[lldb] Store expression evaluator diagnostics in an llvm::Error (NFC)

This patch is a reworking of Pete Lawrence's (@PortalPete) proposal
for better expression evaluator error messages:
#80938

This patch is preparatory patch for improving the rendering of
expression evaluator diagnostics. Currently diagnostics are rendered
into a string and the command interpreter layer then textually parses
words like "error:" to (sometimes) color the output accordingly. In
order to enable user interfaces to do better with diagnostics, we need
to store them in a machine-readable fromat. This patch does this by
adding a new llvm::Error kind wrapping a DiagnosticDetail struct that
is used when the error type is eErrorTypeExpression. Multiple
diagnostics are modeled using llvm::ErrorList.

Right now the extra information is not used by the CommandInterpreter,
this will be added in a follow-up patch!

Author: adrian-prantl

lldb/include/lldb/Expression/DiagnosticManager.h

@@ -12,6 +12,9 @@
 #include "lldb/lldb-defines.h"
 #include "lldb/lldb-types.h"
 
+#include "lldb/Utility/FileSpec.h"
+#include "lldb/Utility/Status.h"
+
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/StringRef.h"
 
@@ -20,6 +23,47 @@
 
 namespace lldb_private {
 
+/// A compiler-independent representation of a Diagnostic. Expression
+/// evaluation failures often have more than one diagnostic that a UI
+/// layer might want to render differently, for example to colorize
+/// it.
+///
+/// Running example:
+///   (lldb) expr 1+foo
+///   error: <user expression 0>:1:3: use of undeclared identifier 'foo'
+///   1+foo
+///     ^
+struct DiagnosticDetail {
+  struct SourceLocation {
+    FileSpec file;
+    unsigned line = 0;
+    uint16_t column = 0;
+    uint16_t length = 0;
+    bool in_user_input = false;
+  };
+  /// Contains {{}, 1, 3, 3, true} in the example above.
+  std::optional<SourceLocation> source_location;
+  /// Contains eSeverityError in the example above.
+  lldb::Severity severity = lldb::eSeverityInfo;
+  /// Contains "use of undeclared identifier 'x'" in the example above.
+  std::string message;
+  /// Contains the fully rendered error message.
+  std::string rendered;
+};
+
+/// An llvm::Error used to communicate diagnostics in Status. Multiple
+/// diagnostics may be chained in an llvm::ErrorList.
+class DetailedExpressionError
+    : public llvm::ErrorInfo<DetailedExpressionError, llvm::ECError> {
+  DiagnosticDetail m_detail;
+
+public:
+  using llvm::ErrorInfo<DetailedExpressionError, llvm::ECError>::ErrorInfo;
+  DetailedExpressionError(DiagnosticDetail detail) {}
+  std::string message() const override;
+  static char ID;
+};
+
 enum DiagnosticOrigin {
   eDiagnosticOriginUnknown = 0,
   eDiagnosticOriginLLDB,
@@ -49,37 +93,28 @@ class Diagnostic {
     }
   }
 
-  Diagnostic(llvm::StringRef message, lldb::Severity severity,
-             DiagnosticOrigin origin, uint32_t compiler_id)
-      : m_message(message), m_severity(severity), m_origin(origin),
-        m_compiler_id(compiler_id) {}
-
-  Diagnostic(const Diagnostic &rhs)
-      : m_message(rhs.m_message), m_severity(rhs.m_severity),
-        m_origin(rhs.m_origin), m_compiler_id(rhs.m_compiler_id) {}
+  Diagnostic(DiagnosticOrigin origin, uint32_t compiler_id,
+             DiagnosticDetail detail)
+      : m_origin(origin), m_compiler_id(compiler_id), m_detail(detail) {}
 
   virtual ~Diagnostic() = default;
 
   virtual bool HasFixIts() const { return false; }
 
-  lldb::Severity GetSeverity() const { return m_severity; }
+  lldb::Severity GetSeverity() const { return m_detail.severity; }
 
   uint32_t GetCompilerID() const { return m_compiler_id; }
 
-  llvm::StringRef GetMessage() const { return m_message; }
+  llvm::StringRef GetMessage() const { return m_detail.message; }
+  llvm::Error GetAsError() const;
 
-  void AppendMessage(llvm::StringRef message,
-                     bool precede_with_newline = true) {
-    if (precede_with_newline)
-      m_message.push_back('\n');
-    m_message += message;
-  }
+  void AppendMessage(llvm::StringRef message, bool precede_with_newline = true);
 
 protected:
-  std::string m_message;
-  lldb::Severity m_severity;
   DiagnosticOrigin m_origin;
-  uint32_t m_compiler_id; // Compiler-specific diagnostic ID
+  /// Compiler-specific diagnostic ID.
+  uint32_t m_compiler_id;
+  DiagnosticDetail m_detail;
 };
 
 typedef std::vector<std::unique_ptr<Diagnostic>> DiagnosticList;
@@ -102,10 +137,7 @@ class DiagnosticManager {
 
   void AddDiagnostic(llvm::StringRef message, lldb::Severity severity,
                      DiagnosticOrigin origin,
-                     uint32_t compiler_id = LLDB_INVALID_COMPILER_ID) {
-    m_diagnostics.emplace_back(
-        std::make_unique<Diagnostic>(message, severity, origin, compiler_id));
-  }
+                     uint32_t compiler_id = LLDB_INVALID_COMPILER_ID);
 
   void AddDiagnostic(std::unique_ptr<Diagnostic> diagnostic) {
     if (diagnostic)
@@ -130,12 +162,13 @@ class DiagnosticManager {
       m_diagnostics.back()->AppendMessage(str);
   }
 
-  // Returns a string containing errors in this format:
-  //
-  // "error: error text\n
-  // warning: warning text\n
-  // remark text\n"
-  std::string GetString(char separator = '\n');
+  /// Copies the diagnostics into an llvm::Error{List}.
+  /// The first diagnostic wraps \c result.
+  llvm::Error GetAsError(lldb::ExpressionResults result) const;
+
+  /// Copies the diagnostics into an llvm::Error, the first diagnostic
+  /// being an llvm::StringError.
+  llvm::Error GetAsError(llvm::Twine msg) const;
 
   void Dump(Log *log);
 

lldb/include/lldb/Utility/Status.h

@@ -144,6 +144,7 @@ class Status {
   Status &operator=(Status &&);
   /// FIXME: Replace this with a takeError() method.
   llvm::Error ToError() const;
+  llvm::Error takeError() const { return std::move(m_error); }
   /// Don't call this function in new code. Redesign the API instead.
   Status clone() const { return Status(ToError()); }
 

lldb/source/Breakpoint/BreakpointLocation.cpp

@@ -264,9 +264,9 @@ bool BreakpointLocation::ConditionSaysStop(ExecutionContext &exe_ctx,
     if (!m_user_expression_sp->Parse(diagnostics, exe_ctx,
                                      eExecutionPolicyOnlyWhenNeeded, true,
                                      false)) {
-      error = Status::FromErrorStringWithFormat(
-          "Couldn't parse conditional expression:\n%s",
-          diagnostics.GetString().c_str());
+      error = Status::FromError(
+          diagnostics.GetAsError("Couldn't parse conditional expression:"));
+
       m_user_expression_sp.reset();
       return true;
     }
@@ -324,8 +324,8 @@ bool BreakpointLocation::ConditionSaysStop(ExecutionContext &exe_ctx,
     }
   } else {
     ret = false;
-    error = Status::FromErrorStringWithFormat(
-        "Couldn't execute expression:\n%s", diagnostics.GetString().c_str());
+    error = Status::FromError(
+        diagnostics.GetAsError("Couldn't execute expression:"));
   }
 
   return ret;

lldb/source/Expression/DiagnosticManager.cpp

@@ -14,23 +14,12 @@
 #include "lldb/Utility/StreamString.h"
 
 using namespace lldb_private;
-
-void DiagnosticManager::Dump(Log *log) {
-  if (!log)
-    return;
-
-  std::string str = GetString();
-
-  // GetString() puts a separator after each diagnostic. We want to remove the
-  // last '\n' because log->PutCString will add one for us.
-
-  if (str.size() && str.back() == '\n') {
-    str.pop_back();
-  }
-
-  log->PutCString(str.c_str());
+char DetailedExpressionError::ID;
+std::string DetailedExpressionError::message() const {
+  return m_detail.rendered;
 }
 
+
 static const char *StringForSeverity(lldb::Severity severity) {
   switch (severity) {
   // this should be exhaustive
@@ -44,9 +33,13 @@ static const char *StringForSeverity(lldb::Severity severity) {
   llvm_unreachable("switch needs another case for lldb::Severity enum");
 }
 
-std::string DiagnosticManager::GetString(char separator) {
-  std::string ret;
-  llvm::raw_string_ostream stream(ret);
+
+void DiagnosticManager::Dump(Log *log) {
+  if (!log)
+    return;
+
+  std::string str;
+  llvm::raw_string_ostream stream(str);
 
   for (const auto &diagnostic : Diagnostics()) {
     llvm::StringRef severity = StringForSeverity(diagnostic->GetSeverity());
@@ -59,10 +52,45 @@ std::string DiagnosticManager::GetString(char separator) {
 
     if (severity_pos != llvm::StringRef::npos)
       stream << message.drop_front(severity_pos + severity.size());
-    stream << separator;
+    stream << '\n';
   }
 
-  return ret;
+  // We want to remove the last '\n' because log->PutCString will add
+  // one for us.
+
+  if (str.size() && str.back() == '\n')
+    str.pop_back();
+
+  log->PutString(str);
+}
+
+llvm::Error Diagnostic::GetAsError() const {
+  return llvm::make_error<DetailedExpressionError>(m_detail);
+}
+
+llvm::Error
+DiagnosticManager::GetAsError(lldb::ExpressionResults result) const {
+  llvm::Error diags = Status::FromExpressionError(result, "").takeError();
+  for (const auto &diagnostic : m_diagnostics)
+    diags = llvm::joinErrors(std::move(diags), diagnostic->GetAsError());
+  return diags;
+}
+
+llvm::Error
+DiagnosticManager::GetAsError(llvm::Twine msg) const {
+  llvm::Error diags = llvm::createStringError(msg);
+  for (const auto &diagnostic : m_diagnostics)
+    diags = llvm::joinErrors(std::move(diags), diagnostic->GetAsError());
+  return diags;
+}
+
+void DiagnosticManager::AddDiagnostic(llvm::StringRef message,
+                                      lldb::Severity severity,
+                                      DiagnosticOrigin origin,
+                                      uint32_t compiler_id) {
+  m_diagnostics.emplace_back(std::make_unique<Diagnostic>(
+      origin, compiler_id,
+      DiagnosticDetail{{}, severity, message.str(), message.str()}));
 }
 
 size_t DiagnosticManager::Printf(lldb::Severity severity, const char *format,
@@ -85,3 +113,13 @@ void DiagnosticManager::PutString(lldb::Severity severity,
     return;
   AddDiagnostic(str, severity, eDiagnosticOriginLLDB);
 }
+
+void Diagnostic::AppendMessage(llvm::StringRef message,
+                               bool precede_with_newline) {
+  if (precede_with_newline) {
+    m_detail.message.push_back('\n');
+    m_detail.rendered.push_back('\n');
+  }
+  m_detail.message += message;
+  m_detail.rendered += message;
+}

lldb/source/Expression/ExpressionParser.cpp

@@ -63,9 +63,8 @@ ExpressionParser::RunStaticInitializers(IRExecutionUnitSP &execution_unit_sp,
             exe_ctx, call_static_initializer, options, execution_errors);
 
     if (results != eExpressionCompleted) {
-      err = Status::FromErrorStringWithFormat(
-          "couldn't run static initializer: %s",
-          execution_errors.GetString().c_str());
+      err = Status::FromError(
+          execution_errors.GetAsError("couldn't run static initializer:"));
       return err;
     }
   }

lldb/source/Expression/UserExpression.cpp

@@ -328,18 +328,20 @@ UserExpression::Evaluate(ExecutionContext &exe_ctx,
     }
 
     if (!parse_success) {
-      std::string msg;
-      {
-        llvm::raw_string_ostream os(msg);
-        if (!diagnostic_manager.Diagnostics().empty())
-          os << diagnostic_manager.GetString();
-        else
-          os << "expression failed to parse (no further compiler diagnostics)";
-        if (target->GetEnableNotifyAboutFixIts() && fixed_expression &&
-            !fixed_expression->empty())
-          os << "\nfixed expression suggested:\n  " << *fixed_expression;
+      if (target->GetEnableNotifyAboutFixIts() && fixed_expression &&
+          !fixed_expression->empty()) {
+        std::string fixit =
+            "fixed expression suggested:\n  " + *fixed_expression;
+        diagnostic_manager.AddDiagnostic(fixit, lldb::eSeverityInfo,
+                                         eDiagnosticOriginLLDB);
       }
-      error = Status::FromExpressionError(execution_results, msg);
+      if (!diagnostic_manager.Diagnostics().size())
+        error = Status::FromExpressionError(
+            execution_results,
+            "expression failed to parse (no further compiler diagnostics)");
+      else
+        error =
+            Status::FromError(diagnostic_manager.GetAsError(execution_results));
     }
   }
 
@@ -384,8 +386,8 @@ UserExpression::Evaluate(ExecutionContext &exe_ctx,
           error = Status::FromExpressionError(
               execution_results, "expression failed to execute, unknown error");
         else
-          error = Status::FromExpressionError(execution_results,
-                                              diagnostic_manager.GetString());
+          error = Status::FromError(
+              diagnostic_manager.GetAsError(execution_results));
       } else {
         if (expr_result) {
           result_valobj_sp = expr_result->GetValueObject();

lldb/source/Expression/UtilityFunction.cpp

@@ -83,19 +83,18 @@ FunctionCaller *UtilityFunction::MakeFunctionCaller(
   m_caller_up.reset(process_sp->GetTarget().GetFunctionCallerForLanguage(
       Language().AsLanguageType(), return_type, impl_code_address,
       arg_value_list, name.c_str(), error));
-  if (error.Fail()) {
-
+  if (error.Fail())
     return nullptr;
-  }
+
   if (m_caller_up) {
     DiagnosticManager diagnostics;
 
     unsigned num_errors =
         m_caller_up->CompileFunction(thread_to_use_sp, diagnostics);
     if (num_errors) {
-      error = Status::FromErrorStringWithFormat(
-          "Error compiling %s caller function: \"%s\".",
-          m_function_name.c_str(), diagnostics.GetString().c_str());
+      error = Status::FromError(diagnostics.GetAsError(
+          "Error compiling " + m_function_name + " caller function:"));
+
       m_caller_up.reset();
       return nullptr;
     }
@@ -104,9 +103,8 @@ FunctionCaller *UtilityFunction::MakeFunctionCaller(
     ExecutionContext exe_ctx(process_sp);
 
     if (!m_caller_up->WriteFunctionWrapper(exe_ctx, diagnostics)) {
-      error = Status::FromErrorStringWithFormat(
-          "Error inserting caller function for %s: \"%s\".",
-          m_function_name.c_str(), diagnostics.GetString().c_str());
+      error = Status::FromError(diagnostics.GetAsError(
+          "Error inserting " + m_function_name + " caller function:"));
       m_caller_up.reset();
       return nullptr;
     }

lldb/source/Plugins/ExpressionParser/Clang/ClangDiagnostic.h

@@ -29,9 +29,8 @@ class ClangDiagnostic : public Diagnostic {
     return diag->getKind() == eDiagnosticOriginClang;
   }
 
-  ClangDiagnostic(llvm::StringRef message, lldb::Severity severity,
-                  uint32_t compiler_id)
-      : Diagnostic(message, severity, eDiagnosticOriginClang, compiler_id) {}
+  ClangDiagnostic(DiagnosticDetail detail, uint32_t compiler_id)
+      : Diagnostic(eDiagnosticOriginClang, compiler_id, detail) {}
 
   ~ClangDiagnostic() override = default;