1 files changed, 535 insertions, 22 deletions

diff --git a/llvm/lib/TableGen/TGLexer.cpp b/llvm/lib/TableGen/TGLexer.cpp
index 652be6e8dbb..fcabce5329c 100644
--- a/llvm/lib/TableGen/TGLexer.cpp
+++ b/llvm/lib/TableGen/TGLexer.cpp
@@ -19,6 +19,7 @@
 #include "llvm/Support/MemoryBuffer.h"
 #include "llvm/Support/SourceMgr.h"
 #include "llvm/TableGen/Error.h"
+#include <algorithm>
 #include <cctype>
 #include <cerrno>
 #include <cstdint>
@@ -28,11 +29,35 @@
 
 using namespace llvm;
 
-TGLexer::TGLexer(SourceMgr &SM) : SrcMgr(SM) {
+namespace {
+// A list of supported preprocessing directives with their
+// internal token kinds and names.
+struct {
+  tgtok::TokKind Kind;
+  const char *Word;
+} PreprocessorDirs[] = {
+  { tgtok::Ifdef, "ifdef" },
+  { tgtok::Else, "else" },
+  { tgtok::Endif, "endif" },
+  { tgtok::Define, "define" }
+};
+} // end anonymous namespace
+
+TGLexer::TGLexer(SourceMgr &SM, ArrayRef<std::string> Macros) : SrcMgr(SM) {
   CurBuffer = SrcMgr.getMainFileID();
   CurBuf = SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer();
   CurPtr = CurBuf.begin();
   TokStart = nullptr;
+
+  // Pretend that we enter the "top-level" include file.
+  PrepIncludeStack.push_back(
+      make_unique<std::vector<PreprocessorControlDesc>>());
+
+  // Put all macros defined in the command line into the DefinedMacros set.
+  std::for_each(Macros.begin(), Macros.end(),
+                [this](const std::string &MacroName) {
+                  DefinedMacros.insert(MacroName);
+                });
 }
 
 SMLoc TGLexer::getLoc() const {
@@ -41,11 +66,42 @@ SMLoc TGLexer::getLoc() const {
 
 /// ReturnError - Set the error to the specified string at the specified
 /// location.  This is defined to always return tgtok::Error.
-tgtok::TokKind TGLexer::ReturnError(const char *Loc, const Twine &Msg) {
+tgtok::TokKind TGLexer::ReturnError(SMLoc Loc, const Twine &Msg) {
   PrintError(Loc, Msg);
   return tgtok::Error;
 }
 
+tgtok::TokKind TGLexer::ReturnError(const char *Loc, const Twine &Msg) {
+  return ReturnError(SMLoc::getFromPointer(Loc), Msg);
+}
+
+bool TGLexer::processEOF() {
+  SMLoc ParentIncludeLoc = SrcMgr.getParentIncludeLoc(CurBuffer);
+  if (ParentIncludeLoc != SMLoc()) {
+    // If prepExitInclude() detects a problem with the preprocessing
+    // control stack, it will return false.  Pretend that we reached
+    // the final EOF and stop lexing more tokens by returning false
+    // to LexToken().
+    if (!prepExitInclude(false))
+      return false;
+
+    CurBuffer = SrcMgr.FindBufferContainingLoc(ParentIncludeLoc);
+    CurBuf = SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer();
+    CurPtr = ParentIncludeLoc.getPointer();
+    // Make sure TokStart points into the parent file's buffer.
+    // LexToken() assigns to it before calling getNextChar(),
+    // so it is pointing into the included file now.
+    TokStart = CurPtr;
+    return true;
+  }
+
+  // Pretend that we exit the "top-level" include file.
+  // Note that in case of an error (e.g. control stack imbalance)
+  // the routine will issue a fatal error.
+  prepExitInclude(true);
+  return false;
+}
+
 int TGLexer::getNextChar() {
   char CurChar = *CurPtr++;
   switch (CurChar) {
@@ -57,16 +113,6 @@ int TGLexer::getNextChar() {
     if (CurPtr-1 != CurBuf.end())
       return 0;  // Just whitespace.
 
-    // If this is the end of an included file, pop the parent file off the
-    // include stack.
-    SMLoc ParentIncludeLoc = SrcMgr.getParentIncludeLoc(CurBuffer);
-    if (ParentIncludeLoc != SMLoc()) {
-      CurBuffer = SrcMgr.FindBufferContainingLoc(ParentIncludeLoc);
-      CurBuf = SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer();
-      CurPtr = ParentIncludeLoc.getPointer();
-      return getNextChar();
-    }
-
     // Otherwise, return end of file.
     --CurPtr;  // Another call to lex will return EOF again.
     return EOF;
@@ -83,11 +129,11 @@ int TGLexer::getNextChar() {
   }
 }
 
-int TGLexer::peekNextChar(int Index) {
+int TGLexer::peekNextChar(int Index) const {
   return *(CurPtr + Index);
 }
 
-tgtok::TokKind TGLexer::LexToken() {
+tgtok::TokKind TGLexer::LexToken(bool FileOrLineStart) {
   TokStart = CurPtr;
   // This always consumes at least one character.
   int CurChar = getNextChar();
@@ -100,7 +146,18 @@ tgtok::TokKind TGLexer::LexToken() {
 
     // Unknown character, emit an error.
     return ReturnError(TokStart, "Unexpected character");
-  case EOF: return tgtok::Eof;
+  case EOF:
+    // Lex next token, if we just left an include file.
+    // Note that leaving an include file means that the next
+    // symbol is located at the end of 'include "..."'
+    // construct, so LexToken() is called with default
+    // false parameter.
+    if (processEOF())
+      return LexToken();
+
+    // Return EOF denoting the end of lexing.
+    return tgtok::Eof;
+
   case ':': return tgtok::colon;
   case ';': return tgtok::semi;
   case '.': return tgtok::period;
@@ -114,15 +171,27 @@ tgtok::TokKind TGLexer::LexToken() {
   case ')': return tgtok::r_paren;
   case '=': return tgtok::equal;
   case '?': return tgtok::question;
-  case '#': return tgtok::paste;
+  case '#':
+    if (FileOrLineStart) {
+      tgtok::TokKind Kind = prepIsDirective();
+      if (Kind != tgtok::Error)
+        return lexPreprocessor(Kind);
+    }
+
+    return tgtok::paste;
+
+  case '\r':
+    PrintFatalError("getNextChar() must never return '\r'");
+    return tgtok::Error;
 
   case 0:
   case ' ':
   case '\t':
-  case '\n':
-  case '\r':
     // Ignore whitespace.
-    return LexToken();
+    return LexToken(FileOrLineStart);
+  case '\n':
+    // Ignore whitespace, and identify the new line.
+    return LexToken(true);
   case '/':
     // If this is the start of a // comment, skip until the end of the line or
     // the end of the buffer.
@@ -133,7 +202,7 @@ tgtok::TokKind TGLexer::LexToken() {
         return tgtok::Error;
     } else // Otherwise, this is an error.
       return ReturnError(TokStart, "Unexpected character");
-    return LexToken();
+    return LexToken(FileOrLineStart);
   case '-': case '+':
   case '0': case '1': case '2': case '3': case '4': case '5': case '6':
   case '7': case '8': case '9': {
@@ -249,10 +318,10 @@ tgtok::TokKind TGLexer::LexVarName() {
 }
 
 tgtok::TokKind TGLexer::LexIdentifier() {
-  // The first letter is [a-zA-Z_#].
+  // The first letter is [a-zA-Z_].
   const char *IdentStart = TokStart;
 
-  // Match the rest of the identifier regex: [0-9a-zA-Z_#]*
+  // Match the rest of the identifier regex: [0-9a-zA-Z_]*
   while (isalpha(*CurPtr) || isdigit(*CurPtr) || *CurPtr == '_')
     ++CurPtr;
 
@@ -322,6 +391,9 @@ bool TGLexer::LexInclude() {
   // Save the line number and lex buffer of the includer.
   CurBuf = SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer();
   CurPtr = CurBuf.begin();
+
+  PrepIncludeStack.push_back(
+      make_unique<std::vector<PreprocessorControlDesc>>());
   return false;
 }
 
@@ -496,3 +568,444 @@ tgtok::TokKind TGLexer::LexExclaim() {
 
   return Kind != tgtok::Error ? Kind : ReturnError(Start-1, "Unknown operator");
 }
+
+bool TGLexer::prepExitInclude(bool IncludeStackMustBeEmpty) {
+  // Report an error, if preprocessor control stack for the current
+  // file is not empty.
+  if (!PrepIncludeStack.back()->empty()) {
+    prepReportPreprocessorStackError();
+
+    return false;
+  }
+
+  // Pop the preprocessing controls from the include stack.
+  if (PrepIncludeStack.empty()) {
+    PrintFatalError("Preprocessor include stack is empty");
+  }
+
+  PrepIncludeStack.pop_back();
+
+  if (IncludeStackMustBeEmpty) {
+    if (!PrepIncludeStack.empty())
+      PrintFatalError("Preprocessor include stack is not empty");
+  } else {
+    if (PrepIncludeStack.empty())
+      PrintFatalError("Preprocessor include stack is empty");
+  }
+
+  return true;
+}
+
+tgtok::TokKind TGLexer::prepIsDirective() const {
+  for (unsigned ID = 0; ID < llvm::array_lengthof(PreprocessorDirs); ++ID) {
+    int NextChar = *CurPtr;
+    bool Match = true;
+    unsigned I = 0;
+    for (; I < strlen(PreprocessorDirs[ID].Word); ++I) {
+      if (NextChar != PreprocessorDirs[ID].Word[I]) {
+        Match = false;
+        break;
+      }
+
+      NextChar = peekNextChar(I + 1);
+    }
+
+    // Check for whitespace after the directive.  If there is no whitespace,
+    // then we do not recognize it as a preprocessing directive.
+    if (Match) {
+      tgtok::TokKind Kind = PreprocessorDirs[ID].Kind;
+
+      // New line and EOF may follow only #else/#endif.  It will be reported
+      // as an error for #ifdef/#define after the call to prepLexMacroName().
+      if (NextChar == ' ' || NextChar == '\t' || NextChar == EOF ||
+          NextChar == '\n' ||
+          // It looks like TableGen does not support '\r' as the actual
+          // carriage return, e.g. getNextChar() treats a single '\r'
+          // as '\n'.  So we do the same here.
+          NextChar == '\r')
+        return Kind;
+
+      // Allow comments after some directives, e.g.:
+      //     #else// OR #else/**/
+      //     #endif// OR #endif/**/
+      //
+      // Note that we do allow comments after #ifdef/#define here, e.g.
+      //     #ifdef/**/ AND #ifdef//
+      //     #define/**/ AND #define//
+      //
+      // These cases will be reported as incorrect after calling
+      // prepLexMacroName().  We could have supported C-style comments
+      // after #ifdef/#define, but this would complicate the code
+      // for little benefit.
+      if (NextChar == '/') {
+        NextChar = peekNextChar(I + 1);
+
+        if (NextChar == '*' || NextChar == '/')
+          return Kind;
+
+        // Pretend that we do not recognize the directive.
+      }
+    }
+  }
+
+  return tgtok::Error;
+}
+
+bool TGLexer::prepEatPreprocessorDirective(tgtok::TokKind Kind) {
+  TokStart = CurPtr;
+
+  for (unsigned ID = 0; ID < llvm::array_lengthof(PreprocessorDirs); ++ID)
+    if (PreprocessorDirs[ID].Kind == Kind) {
+      // Advance CurPtr to the end of the preprocessing word.
+      CurPtr += strlen(PreprocessorDirs[ID].Word);
+      return true;
+    }
+
+  PrintFatalError("Unsupported preprocessing token in "
+                  "prepEatPreprocessorDirective()");
+  return false;
+}
+
+tgtok::TokKind TGLexer::lexPreprocessor(
+    tgtok::TokKind Kind, bool ReturnNextLiveToken) {
+
+  // We must be looking at a preprocessing directive.  Eat it!
+  if (!prepEatPreprocessorDirective(Kind))
+    PrintFatalError("lexPreprocessor() called for unknown "
+                    "preprocessor directive");
+
+  if (Kind == tgtok::Ifdef) {
+    StringRef MacroName = prepLexMacroName();
+    if (MacroName.empty())
+      return ReturnError(TokStart, "Expected macro name after #ifdef");
+
+    bool MacroIsDefined = DefinedMacros.count(MacroName) != 0;
+
+    // Regardless of whether we are processing tokens or not,
+    // we put the #ifdef control on stack.
+    PrepIncludeStack.back()->push_back(
+        {Kind, MacroIsDefined, SMLoc::getFromPointer(TokStart)});
+
+    if (!prepSkipDirectiveEnd())
+      return ReturnError(CurPtr,
+                         "Only comments are supported after #ifdef NAME");
+
+    // If we were not processing tokens before this #ifdef,
+    // then just return back to the lines skipping code.
+    if (!ReturnNextLiveToken)
+      return Kind;
+
+    // If we were processing tokens before this #ifdef,
+    // and the macro is defined, then just return the next token.
+    if (MacroIsDefined)
+      return LexToken();
+
+    // We were processing tokens before this #ifdef, and the macro
+    // is not defined, so we have to start skipping the lines.
+    // If the skipping is successful, it will return the token following
+    // either #else or #endif corresponding to this #ifdef.
+    if (prepSkipRegion(ReturnNextLiveToken))
+      return LexToken();
+
+    return tgtok::Error;
+  } else if (Kind == tgtok::Else) {
+    // Check if this #else is correct before calling prepSkipDirectiveEnd(),
+    // which will move CurPtr away from the beginning of #else.
+    if (PrepIncludeStack.back()->empty())
+      return ReturnError(TokStart, "#else without #ifdef");
+
+    auto &IfdefEntry = PrepIncludeStack.back()->back();
+
+    if (IfdefEntry.Kind != tgtok::Ifdef) {
+      PrintError(TokStart, "double #else");
+      return ReturnError(IfdefEntry.SrcPos, "Previous #else is here");
+    }
+
+    // Replace the corresponding #ifdef's control with its negation
+    // on the control stack.
+    PrepIncludeStack.back()->pop_back();
+    PrepIncludeStack.back()->push_back(
+        {Kind, !IfdefEntry.IsDefined, SMLoc::getFromPointer(TokStart)});
+
+    if (!prepSkipDirectiveEnd())
+      return ReturnError(CurPtr, "Only comments are supported after #else");
+
+    // If we were processing tokens before this #else,
+    // we have to start skipping lines until the matching #endif.
+    if (ReturnNextLiveToken) {
+      if (prepSkipRegion(ReturnNextLiveToken))
+        return LexToken();
+
+      return tgtok::Error;
+    }
+
+    // Return to the lines skipping code.
+    return Kind;
+  } else if (Kind == tgtok::Endif) {
+    // Check if this #endif is correct before calling prepSkipDirectiveEnd(),
+    // which will move CurPtr away from the beginning of #endif.
+    if (PrepIncludeStack.back()->empty())
+      return ReturnError(TokStart, "#endif without #ifdef");
+
+    auto &IfdefOrElseEntry = PrepIncludeStack.back()->back();
+
+    if (IfdefOrElseEntry.Kind != tgtok::Ifdef &&
+        IfdefOrElseEntry.Kind != tgtok::Else) {
+      PrintFatalError("Invalid preprocessor control on the stack");
+      return tgtok::Error;
+    }
+
+    if (!prepSkipDirectiveEnd())
+      return ReturnError(CurPtr, "Only comments are supported after #endif");
+
+    PrepIncludeStack.back()->pop_back();
+
+    // If we were processing tokens before this #endif, then
+    // we should continue it.
+    if (ReturnNextLiveToken) {
+      return LexToken();
+    }
+
+    // Return to the lines skipping code.
+    return Kind;
+  } else if (Kind == tgtok::Define) {
+    StringRef MacroName = prepLexMacroName();
+    if (MacroName.empty())
+      return ReturnError(TokStart, "Expected macro name after #define");
+
+    if (!DefinedMacros.insert(MacroName).second)
+      PrintWarning(getLoc(),
+                   "Duplicate definition of macro: " + Twine(MacroName));
+
+    if (!prepSkipDirectiveEnd())
+      return ReturnError(CurPtr,
+                         "Only comments are supported after #define NAME");
+
+    if (!ReturnNextLiveToken) {
+      PrintFatalError("#define must be ignored during the lines skipping");
+      return tgtok::Error;
+    }
+
+    return LexToken();
+  }
+
+  PrintFatalError("Preprocessing directive is not supported");
+  return tgtok::Error;
+}
+
+bool TGLexer::prepSkipRegion(bool MustNeverBeFalse) {
+  if (!MustNeverBeFalse)
+    PrintFatalError("Invalid recursion.");
+
+  do {
+    // Skip all symbols to the line end.
+    prepSkipToLineEnd();
+
+    // Find the first non-whitespace symbol in the next line(s).
+    if (!prepSkipLineBegin())
+      return false;
+
+    // If the first non-blank/comment symbol on the line is '#',
+    // it may be a start of preprocessing directive.
+    //
+    // If it is not '#' just go to the next line.
+    if (*CurPtr == '#')
+      ++CurPtr;
+    else
+      continue;
+
+    tgtok::TokKind Kind = prepIsDirective();
+
+    // If we did not find a preprocessing directive or it is #define,
+    // then just skip to the next line.  We do not have to do anything
+    // for #define in the line-skipping mode.
+    if (Kind == tgtok::Error || Kind == tgtok::Define)
+      continue;
+
+    tgtok::TokKind ProcessedKind = lexPreprocessor(Kind, false);
+
+    // If lexPreprocessor() encountered an error during lexing this
+    // preprocessor idiom, then return false to the calling lexPreprocessor().
+    // This will force tgtok::Error to be returned to the tokens processing.
+    if (ProcessedKind == tgtok::Error)
+      return false;
+
+    if (Kind != ProcessedKind)
+      PrintFatalError("prepIsDirective() and lexPreprocessor() "
+                      "returned different token kinds");
+
+    // If this preprocessing directive enables tokens processing,
+    // then return to the lexPreprocessor() and get to the next token.
+    // We can move from line-skipping mode to processing tokens only
+    // due to #else or #endif.
+    if (prepIsProcessingEnabled()) {
+      if (Kind != tgtok::Else && Kind != tgtok::Endif) {
+        PrintFatalError("Tokens processing was enabled by an unexpected "
+                        "preprocessing directive");
+        return false;
+      }
+
+      return true;
+    }
+  } while (CurPtr != CurBuf.end());
+
+  // We have reached the end of the file, but never left the lines-skipping
+  // mode.  This means there is no matching #endif.
+  prepReportPreprocessorStackError();
+  return false;
+}
+
+StringRef TGLexer::prepLexMacroName() {
+  // Skip whitespaces between the preprocessing directive and the macro name.
+  while (*CurPtr == ' ' || *CurPtr == '\t')
+    ++CurPtr;
+
+  TokStart = CurPtr;
+  // Macro names start with [a-zA-Z_].
+  if (*CurPtr != '_' && !isalpha(*CurPtr))
+    return "";
+
+  // Match the rest of the identifier regex: [0-9a-zA-Z_]*
+  while (isalpha(*CurPtr) || isdigit(*CurPtr) || *CurPtr == '_')
+    ++CurPtr;
+
+  return StringRef(TokStart, CurPtr - TokStart);
+}
+
+bool TGLexer::prepSkipLineBegin() {
+  while (CurPtr != CurBuf.end()) {
+    switch (*CurPtr) {
+    case ' ':
+    case '\t':
+    case '\n':
+    case '\r':
+      break;
+
+    case '/': {
+      int NextChar = peekNextChar(1);
+      if (NextChar == '*') {
+        // Skip C-style comment.
+        // Note that we do not care about skipping the C++-style comments.
+        // If the line contains "//", it may not contain any processable
+        // preprocessing directive.  Just return CurPtr pointing to
+        // the first '/' in this case.  We also do not care about
+        // incorrect symbols after the first '/' - we are in lines-skipping
+        // mode, so incorrect code is allowed to some extent.
+
+        // Set TokStart to the beginning of the comment to enable proper
+        // diagnostic printing in case of error in SkipCComment().
+        TokStart = CurPtr;
+
+        // CurPtr must point to '*' before call to SkipCComment().
+        ++CurPtr;
+        if (SkipCComment())
+          return false;
+      } else {
+        // CurPtr points to the non-whitespace '/'.
+        return true;
+      }
+
+      // We must not increment CurPtr after the comment was lexed.
+      continue;
+    }
+
+    default:
+      return true;
+    }
+
+    ++CurPtr;
+  }
+
+  // We have reached the end of the file.  Return to the lines skipping
+  // code, and allow it to handle the EOF as needed.
+  return true;
+}
+
+bool TGLexer::prepSkipDirectiveEnd() {
+  while (CurPtr != CurBuf.end()) {
+    switch (*CurPtr) {
+    case ' ':
+    case '\t':
+      break;
+
+    case '\n':
+    case '\r':
+      return true;
+
+    case '/': {
+      int NextChar = peekNextChar(1);
+      if (NextChar == '/') {
+        // Skip C++-style comment.
+        // We may just return true now, but let's skip to the line/buffer end
+        // to simplify the method specification.
+        ++CurPtr;
+        SkipBCPLComment();
+      } else if (NextChar == '*') {
+        // When we are skipping C-style comment at the end of a preprocessing
+        // directive, we can skip several lines.  If any meaningful TD token
+        // follows the end of the C-style comment on the same line, it will
+        // be considered as an invalid usage of TD token.
+        // For example, we want to forbid usages like this one:
+        //     #define MACRO class Class {}
+        // But with C-style comments we also disallow the following:
+        //     #define MACRO /* This macro is used
+        //                      to ... */ class Class {}
+        // One can argue that this should be allowed, but it does not seem
+        // to be worth of the complication.  Moreover, this matches
+        // the C preprocessor behavior.
+
+        // Set TokStart to the beginning of the comment to enable proper
+        // diagnostic printer in case of error in SkipCComment().
+        TokStart = CurPtr;
+        ++CurPtr;
+        if (SkipCComment())
+          return false;
+      } else {
+        TokStart = CurPtr;
+        PrintError(CurPtr, "Unexpected character");
+        return false;
+      }
+
+      // We must not increment CurPtr after the comment was lexed.
+      continue;
+    }
+
+    default:
+      // Do not allow any non-whitespaces after the directive.
+      TokStart = CurPtr;
+      return false;
+    }
+
+    ++CurPtr;
+  }
+
+  return true;
+}
+
+void TGLexer::prepSkipToLineEnd() {
+  while (*CurPtr != '\n' && *CurPtr != '\r' && CurPtr != CurBuf.end())
+    ++CurPtr;
+}
+
+bool TGLexer::prepIsProcessingEnabled() {
+  for (auto I = PrepIncludeStack.back()->rbegin(),
+            E = PrepIncludeStack.back()->rend();
+       I != E; ++I) {
+    if (!I->IsDefined)
+      return false;
+  }
+
+  return true;
+}
+
+void TGLexer::prepReportPreprocessorStackError() {
+  if (PrepIncludeStack.back()->empty())
+    PrintFatalError("prepReportPreprocessorStackError() called with "
+                    "empty control stack");
+
+  auto &PrepControl = PrepIncludeStack.back()->back();
+  PrintError(CurBuf.end(), "Reached EOF without matching #endif");
+  PrintError(PrepControl.SrcPos, "The latest preprocessor control is here");
+
+  TokStart = CurPtr;
+}