGroup<EmptyBody>;
 def warn_empty_for_body : Warning<
@@ -11030,6 +11044,8 @@ def warn_empty_while_body : Warning<
   "while loop has empty body">, InGroup<EmptyBody>;
 def warn_empty_switch_body : Warning<
   "switch statement has empty body">, InGroup<EmptyBody>;
+def warn_empty_defer_body : Warning<
+  "defer statement has empty body">, InGroup<EmptyBody>;
 def note_empty_body_on_separate_line : Note<
   "put the semicolon on a separate line to silence this warning">;
 

clang/include/clang/Basic/IdentifierTable.h

@@ -77,7 +77,8 @@ enum TokenKey : unsigned {
   KEYNOZOS = 0x4000000,
   KEYHLSL = 0x8000000,
   KEYFIXEDPOINT = 0x10000000,
-  KEYMAX = KEYFIXEDPOINT, // The maximum key
+  KEYDEFERTS = 0x20000000,
+  KEYMAX = KEYDEFERTS, // The maximum key
   KEYALLCXX = KEYCXX | KEYCXX11 | KEYCXX20,
   KEYALL = (KEYMAX | (KEYMAX - 1)) & ~KEYNOMS18 & ~KEYNOOPENCL &
            ~KEYNOZOS // KEYNOMS18, KEYNOOPENCL, KEYNOZOS are excluded.

clang/include/clang/Basic/LangOptions.def

@@ -194,6 +194,7 @@ LANGOPT(NoSignedZero      , 1, 0, Benign, "Permit Floating Point optimization wi
 LANGOPT(AllowRecip        , 1, 0, Benign, "Permit Floating Point reciprocal")
 LANGOPT(ApproxFunc        , 1, 0, Benign, "Permit Floating Point approximation")
 LANGOPT(NamedLoops        , 1, 0, Benign, "Permit named break/continue")
+LANGOPT(DeferTS           , 1, 0, Benign, "C '_Defer' Technical Specification")
 
 ENUM_LANGOPT(ComplexRange, ComplexRangeKind, 3, CX_None, NotCompatible, "Enable use of range reduction for complex arithmetics.")
 

clang/include/clang/Basic/StmtNodes.td

@@ -17,6 +17,7 @@ def ForStmt : StmtNode<Stmt>;
 def GotoStmt : StmtNode<Stmt>;
 def IndirectGotoStmt : StmtNode<Stmt>;
 def ReturnStmt : StmtNode<Stmt>;
+def DeferStmt : StmtNode<Stmt>;
 def DeclStmt  : StmtNode<Stmt>;
 def SwitchCase : StmtNode<Stmt, 1>;
 def CaseStmt : StmtNode<SwitchCase>;

clang/include/clang/Basic/TokenKinds.def

@@ -293,6 +293,7 @@ PUNCTUATOR(greatergreatergreater, ">>>")
 //   CHAR8SUPPORT - This is a keyword if 'char8_t' is a built-in type
 //   KEYFIXEDPOINT - This is a keyword according to the N1169 fixed point
 //                   extension.
+//   KEYDEFERTS - This is a keyword if the C '_Defer' TS is enabled
 //   KEYZOS - This is a keyword in C/C++ on z/OS
 //
 KEYWORD(auto                        , KEYALL)
@@ -441,6 +442,9 @@ KEYWORD(_Float16                    , KEYALL)
 C23_KEYWORD(typeof                  , KEYGNU)
 C23_KEYWORD(typeof_unqual           , 0)
 
+// '_Defer' TS
+KEYWORD(_Defer                      , KEYDEFERTS)
+
 // ISO/IEC JTC1 SC22 WG14 N1169 Extension
 KEYWORD(_Accum                      , KEYFIXEDPOINT)
 KEYWORD(_Fract                      , KEYFIXEDPOINT)