[clang-tidy] Escape diagnostic messages before passing to `diag` in Transformer.

[ymandel]

Summary

Messages generated by Transformer rules may have % in them, which
needs to be escaped before being passed to diag, which interprets them
specially (and crashes if they are misused).

[gribozavr2]

clang-tools-extra/unittests/clang-tidy/TransformerClangTidyCheckTest.cpp

107

[ymandel]

address reviewer comments.

[ymandel]

Thanks!

clang-tools-extra/clang-tidy/utils/TransformerClangTidyCheck.cpp

 #endif
 
 static void verifyRule(const RewriteRuleWith<std::string> &Rule) {
   assert(llvm::all_of(Rule.Metadata, hasGenerator) &&
          "clang-tidy checks must have an explanation by default;"
          " explicitly provide an empty explanation if none is desired");
 }
 
+// If a string unintentionally containing '%' is passed as a diagnostic, Clang
+// will claim the string is ill-formed and assert-fail. This function escapes
+// such strings so they can be safely used in diagnostics.
+std::string escapeForDiagnostic(std::string ToEscape) {
+  // Optimize for the common case that the string does not contain `%` at the
+  // cost of an extra scan over the string in the slow case.
+  auto Pos = ToEscape.find('%');
+  if (Pos == ToEscape.npos)
+    return ToEscape;
+
+  std::string Result;
+  Result.reserve(ToEscape.size());
+  // Convert position to a count.
+  ++Pos;
+  Result.append(ToEscape, 0, Pos);
+  Result += '%';
+
+  for (auto N = ToEscape.size(); Pos < N; ++Pos) {
+    const char C = ToEscape.at(Pos);
+    Result += C;
+    if (C == '%')
+      Result += '%';
+  }
+
+  return Result;
+}
+
 TransformerClangTidyCheck::TransformerClangTidyCheck(StringRef Name,
                                                      ClangTidyContext *Context)
     : ClangTidyCheck(Name, Context),
       Inserter(Options.getLocalOrGlobal("IncludeStyle", IncludeSorter::IS_LLVM),
                areDiagsSelfContained()) {}
 
 // This constructor cannot dispatch to the simpler one (below), because, in
 // order to get meaningful results from `getLangOpts` and `Options`, we need the
void TransformerClangTidyCheck::check(
   Expected<std::string> Explanation = Rule.Metadata[I]->eval(Result);
   if (!Explanation) {
     llvm::errs() << "Error in explanation: "
                  << llvm::toString(Explanation.takeError()) << "\n";
     return;
   }
 
   // Associate the diagnostic with the location of the first change.
-  DiagnosticBuilder Diag = diag((*Edits)[0].Range.getBegin(), *Explanation);
+  DiagnosticBuilder Diag =
+      diag((*Edits)[0].Range.getBegin(), escapeForDiagnostic(*Explanation));
   for (const auto &T : *Edits)
     switch (T.Kind) {
     case transformer::EditKind::Range:
       Diag << FixItHint::CreateReplacement(T.Range, T.Replacement);
       break;
     case transformer::EditKind::AddInclude:
       Diag << Inserter.createIncludeInsertion(
           Result.SourceManager->getFileID(T.Range.getBegin()), T.Replacement);

clang-tools-extra/unittests/clang-tidy/TransformerClangTidyCheckTest.cpp

TEST(TransformerClangTidyCheckTest, DiagnosticsCorrectlyGenerated) {
   EXPECT_EQ(Errors.size(), 1U);
   EXPECT_EQ(Errors[0].Message.Message, "message");
   EXPECT_THAT(Errors[0].Message.Ranges, testing::IsEmpty());
 
   // The diagnostic is anchored to the match, "return 5".
   EXPECT_EQ(Errors[0].Message.FileOffset, 10U);
 }
 
+TEST(TransformerClangTidyCheckTest, DiagnosticMessageEscaped) {
+  class GiveDiagWithPercentSymbol : public TransformerClangTidyCheck {
+  public:
+    GiveDiagWithPercentSymbol(StringRef Name, ClangTidyContext *Context)
+        : TransformerClangTidyCheck(makeRule(returnStmt(),
+                                             noopEdit(node(RootID)),
+                                             cat("bad code: x % y % z")),
+                                    Name, Context) {}
+  };
+  std::string Input = "int somecode() { return 0; }";
+  std::vector<ClangTidyError> Errors;
+  EXPECT_EQ(Input,
+            test::runCheckOnCode<GiveDiagWithPercentSymbol>(Input, &Errors));
+  ASSERT_EQ(Errors.size(), 1U);
+  // This should match the input; this test only verifies that the presence of
+  // '%' doesn't crash Clang.
+  EXPECT_EQ(Errors[0].Message.Message, "bad code: x % y % z");
+}
+
 class IntLitCheck : public TransformerClangTidyCheck {
 public:
   IntLitCheck(StringRef Name, ClangTidyContext *Context)
       : TransformerClangTidyCheck(
             makeRule(integerLiteral(), change(cat("LIT")), cat("no message")),
             Name, Context) {}
 };