DiagnosticVerifier.cpp

//===--- DiagnosticVerifier.cpp - Diagnostic Verifier (-verify) -----------===//
//
// This source file is part of the Swift.org open source project
//
// Copyright (c) 2014 - 2020 Apple Inc. and the Swift project authors
// Licensed under Apache License v2.0 with Runtime Library Exception
//
// See https://swift.org/LICENSE.txt for license information
// See https://swift.org/CONTRIBUTORS.txt for the list of Swift project authors
//
//===----------------------------------------------------------------------===//
//
//  This file implements the DiagnosticVerifier class.
//
//===----------------------------------------------------------------------===//

#include "swift/Frontend/DiagnosticVerifier.h"
#include "swift/Basic/Assertions.h"
#include "swift/Basic/ColorUtils.h"
#include "swift/Basic/SourceManager.h"
#include "swift/Parse/Lexer.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/FileSystem.h"
#include "llvm/Support/FormatVariadic.h"
#include "llvm/Support/MemoryBuffer.h"
#include "llvm/Support/Path.h"
#include "llvm/Support/raw_ostream.h"

using namespace swift;

namespace {

struct ExpectedCheckMatchStartParser {
  StringRef MatchStart;
  const char *ClassificationStartLoc = nullptr;
  std::optional<DiagnosticKind> ExpectedClassification;

  ExpectedCheckMatchStartParser(StringRef MatchStart)
      : MatchStart(MatchStart) {}

  bool tryParseClassification() {
    if (MatchStart.starts_with("note")) {
      ClassificationStartLoc = MatchStart.data();
      ExpectedClassification = DiagnosticKind::Note;
      MatchStart = MatchStart.substr(strlen("note"));
      return true;
    }

    if (MatchStart.starts_with("warning")) {
      ClassificationStartLoc = MatchStart.data();
      ExpectedClassification = DiagnosticKind::Warning;
      MatchStart = MatchStart.substr(strlen("warning"));
      return true;
    }

    if (MatchStart.starts_with("error")) {
      ClassificationStartLoc = MatchStart.data();
      ExpectedClassification = DiagnosticKind::Error;
      MatchStart = MatchStart.substr(strlen("error"));
      return true;
    }

    if (MatchStart.starts_with("remark")) {
      ClassificationStartLoc = MatchStart.data();
      ExpectedClassification = DiagnosticKind::Remark;
      MatchStart = MatchStart.substr(strlen("remark"));
      return true;
    }

    return false;
  }

  bool parse(ArrayRef<std::string> prefixes) {
    // First try to parse as if we did not have a prefix. We always parse at
    // least expected-*.
    if (tryParseClassification())
      return true;

    // Otherwise, walk our prefixes until we find one that matches and attempt
    // to check for a note, warning, error, or remark.
    //
    // TODO: We could make this more flexible, but this should work in the
    // short term.
    for (auto &p : prefixes) {
      if (MatchStart.starts_with(p)) {
        MatchStart = MatchStart.substr(p.size());
        return tryParseClassification();
      }
    }

    return false;
  }
};

} // anonymous namespace

namespace swift {

struct ExpectedFixIt {
  const char *StartLoc, *EndLoc; // The loc of the {{ and }}'s.
  LineColumnRange Range;

  std::string Text;
};

struct DiagLoc {
  std::optional<unsigned> bufferID;
  unsigned line;
  unsigned column;
  SourceLoc sourceLoc;

  DiagLoc(SourceManager &diagSM, SourceManager &verifierSM,
          SourceLoc initialSourceLoc, bool wantEnd = false)
      : bufferID(std::nullopt), line(0), column(0), sourceLoc(initialSourceLoc)
  {
    if (sourceLoc.isInvalid())
      return;

    // Walk out of generated code for macros in default arguments so that we
    // register diagnostics emitted in them at the call site instead.
    while (true) {
      bufferID = diagSM.findBufferContainingLoc(sourceLoc);
      ASSERT(bufferID.has_value());

      auto generatedInfo = diagSM.getGeneratedSourceInfo(*bufferID);
      if (!generatedInfo || generatedInfo->originalSourceRange.isInvalid()
            || generatedInfo->kind != GeneratedSourceInfo::DefaultArgument)
        break;

      if (wantEnd)
        sourceLoc = generatedInfo->originalSourceRange.getEnd();
      else
        sourceLoc = generatedInfo->originalSourceRange.getStart();

      ASSERT(sourceLoc.isValid());
    }

    // If this diagnostic came from a different SourceManager (as can happen
    // while compiling a module interface), translate its SourceLoc to match the
    // verifier's SourceManager.
    if (&diagSM != &verifierSM) {
      sourceLoc = verifierSM.getLocForForeignLoc(sourceLoc, diagSM);
      bufferID = verifierSM.findBufferContainingLoc(sourceLoc);
    }

    // At this point, `bufferID` is filled in and `sourceLoc` is a location in
    // that buffer.
    if (sourceLoc.isValid())
      std::tie(line, column) = verifierSM.getLineAndColumnInBuffer(sourceLoc);
  }
};

} // end namespace swift

const LineColumnRange &
CapturedFixItInfo::getLineColumnRange(SourceManager &SM) const {
  if (LineColRange.StartLine != 0) {
    // Already computed.
    return LineColRange;
  }

  auto SrcRange = FixIt.getRange();

  DiagLoc startLoc(*diagSM, SM, SrcRange.getStart());
  LineColRange.StartLine = startLoc.line;
  LineColRange.StartCol = startLoc.column;

  DiagLoc endLoc(*diagSM, SM, SrcRange.getEnd(), /*wantEnd=*/true);
  LineColRange.EndLine = endLoc.line;
  LineColRange.EndCol = endLoc.column;

  return LineColRange;
}

namespace {

static constexpr StringLiteral fixitExpectationNoneString("none");
static constexpr StringLiteral categoryDocFileSpecifier("documentation-file=");

struct ExpectedDiagnosticInfo {
  // This specifies the full range of the "expected-foo {{}}" specifier.
  const char *ExpectedStart, *ExpectedEnd = nullptr;

  // This specifies the full range of the classification string.
  const char *ClassificationStart, *ClassificationEnd = nullptr;

  DiagnosticKind Classification;

  // This is true if a '*' constraint is present to say that the diagnostic
  // may appear (or not) an uncounted number of times.
  bool mayAppear = false;

  // This is true if a '{{none}}' is present to mark that there should be no
  // extra fixits.
  bool noExtraFixitsMayAppear() const { return noneMarkerStartLoc != nullptr; };

  // This is the raw input buffer for the message text, the part in the
  // {{...}}
  StringRef MessageRange;

  // This is the message string with escapes expanded.
  std::string MessageStr;
  unsigned LineNo = ~0U;
  std::optional<unsigned> ColumnNo;

  using AlternativeExpectedFixIts = std::vector<ExpectedFixIt>;
  std::vector<AlternativeExpectedFixIts> Fixits = {};

  // Loc of {{none}}
  const char *noneMarkerStartLoc = nullptr;

  /// Represents a specifier of the form '{{documentation-file=note1}}'.
  struct ExpectedDocumentationFile {
    const char *StartLoc, *EndLoc; // The loc of the {{ and }}'s.
    StringRef Name; // Name of expected documentation file.

    ExpectedDocumentationFile(const char *StartLoc, const char *EndLoc,
                              StringRef Name)
        : StartLoc(StartLoc), EndLoc(EndLoc), Name(Name) {}
  };
  std::optional<ExpectedDocumentationFile> DocumentationFile;

  ExpectedDiagnosticInfo(const char *ExpectedStart,
                         const char *ClassificationStart,
                         const char *ClassificationEnd,
                         DiagnosticKind Classification)
      : ExpectedStart(ExpectedStart), ClassificationStart(ClassificationStart),
        ClassificationEnd(ClassificationEnd), Classification(Classification) {}
};

static std::string getDiagKindString(DiagnosticKind Kind) {
  switch (Kind) {
  case DiagnosticKind::Error:
    return "error";
  case DiagnosticKind::Warning:
    return "warning";
  case DiagnosticKind::Note:
    return "note";
  case DiagnosticKind::Remark:
    return "remark";
  }

  llvm_unreachable("Unhandled DiagKind in switch.");
}

/// Render the verifier syntax for a given documentation file.
static std::string
renderDocumentationFile(const std::string &documentationFile) {
  std::string Result;
  llvm::raw_string_ostream OS(Result);
  OS << "{{" << categoryDocFileSpecifier << documentationFile << "}}";
  return OS.str();
}

/// If we find the specified diagnostic in the list, return it with \c true .
/// If we find a near-match that varies only in classification, return it with
/// \c false.
/// Otherwise return \c CapturedDiagnostics.end() with \c false.
static std::tuple<std::vector<CapturedDiagnosticInfo>::iterator, bool>
findDiagnostic(std::vector<CapturedDiagnosticInfo> &CapturedDiagnostics,
               const ExpectedDiagnosticInfo &Expected, unsigned BufferID) {
  auto fallbackI = CapturedDiagnostics.end();

  for (auto I = CapturedDiagnostics.begin(), E = CapturedDiagnostics.end();
       I != E; ++I) {
    // Verify the file and line of the diagnostic.
    if (I->Line != Expected.LineNo || I->SourceBufferID != BufferID)
      continue;

    // If a specific column was expected, verify it.
    if (Expected.ColumnNo.has_value() && I->Column != *Expected.ColumnNo)
      continue;

    // Verify the classification and string.
    if (I->Message.find(Expected.MessageStr) == StringRef::npos)
      continue;

    // Verify the classification and, if incorrect, remember as a second choice.
    if (I->Classification != Expected.Classification) {
      if (fallbackI == E && !Expected.MessageStr.empty())
        fallbackI = I;
      continue;
    }

    // Okay, we found a match, hurray!
    return { I, true };
  }

  // No perfect match; we'll return the fallback or `end()` instead.
  return { fallbackI, false };
}

/// If there are any -verify errors (e.g. differences between expectations
/// and actual diagnostics produced), apply fixits to the original source
/// file and drop it back in place.
static void autoApplyFixes(SourceManager &SM, unsigned BufferID,
                           ArrayRef<llvm::SMDiagnostic> diags) {
  // Walk the list of diagnostics, pulling out any fixits into an array of just
  // them.
  SmallVector<llvm::SMFixIt, 4> FixIts;
  for (auto &diag : diags)
    FixIts.append(diag.getFixIts().begin(), diag.getFixIts().end());

  // If we have no fixits to apply, avoid touching the file.
  if (FixIts.empty())
    return;

  // Sort the fixits by their start location.
  std::sort(FixIts.begin(), FixIts.end(),
            [&](const llvm::SMFixIt &lhs, const llvm::SMFixIt &rhs) -> bool {
              return lhs.getRange().Start.getPointer() <
                     rhs.getRange().Start.getPointer();
            });
  // Coalesce identical fix-its. This happens most often with "expected-error 2"
  // syntax.
  FixIts.erase(std::unique(FixIts.begin(), FixIts.end(),
                           [](const llvm::SMFixIt &lhs,
                              const llvm::SMFixIt &rhs) -> bool {
                             return lhs.getRange().Start ==
                                        rhs.getRange().Start &&
                                    lhs.getRange().End == rhs.getRange().End &&
                                    lhs.getText() == rhs.getText();
                           }),
               FixIts.end());
  // Filter out overlapping fix-its. This allows the compiler to apply changes
  // to the easy parts of the file, and leave in the tricky cases for the
  // developer to handle manually.
  FixIts.erase(swift::removeAdjacentIf(
                   FixIts.begin(), FixIts.end(),
                   [](const llvm::SMFixIt &lhs, const llvm::SMFixIt &rhs) {
                     return lhs.getRange().End.getPointer() >
                            rhs.getRange().Start.getPointer();
                   }),
               FixIts.end());

  // Get the contents of the original source file.
  auto memBuffer = SM.getLLVMSourceMgr().getMemoryBuffer(BufferID);
  auto bufferRange = memBuffer->getBuffer();

  // Apply the fixes, building up a new buffer as an std::string.
  const char *LastPos = bufferRange.begin();
  std::string Result;

  for (auto &fix : FixIts) {
    // We cannot handle overlapping fixits, so assert that they don't happen.
    assert(LastPos <= fix.getRange().Start.getPointer() &&
           "Cannot handle overlapping fixits");

    // Keep anything from the last spot we've checked to the start of the fixit.
    Result.append(LastPos, fix.getRange().Start.getPointer());

    // Replace the content covered by the fixit with the replacement text.
    Result.append(fix.getText().begin(), fix.getText().end());

    // Next character to consider is at the end of the fixit.
    LastPos = fix.getRange().End.getPointer();
  }

  // Retain the end of the file.
  Result.append(LastPos, bufferRange.end());

  std::error_code error;
  llvm::raw_fd_ostream outs(memBuffer->getBufferIdentifier(), error,
                            llvm::sys::fs::OpenFlags::OF_None);
  if (!error)
    outs << Result;
}
} // end anonymous namespace

/// diagnostics for '<unknown>:0' should be considered as unexpected.
bool DiagnosticVerifier::verifyUnknown(
    std::vector<CapturedDiagnosticInfo> &CapturedDiagnostics) const {
  bool HadError = false;
  for (unsigned i = 0, e = CapturedDiagnostics.size(); i != e; ++i) {
    if (CapturedDiagnostics[i].Loc.isValid())
      continue;

    HadError = true;
    std::string Message =
        ("unexpected " +
         getDiagKindString(CapturedDiagnostics[i].Classification) +
         " produced: " + CapturedDiagnostics[i].Message)
            .str();

    auto diag = SM.GetMessage({}, llvm::SourceMgr::DK_Error, Message, {}, {});
    printDiagnostic(diag);
  }
  return HadError;
}

/// Return true if the given \p ExpectedFixIt is in the fix-its emitted by
/// diagnostic \p D.
bool DiagnosticVerifier::checkForFixIt(
    const ExpectedDiagnosticInfo::AlternativeExpectedFixIts &ExpectedAlts,
    const CapturedDiagnosticInfo &D, unsigned BufferID) const {
  for (auto &ActualFixIt : D.FixIts) {
    for (auto &Expected : ExpectedAlts) {
      if (ActualFixIt.getText() != Expected.Text)
        continue;

      auto &ActualRange = ActualFixIt.getLineColumnRange(SM);

      if (Expected.Range.StartCol != ActualRange.StartCol ||
          Expected.Range.EndCol != ActualRange.EndCol ||
          Expected.Range.StartLine != ActualRange.StartLine ||
          Expected.Range.EndLine != ActualRange.EndLine) {
        continue;
      }
      return true;
    }
  }

  return false;
}

void DiagnosticVerifier::printDiagnostic(const llvm::SMDiagnostic &Diag) const {
  raw_ostream &stream = llvm::errs();
  ColoredStream coloredStream{stream};
  raw_ostream &out = UseColor ? coloredStream : stream;
  SM.getLLVMSourceMgr().PrintMessage(out, Diag);
}

std::string
DiagnosticVerifier::renderFixits(ArrayRef<CapturedFixItInfo> ActualFixIts,
                                 unsigned BufferID,
                                 unsigned DiagnosticLineNo) const {
  std::string Result;
  llvm::raw_string_ostream OS(Result);
  interleave(
      ActualFixIts,
      [&](const CapturedFixItInfo &ActualFixIt) {
        auto &ActualRange = ActualFixIt.getLineColumnRange(SM);
        OS << "{{";

        if (ActualRange.StartLine != DiagnosticLineNo)
          OS << ActualRange.StartLine << ':';
        OS << ActualRange.StartCol;

        OS << '-';

        if (ActualRange.EndLine != ActualRange.StartLine)
          OS << ActualRange.EndLine << ':';
        OS << ActualRange.EndCol;

        OS << '=';

        for (auto C : ActualFixIt.getText()) {
          if (C == '\n')
            OS << "\\n";
          else if (C == '}' || C == '\\')
            OS << '\\' << C;
          else
            OS << C;
        }
        OS << "}}";
      },
      [&] { OS << ' '; });
  return OS.str();
}

/// Parse the introductory line-column range of an expected fix-it by consuming
/// the given input string. The range format is \c ([+-]?N:)?N-([+-]?N:)?N
/// where \c 'N' is \c [0-9]+.
///
/// \param DiagnosticLineNo The line number of the associated expected
/// diagnostic; used to turn line offsets into line numbers.
static std::optional<LineColumnRange> parseExpectedFixItRange(
    StringRef &Str, unsigned DiagnosticLineNo,
    llvm::function_ref<void(const char *, const Twine &)> diagnoseError) {
  assert(!Str.empty());

  struct ParsedLineAndColumn {
    std::optional<unsigned> Line;
    unsigned Column;
  };

  const auto parseLineAndColumn = [&]() -> std::optional<ParsedLineAndColumn> {
    enum class OffsetKind : uint8_t { None, Plus, Minus };

    OffsetKind LineOffsetKind = OffsetKind::None;
    if (!Str.empty()) {
      switch (Str.front()) {
      case '+':
        LineOffsetKind = OffsetKind::Plus;
        Str = Str.drop_front();
        break;
      case '-':
        LineOffsetKind = OffsetKind::Minus;
        Str = Str.drop_front();
        break;
      default:
        break;
      }
    }

    unsigned FirstVal = 0;
    if (Str.consumeInteger(10, FirstVal)) {
      if (LineOffsetKind == OffsetKind::None) {
        diagnoseError(Str.data(),
                      "expected line or column number in fix-it verification");
      } else {
        diagnoseError(Str.data(),
                      "expected line offset after leading '+' or '-' in fix-it "
                      "verification");
      }
      return std::nullopt;
    }

    // If the first value is not followed by a colon, it is either a column or a
    // line offset that is missing a column.
    if (Str.empty() || Str.front() != ':') {
      if (LineOffsetKind == OffsetKind::None) {
        return ParsedLineAndColumn{std::nullopt, FirstVal};
      }

      diagnoseError(Str.data(),
                    "expected colon-separated column number after line offset "
                    "in fix-it verification");
      return std::nullopt;
    }

    unsigned Column = 0;
    Str = Str.drop_front();
    if (Str.consumeInteger(10, Column)) {
      diagnoseError(Str.data(),
                    "expected column number after ':' in fix-it verification");
      return std::nullopt;
    }

    // Apply the offset relative to the line of the expected diagnostic.
    switch (LineOffsetKind) {
    case OffsetKind::None:
      break;
    case OffsetKind::Plus:
      FirstVal += DiagnosticLineNo;
      break;
    case OffsetKind::Minus:
      FirstVal = DiagnosticLineNo - FirstVal;
      break;
    }

    return ParsedLineAndColumn{FirstVal, Column};
  };

  LineColumnRange Range;

  if (const auto LineAndCol = parseLineAndColumn()) {
    // The start line defaults to the line of the expected diagnostic.
    Range.StartLine = LineAndCol->Line.value_or(DiagnosticLineNo);
    Range.StartCol = LineAndCol->Column;
  } else {
    return std::nullopt;
  }

  if (!Str.empty() && Str.front() == '-') {
    Str = Str.drop_front();
  } else {
    diagnoseError(Str.data(),
                  "expected '-' range separator in fix-it verification");
    return std::nullopt;
  }

  if (const auto LineAndCol = parseLineAndColumn()) {
    // The end line defaults to the start line.
    Range.EndLine = LineAndCol->Line.value_or(Range.StartLine);
    Range.EndCol = LineAndCol->Column;
  } else {
    return std::nullopt;
  }

  return Range;
}

/// Before we do anything, check if any of our prefixes are prefixes of later
/// prefixes. In such a case, we will never actually pattern match the later
/// prefix. In such a case, crash with a nice error message.
static void validatePrefixList(ArrayRef<std::string> prefixes) {
  // Work backwards through the prefix list.
  while (!prefixes.empty()) {
    auto target = StringRef(prefixes.front());
    prefixes = prefixes.drop_front();

    for (auto &p : prefixes) {
      if (StringRef(p).starts_with(target)) {
        llvm::errs() << "Error! Found a verifier diagnostic additional prefix "
                        "that is a prefix of a later prefix. The later prefix "
                        "will never be pattern matched!\n"
                     << "First Prefix: " << target << '\n'
                     << "Second Prefix: " << p << '\n';
        llvm::report_fatal_error("Standard compiler error!\n");
      }
    }
  }
}

/// After the file has been processed, check to see if we got all of
/// the expected diagnostics and check to see if there were any unexpected
/// ones.
DiagnosticVerifier::Result DiagnosticVerifier::verifyFile(unsigned BufferID) {
  using llvm::SMLoc;
  
  const SourceLoc BufferStartLoc = SM.getLocForBufferStart(BufferID);
  StringRef InputFile = SM.getEntireTextForBuffer(BufferID);

  // Queue up all of the diagnostics, allowing us to sort them and emit them in
  // file order.
  std::vector<llvm::SMDiagnostic> Errors;

  unsigned PrevExpectedContinuationLine = 0;

  std::vector<ExpectedDiagnosticInfo> ExpectedDiagnostics;

  auto addError = [&](const char *Loc, const Twine &message,
                      ArrayRef<llvm::SMFixIt> FixIts = {}) {
    auto loc = SourceLoc(SMLoc::getFromPointer(Loc));
    auto diag = SM.GetMessage(loc, llvm::SourceMgr::DK_Error, message,
                              {}, FixIts);
    Errors.push_back(diag);
  };

  // Validate that earlier prefixes are not prefixes of alter
  // prefixes... otherwise, we will never pattern match the later prefix.
  validatePrefixList(AdditionalExpectedPrefixes);

  // Scan the memory buffer looking for expected-note/warning/error.
  for (size_t Match = InputFile.find("expected-");
       Match != StringRef::npos; Match = InputFile.find("expected-", Match+1)) {
    // Process this potential match.  If we fail to process it, just move on to
    // the next match.
    StringRef MatchStart = InputFile.substr(Match);
    const char *DiagnosticLoc = MatchStart.data();
    MatchStart = MatchStart.substr(strlen("expected-"));

    const char *ClassificationStartLoc = nullptr;
    std::optional<DiagnosticKind> ExpectedClassification;
    {
      ExpectedCheckMatchStartParser parser(MatchStart);
      // If we fail to parse... continue.
      if (!parser.parse(AdditionalExpectedPrefixes)) {
        continue;
      }
      MatchStart = parser.MatchStart;
      ClassificationStartLoc = parser.ClassificationStartLoc;
      ExpectedClassification = parser.ExpectedClassification;
    }
    assert(ClassificationStartLoc);
    assert(bool(ExpectedClassification));

    // Skip any whitespace before the {{.
    MatchStart = MatchStart.substr(MatchStart.find_first_not_of(" \t"));

    size_t TextStartIdx = MatchStart.find("{{");
    if (TextStartIdx >=
        MatchStart.find("\n")) { // Either not found, or found beyond next \n
      addError(MatchStart.data(),
               "expected {{ in expected-warning/note/error line");
      continue;
    }

    ExpectedDiagnosticInfo Expected(DiagnosticLoc, ClassificationStartLoc,
                                    /*ClassificationEndLoc=*/MatchStart.data(),
                                    *ExpectedClassification);
    int LineOffset = 0;

    if (TextStartIdx > 0 && MatchStart[0] == '@') {
      if (MatchStart[1] != '+' && MatchStart[1] != '-' &&
          MatchStart[1] != ':') {
        addError(MatchStart.data(),
                 "expected '+'/'-' for line offset, or ':' for column");
        continue;
      }
      StringRef Offs;
      if (MatchStart[1] == '+')
        Offs = MatchStart.slice(2, TextStartIdx).rtrim();
      else
        Offs = MatchStart.slice(1, TextStartIdx).rtrim();

      size_t SpaceIndex = Offs.find(' ');
      if (SpaceIndex != StringRef::npos && SpaceIndex < TextStartIdx) {
        size_t Delta = Offs.size() - SpaceIndex;
        MatchStart = MatchStart.substr(TextStartIdx - Delta);
        TextStartIdx = Delta;
        Offs = Offs.slice(0, SpaceIndex);
      } else {
        MatchStart = MatchStart.substr(TextStartIdx);
        TextStartIdx = 0;
      }

      size_t ColonIndex = Offs.find(':');
      // Check whether a line offset was provided
      if (ColonIndex != 0) {
        StringRef LineOffs = Offs.slice(0, ColonIndex);
        if (LineOffs.getAsInteger(10, LineOffset)) {
          addError(MatchStart.data(), "expected line offset before '{{'");
          continue;
        }
      }

      // Check whether a column was provided
      if (ColonIndex != StringRef::npos) {
        Offs = Offs.slice(ColonIndex + 1, Offs.size());
        int Column = 0;
        if (Offs.getAsInteger(10, Column)) {
          addError(MatchStart.data(), "expected column before '{{'");
          continue;
        }
        Expected.ColumnNo = Column;
      }
    }

    unsigned Count = 1;
    if (TextStartIdx > 0) {
      StringRef CountStr = MatchStart.substr(0, TextStartIdx).trim(" \t");
      if (CountStr == "*") {
        Expected.mayAppear = true;
      } else {
        if (CountStr.getAsInteger(10, Count)) {
          addError(MatchStart.data(), "expected match count before '{{'");
          continue;
        }
        if (Count == 0) {
          addError(MatchStart.data(),
                   "expected positive match count before '{{'");
          continue;
        }
      }

      // Resync up to the '{{'.
      MatchStart = MatchStart.substr(TextStartIdx);
    }

    size_t End = MatchStart.find("}}");
    if (End == StringRef::npos) {
      addError(MatchStart.data(),
          "didn't find '}}' to match '{{' in expected-warning/note/error line");
      continue;
    }

    llvm::SmallString<256> Buf;
    Expected.MessageRange = MatchStart.slice(2, End);
    Expected.MessageStr =
        Lexer::getEncodedStringSegment(Expected.MessageRange, Buf).str();
    if (PrevExpectedContinuationLine)
      Expected.LineNo = PrevExpectedContinuationLine;
    else
      Expected.LineNo = SM.getLineAndColumnInBuffer(
                              BufferStartLoc.getAdvancedLoc(MatchStart.data() -
                                                            InputFile.data()),
                              BufferID)
                            .first;
    Expected.LineNo += LineOffset;

    // Check if the next expected diagnostic should be in the same line.
    StringRef AfterEnd = MatchStart.substr(End + strlen("}}"));
    AfterEnd = AfterEnd.substr(AfterEnd.find_first_not_of(" \t"));
    if (AfterEnd.starts_with("\\"))
      PrevExpectedContinuationLine = Expected.LineNo;
    else
      PrevExpectedContinuationLine = 0;

    
    // Scan for fix-its: {{10-14=replacement text}}
    bool startNewAlternatives = true;
    StringRef ExtraChecks = MatchStart.substr(End+2).ltrim(" \t");
    while (ExtraChecks.starts_with("{{")) {
      // First make sure we have a closing "}}".
      size_t EndIndex = ExtraChecks.find("}}");
      if (EndIndex == StringRef::npos) {
        addError(ExtraChecks.data(),
                 "didn't find '}}' to match '{{' in diagnostic verification");
        break;
      }

      // Allow for close braces to appear in the replacement text.
      while (EndIndex + 2 < ExtraChecks.size() &&
             ExtraChecks[EndIndex + 2] == '}')
        ++EndIndex;

      const char *OpenLoc = ExtraChecks.data(); // Beginning of opening '{{'.
      const char *CloseLoc =
          ExtraChecks.data() + EndIndex + 2; // End of closing '}}'.

      StringRef CheckStr = ExtraChecks.slice(2, EndIndex);
      // Check for matching a later "}}" on a different line.
      if (CheckStr.find_first_of("\r\n") != StringRef::npos) {
        addError(ExtraChecks.data(), "didn't find '}}' to match '{{' in "
                                     "diagnostic verification");
        break;
      }

      // Prepare for the next round of checks.
      ExtraChecks = ExtraChecks.substr(EndIndex + 2).ltrim(" \t");

      // Handle fix-it alternation.
      // If two fix-its are separated by `||`, we can match either of the two.
      // This is represented by putting them in the same subarray of `Fixits`.
      // If they are not separated by `||`, we must match both of them.
      // This is represented by putting them in separate subarrays of `Fixits`.
      if (startNewAlternatives &&
          (Expected.Fixits.empty() || !Expected.Fixits.back().empty()))
        Expected.Fixits.push_back({});

      if (ExtraChecks.starts_with("||")) {
        startNewAlternatives = false;
        ExtraChecks = ExtraChecks.substr(2).ltrim(" \t");
      } else {
        startNewAlternatives = true;
      }

      // If this check starts with 'documentation-file=', check for a
      // documentation file name instead of a fix-it.
      if (CheckStr.starts_with(categoryDocFileSpecifier)) {
        if (Expected.DocumentationFile.has_value()) {
          addError(CheckStr.data(),
                   "each verified diagnostic may only have one "
                   "{{documentation-file=<#notes#>}} declaration");
          continue;
        }

        // Trim 'documentation-file='.
        StringRef name = CheckStr.substr(categoryDocFileSpecifier.size());
        Expected.DocumentationFile = { OpenLoc, CloseLoc, name };
        continue;
      }

      // This wasn't a documentation file specifier, so it must be a fix-it.
      // Special case for specifying no fixits should appear.
      if (CheckStr == fixitExpectationNoneString) {
        if (Expected.noneMarkerStartLoc) {
          addError(CheckStr.data() - 2,
                   Twine("A second {{") + fixitExpectationNoneString +
                       "}} was found. It may only appear once in an expectation.");
          break;
        }

        Expected.noneMarkerStartLoc = CheckStr.data() - 2;
        continue;
      }

      if (Expected.noneMarkerStartLoc) {
        addError(Expected.noneMarkerStartLoc, Twine("{{") +
                                                  fixitExpectationNoneString +
                                                  "}} must be at the end.");
        break;
      }

      if (CheckStr.empty()) {
        addError(CheckStr.data(), Twine("expected fix-it verification within "
                                        "braces; example: '1-2=text' or '") +
                                      fixitExpectationNoneString + Twine("'"));
        continue;
      }

      // Parse the pieces of the fix-it.
      ExpectedFixIt FixIt;
      FixIt.StartLoc = OpenLoc;
      FixIt.EndLoc = CloseLoc;

      if (const auto range =
              parseExpectedFixItRange(CheckStr, Expected.LineNo, addError)) {
        FixIt.Range = range.value();
      } else {
        continue;
      }

      if (!CheckStr.empty() && CheckStr.front() == '=') {
        CheckStr = CheckStr.drop_front();
      } else {
        addError(CheckStr.data(),
                 "expected '=' after range in fix-it verification");
        continue;
      }

      // Translate literal "\\n" into '\n', inefficiently.
      for (const char *current = CheckStr.begin(), *end = CheckStr.end();
           current != end; /* in loop */) {
        if (*current == '\\' && current + 1 < end) {
          if (current[1] == 'n') {
            FixIt.Text += '\n';
            current += 2;
          } else {  // Handle \}, \\, etc.
            FixIt.Text += current[1];
            current += 2;
          }

        } else {
          FixIt.Text += *current++;
        }
      }

      Expected.Fixits.back().push_back(FixIt);
    }

    // If there's a trailing empty alternation, remove it.
    if (!Expected.Fixits.empty() && Expected.Fixits.back().empty())
      Expected.Fixits.pop_back();

    Expected.ExpectedEnd = ExtraChecks.data();
    
    // Don't include trailing whitespace in the expected-foo{{}} range.
    while (isspace(Expected.ExpectedEnd[-1]))
      --Expected.ExpectedEnd;

    // Add the diagnostic the expected number of times.
    for (; Count; --Count)
      ExpectedDiagnostics.push_back(Expected);
  }

  // Make sure all the expected diagnostics appeared.
  std::reverse(ExpectedDiagnostics.begin(), ExpectedDiagnostics.end());

  for (unsigned i = ExpectedDiagnostics.size(); i != 0; ) {
    --i;
    auto &expected = ExpectedDiagnostics[i];

    // Check to see if we had this expected diagnostic.
    auto FoundDiagnosticInfo =
        findDiagnostic(CapturedDiagnostics, expected, BufferID);
    auto FoundDiagnosticIter = std::get<0>(FoundDiagnosticInfo);
    if (FoundDiagnosticIter == CapturedDiagnostics.end()) {
      // Diagnostic didn't exist.  If this is a 'mayAppear' diagnostic, then
      // we're ok.  Otherwise, leave it in the list.
      if (expected.mayAppear)
        ExpectedDiagnostics.erase(ExpectedDiagnostics.begin()+i);
      continue;
    }

    auto emitFixItsError = [&](const char *location, const Twine &message,
                               const char *replStartLoc, const char *replEndLoc,
                               const std::string &replStr) {
      llvm::SMFixIt fix(llvm::SMRange(SMLoc::getFromPointer(replStartLoc),
                                      SMLoc::getFromPointer(replEndLoc)),
                        replStr);
      addError(location, message, fix);
    };

    auto &FoundDiagnostic = *FoundDiagnosticIter;

    if (!std::get<1>(FoundDiagnosticInfo)) {
      // Found a diagnostic with the right location and text but the wrong
      // classification. We'll emit an error about the mismatch and
      // thereafter pretend that the diagnostic fully matched.
      auto expectedKind = getDiagKindString(expected.Classification);
      auto actualKind = getDiagKindString(FoundDiagnostic.Classification);
      emitFixItsError(expected.ClassificationStart,
          llvm::Twine("expected ") + expectedKind + ", not " + actualKind,
          expected.ClassificationStart, expected.ClassificationEnd,
          actualKind);
    }

    const char *missedFixitLoc = nullptr;
    // Verify that any expected fix-its are present in the diagnostic.
    for (auto fixitAlternates : expected.Fixits) {
      assert(!fixitAlternates.empty() && "an empty alternation survived");

      // If we found it, we're ok.
      if (!checkForFixIt(fixitAlternates, FoundDiagnostic, BufferID)) {
        missedFixitLoc = fixitAlternates.front().StartLoc;
        break;
      }
    }

    const bool isUnexpectedFixitsSeen =
        expected.Fixits.size() < FoundDiagnostic.FixIts.size();

    struct ActualFixitsPhrase {
      std::string phrase;
      std::string actualFixits;
    };

    auto makeActualFixitsPhrase =
        [&](ArrayRef<CapturedFixItInfo> actualFixits) -> ActualFixitsPhrase {
      std::string actualFixitsStr =
          renderFixits(actualFixits, BufferID, expected.LineNo);

      return ActualFixitsPhrase{(Twine("actual fix-it") +
                                 (actualFixits.size() >= 2 ? "s" : "") +
                                 " seen: " + actualFixitsStr).str(),
                                actualFixitsStr};
    };

    // If we have any expected fixits that didn't get matched, then they are
    // wrong.  Replace the failed fixit with what actually happened.

    if (missedFixitLoc) {
      // If we had an incorrect expected fixit, render it and produce a fixit
      // of our own.

      assert(!expected.Fixits.empty() &&
             "some fix-its should be expected here");

      const char *replStartLoc = expected.Fixits.front().front().StartLoc;
      const char *replEndLoc = expected.Fixits.back().back().EndLoc;

      std::string message = "expected fix-it not seen";
      std::string actualFixits;

      if (FoundDiagnostic.FixIts.empty()) {
        /// If actual fix-its is empty,
        /// eat a space before first marker.
        /// For example,
        ///
        /// @code
        /// expected-error {{message}} {{1-2=aa}}
        ///                           ~~~~~~~~~~~
        ///                           ^ remove
        /// @endcode
        if (replStartLoc[-1] == ' ') {
          --replStartLoc;
        }
      } else {
        auto phrase = makeActualFixitsPhrase(FoundDiagnostic.FixIts);
        actualFixits = phrase.actualFixits;
        message += "; " + phrase.phrase;
      }

      emitFixItsError(missedFixitLoc, message, replStartLoc, replEndLoc,
                      actualFixits);
    } else if (expected.noExtraFixitsMayAppear() && isUnexpectedFixitsSeen) {
      // If unexpected fixit were produced, add a fixit to add them in.

      assert(!FoundDiagnostic.FixIts.empty() &&
             "some fix-its should be produced here");
      assert(expected.noneMarkerStartLoc && "none marker location is null");

      const char *replStartLoc = nullptr, *replEndLoc = nullptr;
      std::string message;
      if (expected.Fixits.empty()) {
        message = "expected no fix-its";
        replStartLoc = expected.noneMarkerStartLoc;
        replEndLoc = expected.noneMarkerStartLoc;
      } else {
        message = "unexpected fix-it seen";
        replStartLoc = expected.Fixits.front().front().StartLoc;
        replEndLoc = expected.Fixits.back().back().EndLoc;
      }

      auto phrase = makeActualFixitsPhrase(FoundDiagnostic.FixIts);
      std::string actualFixits = phrase.actualFixits;
      message += "; " + phrase.phrase;

      if (replStartLoc == replEndLoc) {
        /// If no fix-its was expected and range of replacement is empty,
        /// insert space after new last marker.
        /// For example:
        ///
        /// @code
        /// expected-error {{message}} {{none}}
        ///                            ^
        ///                    insert `{{1-2=aa}} `
        /// @endcode
        actualFixits += " ";
      }

      emitFixItsError(expected.noneMarkerStartLoc, message, replStartLoc,
                      replEndLoc, actualFixits);
    }

    if (auto expectedDocFile = expected.DocumentationFile) {
      // Verify diagnostic file.
      if (FoundDiagnostic.CategoryDocFile == expectedDocFile->Name)
        expectedDocFile = std::nullopt;

      if (expectedDocFile) {
        if (FoundDiagnostic.CategoryDocFile.empty()) {
          addError(expectedDocFile->StartLoc,
                   "expected documentation file not seen");
        } else {
          // If we had an incorrect expected document file, render it and
          // produce a fixit of our own.
          auto actual =
              renderDocumentationFile(FoundDiagnostic.CategoryDocFile);
          auto replStartLoc = SMLoc::getFromPointer(expectedDocFile->StartLoc);
          auto replEndLoc = SMLoc::getFromPointer(expectedDocFile->EndLoc);

          llvm::SMFixIt fix(llvm::SMRange(replStartLoc, replEndLoc), actual);
          addError(expectedDocFile->StartLoc,
                   "expected documentation file not seen; actual documentation "
                   "file: " + actual, fix);
        }
      }
    }

    // Actually remove the diagnostic from the list, so we don't match it
    // again. We do have to do this after checking fix-its, though, because
    // the diagnostic owns its fix-its.
    CapturedDiagnostics.erase(FoundDiagnosticIter);
    
    // We found the diagnostic, so remove it... unless we allow an arbitrary
    // number of diagnostics, in which case we want to reprocess this.
    if (expected.mayAppear)
      ++i;
    else
      ExpectedDiagnostics.erase(ExpectedDiagnostics.begin()+i);
  }
  
  // Check to see if we have any incorrect diagnostics.  If so, diagnose them as
  // such.
  auto expectedDiagIter = ExpectedDiagnostics.begin();
  while (expectedDiagIter != ExpectedDiagnostics.end()) {
    // Check to see if any found diagnostics have the right line and
    // classification, but the wrong text.
    auto I = CapturedDiagnostics.begin();
    for (auto E = CapturedDiagnostics.end(); I != E; ++I) {
      // Verify the file and line of the diagnostic.
      if (I->Line != expectedDiagIter->LineNo || I->SourceBufferID != BufferID
            || I->Classification != expectedDiagIter->Classification)
        continue;
      
      // Otherwise, we found it, break out.
      break;
    }

    if (I == CapturedDiagnostics.end()) {
      ++expectedDiagIter;
      continue;
    }

    if (I->Message.find(expectedDiagIter->MessageStr) == StringRef::npos) {
      auto StartLoc =
          SMLoc::getFromPointer(expectedDiagIter->MessageRange.begin());
      auto EndLoc = SMLoc::getFromPointer(expectedDiagIter->MessageRange.end());

      llvm::SMFixIt fixIt(llvm::SMRange{StartLoc, EndLoc}, I->Message);
      addError(expectedDiagIter->MessageRange.begin(),
               "incorrect message found", fixIt);
    } else if (I->Column != *expectedDiagIter->ColumnNo) {
      // The difference must be only in the column
      addError(expectedDiagIter->MessageRange.begin(),
               llvm::formatv("message found at column {0} but was expected to "
                             "appear at column {1}",
                             I->Column, *expectedDiagIter->ColumnNo));
    } else {
      llvm_unreachable("unhandled difference from expected diagnostic");
    }
    CapturedDiagnostics.erase(I);
    expectedDiagIter = ExpectedDiagnostics.erase(expectedDiagIter);
  }

  // Diagnose expected diagnostics that didn't appear.
  std::reverse(ExpectedDiagnostics.begin(), ExpectedDiagnostics.end());
  for (auto const &expected : ExpectedDiagnostics) {
    std::string message = "expected "+getDiagKindString(expected.Classification)
      + " not produced";

    // Get the range of the expected-foo{{}} diagnostic specifier.
    auto StartLoc = expected.ExpectedStart;
    auto EndLoc = expected.ExpectedEnd;

    // A very common case if for the specifier to be the last thing on the line.
    // In this case, eat any trailing whitespace.
    while (isspace(*EndLoc) && *EndLoc != '\n' && *EndLoc != '\r')
      ++EndLoc;

    // If we found the end of the line, we can do great things.  Otherwise,
    // avoid nuking whitespace that might be zapped through other means.
    if (*EndLoc != '\n' && *EndLoc != '\r') {
      EndLoc = expected.ExpectedEnd;
    } else {
      // If we hit the end of line, then zap whitespace leading up to it.
      auto FileStart = InputFile.data();
      while (StartLoc-1 != FileStart && isspace(StartLoc[-1]) &&
             StartLoc[-1] != '\n' && StartLoc[-1] != '\r')
        --StartLoc;

      // If we got to the end of the line, and the thing before this diagnostic
      // is a "//" then we can remove it too.
      if (StartLoc-2 >= FileStart && StartLoc[-1] == '/' && StartLoc[-2] == '/')
        StartLoc -= 2;

      // Perform another round of general whitespace nuking to cleanup
      // whitespace before the //.
      while (StartLoc-1 != FileStart && isspace(StartLoc[-1]) &&
             StartLoc[-1] != '\n' && StartLoc[-1] != '\r')
        --StartLoc;

      // If we found a \n, then we can nuke the entire line.
      if (StartLoc-1 != FileStart &&
          (StartLoc[-1] == '\n' || StartLoc[-1] == '\r'))
        --StartLoc;
    }

    // Remove the expected-foo{{}} as a fixit.
    llvm::SMFixIt fixIt(llvm::SMRange{
      SMLoc::getFromPointer(StartLoc),
      SMLoc::getFromPointer(EndLoc)
    }, "");
    addError(expected.ExpectedStart, message, fixIt);
  }
  
  // Verify that there are no diagnostics (in MemoryBuffer) left in the list.
  bool HadUnexpectedDiag = false;
  auto CapturedDiagIter = CapturedDiagnostics.begin();
  while (CapturedDiagIter != CapturedDiagnostics.end()) {
    if (CapturedDiagIter->SourceBufferID != BufferID) {
      ++CapturedDiagIter;
      continue;
    }

    HadUnexpectedDiag = true;
    std::string Message =
        ("unexpected " + getDiagKindString(CapturedDiagIter->Classification) +
         " produced: " + CapturedDiagIter->Message)
            .str();
    addError(getRawLoc(CapturedDiagIter->Loc).getPointer(), Message);
    CapturedDiagIter = CapturedDiagnostics.erase(CapturedDiagIter);
  }

  // Sort the diagnostics by their address in the memory buffer as the primary
  // key.  This ensures that an "unexpected diagnostic" and
  // "expected diagnostic" in the same place are emitted next to each other.
  std::sort(Errors.begin(), Errors.end(),
            [&](const llvm::SMDiagnostic &lhs,
                const llvm::SMDiagnostic &rhs) -> bool {
              return lhs.getLoc().getPointer() < rhs.getLoc().getPointer();
            });

  // Emit all of the queue'd up errors.
  for (auto Err : Errors)
    printDiagnostic(Err);

  // If auto-apply fixits is on, rewrite the original source file.
  if (AutoApplyFixes)
    autoApplyFixes(SM, BufferID, Errors);

  return Result{!Errors.empty(), HadUnexpectedDiag};
}

void DiagnosticVerifier::printRemainingDiagnostics() const {
  for (const auto &diag : CapturedDiagnostics) {
    // Determine what kind of diagnostic we're emitting.
    llvm::SourceMgr::DiagKind SMKind;
    switch (diag.Classification) {
    case DiagnosticKind::Error:
      SMKind = llvm::SourceMgr::DK_Error;
      break;
    case DiagnosticKind::Warning:
      SMKind = llvm::SourceMgr::DK_Warning;
      break;

    case DiagnosticKind::Note:
      SMKind = llvm::SourceMgr::DK_Note;
      break;

    case DiagnosticKind::Remark:
      SMKind = llvm::SourceMgr::DK_Remark;
      break;
    }

    auto message =
        SM.GetMessage(diag.Loc, SMKind,
                      "diagnostic produced elsewhere: " + diag.Message.str(),
                      /*Ranges=*/{}, {});
    printDiagnostic(message);
  }
}

//===----------------------------------------------------------------------===//
// Main entrypoints
//===----------------------------------------------------------------------===//

/// Every time a diagnostic is generated in -verify mode, this function is
/// called with the diagnostic.  We just buffer them up until the end of the
/// file.
void DiagnosticVerifier::handleDiagnostic(SourceManager &SM,
                                          const DiagnosticInfo &Info) {
  SmallVector<CapturedFixItInfo, 2> fixIts;
  for (const auto &fixIt : Info.FixIts) {
    fixIts.emplace_back(SM, fixIt);
  }

  llvm::SmallString<128> message;
  {
    llvm::raw_svector_ostream Out(message);
    DiagnosticEngine::formatDiagnosticText(Out, Info.FormatString,
                                           Info.FormatArgs);
  }

  DiagLoc loc(SM, this->SM, Info.Loc);
  CapturedDiagnostics.emplace_back(message, loc.bufferID, Info.Kind,
                                   loc.sourceLoc, loc.line, loc.column, fixIts,
                                   llvm::sys::path::stem(
                                      Info.CategoryDocumentationURL).str());
}

/// Once all diagnostics have been captured, perform verification.
bool DiagnosticVerifier::finishProcessing() {
  DiagnosticVerifier::Result Result = {false, false};

  SmallVector<unsigned, 4> additionalBufferIDs;
  for (auto path : AdditionalFilePaths) {
    llvm::SmallString<261> buffer{path};
    llvm::sys::path::make_preferred(buffer);
    auto bufferID = SM.getIDForBufferIdentifier(buffer);
    if (!bufferID) {
      // Still need to scan this file for expectations.
      auto result = SM.getFileSystem()->getBufferForFile(buffer);
      if (!result) {
        auto message = SM.GetMessage(
           SourceLoc(), llvm::SourceMgr::DiagKind::DK_Error,
           llvm::Twine("unable to open file in '-verify-additional-file ") +
           buffer + "': " + result.getError().message(), {}, {});
        printDiagnostic(message);
        Result.HadError |= true;
        continue;
      }
      bufferID = SM.addNewSourceBuffer(std::move(result.get()));
    }
    if (bufferID) {
      additionalBufferIDs.push_back(*bufferID);
    }
  }

  ArrayRef<unsigned> BufferIDLists[2] = { BufferIDs, additionalBufferIDs };
  for (ArrayRef<unsigned> BufferIDList : BufferIDLists)
    for (auto &BufferID : BufferIDList) {
      DiagnosticVerifier::Result FileResult = verifyFile(BufferID);
      Result.HadError |= FileResult.HadError;
      Result.HadUnexpectedDiag |= FileResult.HadUnexpectedDiag;
    }
  if (!IgnoreUnknown) {
    bool HadError = verifyUnknown(CapturedDiagnostics);
    Result.HadError |= HadError;
    // For <unknown>, all errors are unexpected.
    Result.HadUnexpectedDiag |= HadError;
  }

  if (Result.HadUnexpectedDiag)
    printRemainingDiagnostics();

  return Result.HadError;
}