Add support for __builtin_verbose_trap (#79230)

The builtin causes the program to stop its execution abnormally and
shows a human-readable description of the reason for the termination
when a debugger is attached or in a symbolicated crash log.

The motivation for the builtin is explained in the following RFC:

https://discourse.llvm.org/t/rfc-adding-builtin-verbose-trap-string-literal/75845

clang's CodeGen lowers the builtin to `llvm.trap` and emits debugging
information that represents an artificial inline frame whose name
encodes the category and reason strings passed to the builtin.

Author: ahatanak

clang/docs/LanguageExtensions.rst

@@ -3483,6 +3483,60 @@ Query for this feature with ``__has_builtin(__builtin_trap)``.
 
 ``__builtin_arm_trap`` is lowered to the ``llvm.aarch64.break`` builtin, and then to ``brk #payload``.
 
+``__builtin_verbose_trap``
+--------------------------
+
+``__builtin_verbose_trap`` causes the program to stop its execution abnormally
+and shows a human-readable description of the reason for the termination when a
+debugger is attached or in a symbolicated crash log.
+
+**Syntax**:
+
+.. code-block:: c++
+
+    __builtin_verbose_trap(const char *category, const char *reason)
+
+**Description**
+
+``__builtin_verbose_trap`` is lowered to the ` ``llvm.trap`` <https://llvm.org/docs/LangRef.html#llvm-trap-intrinsic>`_ builtin.
+Additionally, clang emits debugging information that represents an artificial
+inline frame whose name encodes the category and reason strings passed to the builtin,
+prefixed by a "magic" prefix.
+
+For example, consider the following code:
+
+.. code-block:: c++
+
+    void foo(int* p) {
+      if (p == nullptr)
+        __builtin_verbose_trap("check null", "Argument must not be null!");
+    }
+
+The debugging information would look as if it were produced for the following code:
+
+.. code-block:: c++
+
+    __attribute__((always_inline))
+    inline void "__clang_trap_msg$check null$Argument must not be null!"() {
+      __builtin_trap();
+    }
+
+    void foo(int* p) {
+      if (p == nullptr)
+        "__clang_trap_msg$check null$Argument must not be null!"();
+    }
+
+However, the generated code would not actually contain a call to the artificial
+function — it only exists in the debugging information.
+
+Query for this feature with ``__has_builtin(__builtin_verbose_trap)``. Note that
+users need to enable debug information to enable this feature. A call to this
+builtin is equivalent to a call to ``__builtin_trap`` if debug information isn't
+enabled.
+
+The optimizer can merge calls to trap with different messages, which degrades
+the debugging experience.
+
 ``__builtin_allow_runtime_check``
 ---------------------------------
 

clang/include/clang/AST/Expr.h

@@ -787,6 +787,11 @@ class Expr : public ValueStmt {
                                  const Expr *PtrExpression, ASTContext &Ctx,
                                  EvalResult &Status) const;
 
+  /// If the current Expr can be evaluated to a pointer to a null-terminated
+  /// constant string, return the constant string (without the terminating
+  /// null).
+  std::optional<std::string> tryEvaluateString(ASTContext &Ctx) const;
+
   /// Enumeration used to describe the kind of Null pointer constant
   /// returned from \c isNullPointerConstant().
   enum NullPointerConstantKind {

clang/include/clang/Basic/Builtins.td

@@ -1152,6 +1152,12 @@ def Trap : Builtin {
   let Prototype = "void()";
 }
 
+def VerboseTrap : Builtin {
+  let Spellings = ["__builtin_verbose_trap"];
+  let Attributes = [NoThrow, NoReturn];
+  let Prototype = "void(char const*, char const*)";
+}
+
 def Debugtrap : Builtin {
   let Spellings = ["__builtin_debugtrap"];
   let Attributes = [NoThrow];

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -8956,6 +8956,8 @@ def err_expected_callable_argument : Error<
   "expected a callable expression as %ordinal0 argument to %1, found %2">;
 def note_building_builtin_dump_struct_call : Note<
   "in call to printing function with arguments '(%0)' while dumping struct">;
+def err_builtin_verbose_trap_arg : Error<
+  "argument to __builtin_verbose_trap must %select{be a pointer to a constant string|not contain $}0">;
 
 def err_atomic_load_store_uses_lib : Error<
   "atomic %select{load|store}0 requires runtime support that is not "

clang/include/clang/CodeGen/ModuleBuilder.h

@@ -15,6 +15,7 @@
 
 #include "clang/AST/ASTConsumer.h"
 #include "clang/Basic/LLVM.h"
+#include "llvm/ADT/StringRef.h"
 
 namespace llvm {
   class Constant;
@@ -27,6 +28,9 @@ namespace llvm {
   }
 }
 
+// Prefix of the name of the artificial inline frame.
+inline constexpr llvm::StringRef ClangTrapPrefix = "__clang_trap_msg";
+
 namespace clang {
   class CodeGenOptions;
   class CoverageSourceInfo;

clang/lib/AST/ExprConstant.cpp

@@ -1885,7 +1885,8 @@ static bool EvaluateAtomic(const Expr *E, const LValue *This, APValue &Result,
                            EvalInfo &Info);
 static bool EvaluateAsRValue(EvalInfo &Info, const Expr *E, APValue &Result);
 static bool EvaluateBuiltinStrLen(const Expr *E, uint64_t &Result,
-                                  EvalInfo &Info);
+                                  EvalInfo &Info,
+                                  std::string *StringResult = nullptr);
 
 /// Evaluate an integer or fixed point expression into an APResult.
 static bool EvaluateFixedPointOrInteger(const Expr *E, APFixedPoint &Result,
@@ -17009,7 +17010,7 @@ bool Expr::tryEvaluateObjectSize(uint64_t &Result, ASTContext &Ctx,
 }
 
 static bool EvaluateBuiltinStrLen(const Expr *E, uint64_t &Result,
-                                  EvalInfo &Info) {
+                                  EvalInfo &Info, std::string *StringResult) {
   if (!E->getType()->hasPointerRepresentation() || !E->isPRValue())
     return false;
 
@@ -17036,6 +17037,8 @@ static bool EvaluateBuiltinStrLen(const Expr *E, uint64_t &Result,
         Str = Str.substr(0, Pos);
 
       Result = Str.size();
+      if (StringResult)
+        *StringResult = Str;
       return true;
     }
 
@@ -17051,12 +17054,24 @@ static bool EvaluateBuiltinStrLen(const Expr *E, uint64_t &Result,
     if (!Char.getInt()) {
       Result = Strlen;
       return true;
-    }
+    } else if (StringResult)
+      StringResult->push_back(Char.getInt().getExtValue());
     if (!HandleLValueArrayAdjustment(Info, E, String, CharTy, 1))
       return false;
   }
 }
 
+std::optional<std::string> Expr::tryEvaluateString(ASTContext &Ctx) const {
+  Expr::EvalStatus Status;
+  EvalInfo Info(Ctx, Status, EvalInfo::EM_ConstantFold);
+  uint64_t Result;
+  std::string StringResult;
+
+  if (EvaluateBuiltinStrLen(this, Result, Info, &StringResult))
+    return StringResult;
+  return {};
+}
+
 bool Expr::EvaluateCharRangeAsString(std::string &Result,
                                      const Expr *SizeExpression,
                                      const Expr *PtrExpression, ASTContext &Ctx,

clang/lib/CodeGen/CGBuiltin.cpp

@@ -3505,6 +3505,18 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
   case Builtin::BI__builtin_trap:
     EmitTrapCall(Intrinsic::trap);
     return RValue::get(nullptr);
+  case Builtin::BI__builtin_verbose_trap: {
+    llvm::DILocation *TrapLocation = Builder.getCurrentDebugLocation();
+    if (getDebugInfo()) {
+      TrapLocation = getDebugInfo()->CreateTrapFailureMessageFor(
+          TrapLocation, *E->getArg(0)->tryEvaluateString(getContext()),
+          *E->getArg(1)->tryEvaluateString(getContext()));
+    }
+    ApplyDebugLocation ApplyTrapDI(*this, TrapLocation);
+    // Currently no attempt is made to prevent traps from being merged.
+    EmitTrapCall(Intrinsic::trap);
+    return RValue::get(nullptr);
+  }
   case Builtin::BI__debugbreak:
     EmitTrapCall(Intrinsic::debugtrap);
     return RValue::get(nullptr);

clang/lib/CodeGen/CGDebugInfo.cpp

@@ -32,6 +32,7 @@
 #include "clang/Basic/FileManager.h"
 #include "clang/Basic/SourceManager.h"
 #include "clang/Basic/Version.h"
+#include "clang/CodeGen/ModuleBuilder.h"
 #include "clang/Frontend/FrontendOptions.h"
 #include "clang/Lex/HeaderSearchOptions.h"
 #include "clang/Lex/ModuleMap.h"
@@ -1731,6 +1732,28 @@ llvm::DIType *CGDebugInfo::createFieldType(
                                    offsetInBits, flags, debugType, Annotations);
 }
 
+llvm::DISubprogram *
+CGDebugInfo::createInlinedTrapSubprogram(StringRef FuncName,
+                                         llvm::DIFile *FileScope) {
+  // We are caching the subprogram because we don't want to duplicate
+  // subprograms with the same message. Note that `SPFlagDefinition` prevents
+  // subprograms from being uniqued.
+  llvm::DISubprogram *&SP = InlinedTrapFuncMap[FuncName];
+
+  if (!SP) {
+    llvm::DISubroutineType *DIFnTy = DBuilder.createSubroutineType(nullptr);
+    SP = DBuilder.createFunction(
+        /*Scope=*/FileScope, /*Name=*/FuncName, /*LinkageName=*/StringRef(),
+        /*File=*/FileScope, /*LineNo=*/0, /*Ty=*/DIFnTy,
+        /*ScopeLine=*/0,
+        /*Flags=*/llvm::DINode::FlagArtificial,
+        /*SPFlags=*/llvm::DISubprogram::SPFlagDefinition,
+        /*TParams=*/nullptr, /*ThrownTypes=*/nullptr, /*Annotations=*/nullptr);
+  }
+
+  return SP;
+}
+
 void CGDebugInfo::CollectRecordLambdaFields(
     const CXXRecordDecl *CXXDecl, SmallVectorImpl<llvm::Metadata *> &elements,
     llvm::DIType *RecordTy) {
@@ -3527,6 +3550,23 @@ llvm::DIMacroFile *CGDebugInfo::CreateTempMacroFile(llvm::DIMacroFile *Parent,
   return DBuilder.createTempMacroFile(Parent, Line, FName);
 }
 
+llvm::DILocation *CGDebugInfo::CreateTrapFailureMessageFor(
+    llvm::DebugLoc TrapLocation, StringRef Category, StringRef FailureMsg) {
+  // Create a debug location from `TrapLocation` that adds an artificial inline
+  // frame.
+  SmallString<64> FuncName(ClangTrapPrefix);
+
+  FuncName += "$";
+  FuncName += Category;
+  FuncName += "$";
+  FuncName += FailureMsg;
+
+  llvm::DISubprogram *TrapSP =
+      createInlinedTrapSubprogram(FuncName, TrapLocation->getFile());
+  return llvm::DILocation::get(CGM.getLLVMContext(), /*Line=*/0, /*Column=*/0,
+                               /*Scope=*/TrapSP, /*InlinedAt=*/TrapLocation);
+}
+
 static QualType UnwrapTypeForDebugInfo(QualType T, const ASTContext &C) {
   Qualifiers Quals;
   do {

clang/lib/CodeGen/CGDebugInfo.h

@@ -29,7 +29,9 @@
 #include "llvm/IR/DebugInfo.h"
 #include "llvm/IR/ValueHandle.h"
 #include "llvm/Support/Allocator.h"
+#include <map>
 #include <optional>
+#include <string>
 
 namespace llvm {
 class MDNode;
@@ -346,6 +348,14 @@ class CGDebugInfo {
       const FieldDecl *BitFieldDecl, const llvm::DIDerivedType *BitFieldDI,
       llvm::ArrayRef<llvm::Metadata *> PreviousFieldsDI, const RecordDecl *RD);
 
+  /// A cache that maps names of artificial inlined functions to subprograms.
+  llvm::StringMap<llvm::DISubprogram *> InlinedTrapFuncMap;
+
+  /// A function that returns the subprogram corresponding to the artificial
+  /// inlined function for traps.
+  llvm::DISubprogram *createInlinedTrapSubprogram(StringRef FuncName,
+                                                  llvm::DIFile *FileScope);
+
   /// Helpers for collecting fields of a record.
   /// @{
   void CollectRecordLambdaFields(const CXXRecordDecl *CXXDecl,
@@ -608,6 +618,18 @@ class CGDebugInfo {
     return CoroutineParameterMappings;
   }
 
+  /// Create a debug location from `TrapLocation` that adds an artificial inline
+  /// frame where the frame name is
+  ///
+  /// * `<Prefix>:<Category>:<FailureMsg>`
+  ///
+  /// `<Prefix>` is "__clang_trap_msg".
+  ///
+  /// This is used to store failure reasons for traps.
+  llvm::DILocation *CreateTrapFailureMessageFor(llvm::DebugLoc TrapLocation,
+                                                StringRef Category,
+                                                StringRef FailureMsg);
+
 private:
   /// Emit call to llvm.dbg.declare for a variable declaration.
   /// Returns a pointer to the DILocalVariable associated with the

clang/lib/Sema/SemaChecking.cpp

@@ -183,6 +183,33 @@ bool Sema::checkArgCount(CallExpr *Call, unsigned DesiredArgCount) {
          << /*is non object*/ 0 << Call->getArg(1)->getSourceRange();
 }
 
+static bool checkBuiltinVerboseTrap(CallExpr *Call, Sema &S) {
+  bool HasError = false;
+
+  for (unsigned I = 0; I < Call->getNumArgs(); ++I) {
+    Expr *Arg = Call->getArg(I);
+
+    if (Arg->isValueDependent())
+      continue;
+
+    std::optional<std::string> ArgString = Arg->tryEvaluateString(S.Context);
+    int DiagMsgKind = -1;
+    // Arguments must be pointers to constant strings and cannot use '$'.
+    if (!ArgString.has_value())
+      DiagMsgKind = 0;
+    else if (ArgString->find('$') != std::string::npos)
+      DiagMsgKind = 1;
+
+    if (DiagMsgKind >= 0) {
+      S.Diag(Arg->getBeginLoc(), diag::err_builtin_verbose_trap_arg)
+          << DiagMsgKind << Arg->getSourceRange();
+      HasError = true;
+    }
+  }
+
+  return !HasError;
+}
+
 static bool convertArgumentToType(Sema &S, Expr *&Value, QualType Ty) {
   if (Value->isTypeDependent())
     return false;
@@ -3351,6 +3378,11 @@ Sema::CheckBuiltinFunctionCall(FunctionDecl *FDecl, unsigned BuiltinID,
   case Builtin::BI__builtin_matrix_column_major_store:
     return BuiltinMatrixColumnMajorStore(TheCall, TheCallResult);
 
+  case Builtin::BI__builtin_verbose_trap:
+    if (!checkBuiltinVerboseTrap(TheCall, *this))
+      return ExprError();
+    break;
+
   case Builtin::BI__builtin_get_device_side_mangled_name: {
     auto Check = [](CallExpr *TheCall) {
       if (TheCall->getNumArgs() != 1)