src/llvm/lib/TableGen/TGLexer.cpp

   1 //===- TGLexer.cpp - Lexer for TableGen -----------------------------------===//
   2 //
   3 //                     The LLVM Compiler Infrastructure
   4 //
   5 // This file is distributed under the University of Illinois Open Source
   6 // License. See LICENSE.TXT for details.
   7 //
   8 //===----------------------------------------------------------------------===//
   9 //
  10 // Implement the Lexer for TableGen.
  11 //
  12 //===----------------------------------------------------------------------===//
  13
  14 #include "TGLexer.h"
  15 #include "llvm/ADT/StringSwitch.h"
  16 #include "llvm/ADT/Twine.h"
  17 #include "llvm/Config/config.h" // for strtoull()/strtoll() define
  18 #include "llvm/Support/MemoryBuffer.h"
  19 #include "llvm/Support/SourceMgr.h"
  20 #include "llvm/TableGen/Error.h"
  21 #include <cctype>
  22 #include <cerrno>
  23 #include <cstdio>
  24 #include <cstdlib>
  25 #include <cstring>
  26
  27 using namespace llvm;
  28
  29 TGLexer::TGLexer(SourceMgr &SM) : SrcMgr(SM) {
  30   CurBuffer = SrcMgr.getMainFileID();
  31   CurBuf = SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer();
  32   CurPtr = CurBuf.begin();
  33   TokStart = nullptr;
  34 }
  35
  36 SMLoc TGLexer::getLoc() const {
  37   return SMLoc::getFromPointer(TokStart);
  38 }
  39
  40 /// ReturnError - Set the error to the specified string at the specified
  41 /// location.  This is defined to always return tgtok::Error.
  42 tgtok::TokKind TGLexer::ReturnError(const char *Loc, const Twine &Msg) {
  43   PrintError(Loc, Msg);
  44   return tgtok::Error;
  45 }
  46
  47 int TGLexer::getNextChar() {
  48   char CurChar = *CurPtr++;
  49   switch (CurChar) {
  50   default:
  51     return (unsigned char)CurChar;
  52   case 0: {
  53     // A nul character in the stream is either the end of the current buffer or
  54     // a random nul in the file.  Disambiguate that here.
  55     if (CurPtr-1 != CurBuf.end())
  56       return 0;  // Just whitespace.
  57
  58     // If this is the end of an included file, pop the parent file off the
  59     // include stack.
  60     SMLoc ParentIncludeLoc = SrcMgr.getParentIncludeLoc(CurBuffer);
  61     if (ParentIncludeLoc != SMLoc()) {
  62       CurBuffer = SrcMgr.FindBufferContainingLoc(ParentIncludeLoc);
  63       CurBuf = SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer();
  64       CurPtr = ParentIncludeLoc.getPointer();
  65       return getNextChar();
  66     }
  67
  68     // Otherwise, return end of file.
  69     --CurPtr;  // Another call to lex will return EOF again.
  70     return EOF;
  71   }
  72   case '\n':
  73   case '\r':
  74     // Handle the newline character by ignoring it and incrementing the line
  75     // count.  However, be careful about 'dos style' files with \n\r in them.
  76     // Only treat a \n\r or \r\n as a single line.
  77     if ((*CurPtr == '\n' || (*CurPtr == '\r')) &&
  78         *CurPtr != CurChar)
  79       ++CurPtr;  // Eat the two char newline sequence.
  80     return '\n';
  81   }
  82 }
  83
  84 int TGLexer::peekNextChar(int Index) {
  85   return *(CurPtr + Index);
  86 }
  87
  88 tgtok::TokKind TGLexer::LexToken() {
  89   TokStart = CurPtr;
  90   // This always consumes at least one character.
  91   int CurChar = getNextChar();
  92
  93   switch (CurChar) {
  94   default:
  95     // Handle letters: [a-zA-Z_]
  96     if (isalpha(CurChar) || CurChar == '_')
  97       return LexIdentifier();
  98
  99     // Unknown character, emit an error.
 100     return ReturnError(TokStart, "Unexpected character");
 101   case EOF: return tgtok::Eof;
 102   case ':': return tgtok::colon;
 103   case ';': return tgtok::semi;
 104   case '.': return tgtok::period;
 105   case ',': return tgtok::comma;
 106   case '<': return tgtok::less;
 107   case '>': return tgtok::greater;
 108   case ']': return tgtok::r_square;
 109   case '{': return tgtok::l_brace;
 110   case '}': return tgtok::r_brace;
 111   case '(': return tgtok::l_paren;
 112   case ')': return tgtok::r_paren;
 113   case '=': return tgtok::equal;
 114   case '?': return tgtok::question;
 115   case '#': return tgtok::paste;
 116
 117   case 0:
 118   case ' ':
 119   case '\t':
 120   case '\n':
 121   case '\r':
 122     // Ignore whitespace.
 123     return LexToken();
 124   case '/':
 125     // If this is the start of a // comment, skip until the end of the line or
 126     // the end of the buffer.
 127     if (*CurPtr == '/')
 128       SkipBCPLComment();
 129     else if (*CurPtr == '*') {
 130       if (SkipCComment())
 131         return tgtok::Error;
 132     } else // Otherwise, this is an error.
 133       return ReturnError(TokStart, "Unexpected character");
 134     return LexToken();
 135   case '-': case '+':
 136   case '0': case '1': case '2': case '3': case '4': case '5': case '6':
 137   case '7': case '8': case '9': {
 138     int NextChar = 0;
 139     if (isdigit(CurChar)) {
 140       // Allow identifiers to start with a number if it is followed by
 141       // an identifier.  This can happen with paste operations like
 142       // foo#8i.
 143       int i = 0;
 144       do {
 145         NextChar = peekNextChar(i++);
 146       } while (isdigit(NextChar));
 147
 148       if (NextChar == 'x' || NextChar == 'b') {
 149         // If this is [0-9]b[01] or [0-9]x[0-9A-fa-f] this is most
 150         // likely a number.
 151         int NextNextChar = peekNextChar(i);
 152         switch (NextNextChar) {
 153         default:
 154           break;
 155         case '0': case '1':
 156           if (NextChar == 'b')
 157             return LexNumber();
 158           // Fallthrough
 159         case '2': case '3': case '4': case '5':
 160         case '6': case '7': case '8': case '9':
 161         case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
 162         case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
 163           if (NextChar == 'x')
 164             return LexNumber();
 165           break;
 166         }
 167       }
 168     }
 169
 170     if (isalpha(NextChar) || NextChar == '_')
 171       return LexIdentifier();
 172
 173     return LexNumber();
 174   }
 175   case '"': return LexString();
 176   case '$': return LexVarName();
 177   case '[': return LexBracket();
 178   case '!': return LexExclaim();
 179   }
 180 }
 181
 182 /// LexString - Lex "[^"]*"
 183 tgtok::TokKind TGLexer::LexString() {
 184   const char *StrStart = CurPtr;
 185
 186   CurStrVal = "";
 187
 188   while (*CurPtr != '"') {
 189     // If we hit the end of the buffer, report an error.
 190     if (*CurPtr == 0 && CurPtr == CurBuf.end())
 191       return ReturnError(StrStart, "End of file in string literal");
 192
 193     if (*CurPtr == '\n' || *CurPtr == '\r')
 194       return ReturnError(StrStart, "End of line in string literal");
 195
 196     if (*CurPtr != '\\') {
 197       CurStrVal += *CurPtr++;
 198       continue;
 199     }
 200
 201     ++CurPtr;
 202
 203     switch (*CurPtr) {
 204     case '\\': case '\'': case '"':
 205       // These turn into their literal character.
 206       CurStrVal += *CurPtr++;
 207       break;
 208     case 't':
 209       CurStrVal += '\t';
 210       ++CurPtr;
 211       break;
 212     case 'n':
 213       CurStrVal += '\n';
 214       ++CurPtr;
 215       break;
 216
 217     case '\n':
 218     case '\r':
 219       return ReturnError(CurPtr, "escaped newlines not supported in tblgen");
 220
 221     // If we hit the end of the buffer, report an error.
 222     case '\0':
 223       if (CurPtr == CurBuf.end())
 224         return ReturnError(StrStart, "End of file in string literal");
 225       // FALL THROUGH
 226     default:
 227       return ReturnError(CurPtr, "invalid escape in string literal");
 228     }
 229   }
 230
 231   ++CurPtr;
 232   return tgtok::StrVal;
 233 }
 234
 235 tgtok::TokKind TGLexer::LexVarName() {
 236   if (!isalpha(CurPtr[0]) && CurPtr[0] != '_')
 237     return ReturnError(TokStart, "Invalid variable name");
 238
 239   // Otherwise, we're ok, consume the rest of the characters.
 240   const char *VarNameStart = CurPtr++;
 241
 242   while (isalpha(*CurPtr) || isdigit(*CurPtr) || *CurPtr == '_')
 243     ++CurPtr;
 244
 245   CurStrVal.assign(VarNameStart, CurPtr);
 246   return tgtok::VarName;
 247 }
 248
 249
 250 tgtok::TokKind TGLexer::LexIdentifier() {
 251   // The first letter is [a-zA-Z_#].
 252   const char *IdentStart = TokStart;
 253
 254   // Match the rest of the identifier regex: [0-9a-zA-Z_#]*
 255   while (isalpha(*CurPtr) || isdigit(*CurPtr) || *CurPtr == '_')
 256     ++CurPtr;
 257
 258   // Check to see if this identifier is a keyword.
 259   StringRef Str(IdentStart, CurPtr-IdentStart);
 260
 261   if (Str == "include") {
 262     if (LexInclude()) return tgtok::Error;
 263     return Lex();
 264   }
 265
 266   tgtok::TokKind Kind = StringSwitch<tgtok::TokKind>(Str)
 267     .Case("int", tgtok::Int)
 268     .Case("bit", tgtok::Bit)
 269     .Case("bits", tgtok::Bits)
 270     .Case("string", tgtok::String)
 271     .Case("list", tgtok::List)
 272     .Case("code", tgtok::Code)
 273     .Case("dag", tgtok::Dag)
 274     .Case("class", tgtok::Class)
 275     .Case("def", tgtok::Def)
 276     .Case("foreach", tgtok::Foreach)
 277     .Case("defm", tgtok::Defm)
 278     .Case("multiclass", tgtok::MultiClass)
 279     .Case("field", tgtok::Field)
 280     .Case("let", tgtok::Let)
 281     .Case("in", tgtok::In)
 282     .Default(tgtok::Id);
 283
 284   if (Kind == tgtok::Id)
 285     CurStrVal.assign(Str.begin(), Str.end());
 286   return Kind;
 287 }
 288
 289 /// LexInclude - We just read the "include" token.  Get the string token that
 290 /// comes next and enter the include.
 291 bool TGLexer::LexInclude() {
 292   // The token after the include must be a string.
 293   tgtok::TokKind Tok = LexToken();
 294   if (Tok == tgtok::Error) return true;
 295   if (Tok != tgtok::StrVal) {
 296     PrintError(getLoc(), "Expected filename after include");
 297     return true;
 298   }
 299
 300   // Get the string.
 301   std::string Filename = CurStrVal;
 302   std::string IncludedFile;
 303
 304
 305   CurBuffer = SrcMgr.AddIncludeFile(Filename, SMLoc::getFromPointer(CurPtr),
 306                                     IncludedFile);
 307   if (!CurBuffer) {
 308     PrintError(getLoc(), "Could not find include file '" + Filename + "'");
 309     return true;
 310   }
 311
 312   DependenciesMapTy::const_iterator Found = Dependencies.find(IncludedFile);
 313   if (Found != Dependencies.end()) {
 314     PrintError(getLoc(),
 315                "File '" + IncludedFile + "' has already been included.");
 316     SrcMgr.PrintMessage(Found->second, SourceMgr::DK_Note,
 317                         "previously included here");
 318     return true;
 319   }
 320   Dependencies.insert(std::make_pair(IncludedFile, getLoc()));
 321   // Save the line number and lex buffer of the includer.
 322   CurBuf = SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer();
 323   CurPtr = CurBuf.begin();
 324   return false;
 325 }
 326
 327 void TGLexer::SkipBCPLComment() {
 328   ++CurPtr;  // skip the second slash.
 329   while (1) {
 330     switch (*CurPtr) {
 331     case '\n':
 332     case '\r':
 333       return;  // Newline is end of comment.
 334     case 0:
 335       // If this is the end of the buffer, end the comment.
 336       if (CurPtr == CurBuf.end())
 337         return;
 338       break;
 339     }
 340     // Otherwise, skip the character.
 341     ++CurPtr;
 342   }
 343 }
 344
 345 /// SkipCComment - This skips C-style /**/ comments.  The only difference from C
 346 /// is that we allow nesting.
 347 bool TGLexer::SkipCComment() {
 348   ++CurPtr;  // skip the star.
 349   unsigned CommentDepth = 1;
 350
 351   while (1) {
 352     int CurChar = getNextChar();
 353     switch (CurChar) {
 354     case EOF:
 355       PrintError(TokStart, "Unterminated comment!");
 356       return true;
 357     case '*':
 358       // End of the comment?
 359       if (CurPtr[0] != '/') break;
 360
 361       ++CurPtr;   // End the */.
 362       if (--CommentDepth == 0)
 363         return false;
 364       break;
 365     case '/':
 366       // Start of a nested comment?
 367       if (CurPtr[0] != '*') break;
 368       ++CurPtr;
 369       ++CommentDepth;
 370       break;
 371     }
 372   }
 373 }
 374
 375 /// LexNumber - Lex:
 376 ///    [-+]?[0-9]+
 377 ///    0x[0-9a-fA-F]+
 378 ///    0b[01]+
 379 tgtok::TokKind TGLexer::LexNumber() {
 380   if (CurPtr[-1] == '0') {
 381     if (CurPtr[0] == 'x') {
 382       ++CurPtr;
 383       const char *NumStart = CurPtr;
 384       while (isxdigit(CurPtr[0]))
 385         ++CurPtr;
 386
 387       // Requires at least one hex digit.
 388       if (CurPtr == NumStart)
 389         return ReturnError(TokStart, "Invalid hexadecimal number");
 390
 391       errno = 0;
 392       CurIntVal = strtoll(NumStart, nullptr, 16);
 393       if (errno == EINVAL)
 394         return ReturnError(TokStart, "Invalid hexadecimal number");
 395       if (errno == ERANGE) {
 396         errno = 0;
 397         CurIntVal = (int64_t)strtoull(NumStart, nullptr, 16);
 398         if (errno == EINVAL)
 399           return ReturnError(TokStart, "Invalid hexadecimal number");
 400         if (errno == ERANGE)
 401           return ReturnError(TokStart, "Hexadecimal number out of range");
 402       }
 403       return tgtok::IntVal;
 404     } else if (CurPtr[0] == 'b') {
 405       ++CurPtr;
 406       const char *NumStart = CurPtr;
 407       while (CurPtr[0] == '0' || CurPtr[0] == '1')
 408         ++CurPtr;
 409
 410       // Requires at least one binary digit.
 411       if (CurPtr == NumStart)
 412         return ReturnError(CurPtr-2, "Invalid binary number");
 413       CurIntVal = strtoll(NumStart, nullptr, 2);
 414       return tgtok::BinaryIntVal;
 415     }
 416   }
 417
 418   // Check for a sign without a digit.
 419   if (!isdigit(CurPtr[0])) {
 420     if (CurPtr[-1] == '-')
 421       return tgtok::minus;
 422     else if (CurPtr[-1] == '+')
 423       return tgtok::plus;
 424   }
 425
 426   while (isdigit(CurPtr[0]))
 427     ++CurPtr;
 428   CurIntVal = strtoll(TokStart, nullptr, 10);
 429   return tgtok::IntVal;
 430 }
 431
 432 /// LexBracket - We just read '['.  If this is a code block, return it,
 433 /// otherwise return the bracket.  Match: '[' and '[{ ( [^}]+ | }[^]] )* }]'
 434 tgtok::TokKind TGLexer::LexBracket() {
 435   if (CurPtr[0] != '{')
 436     return tgtok::l_square;
 437   ++CurPtr;
 438   const char *CodeStart = CurPtr;
 439   while (1) {
 440     int Char = getNextChar();
 441     if (Char == EOF) break;
 442
 443     if (Char != '}') continue;
 444
 445     Char = getNextChar();
 446     if (Char == EOF) break;
 447     if (Char == ']') {
 448       CurStrVal.assign(CodeStart, CurPtr-2);
 449       return tgtok::CodeFragment;
 450     }
 451   }
 452
 453   return ReturnError(CodeStart-2, "Unterminated Code Block");
 454 }
 455
 456 /// LexExclaim - Lex '!' and '![a-zA-Z]+'.
 457 tgtok::TokKind TGLexer::LexExclaim() {
 458   if (!isalpha(*CurPtr))
 459     return ReturnError(CurPtr - 1, "Invalid \"!operator\"");
 460
 461   const char *Start = CurPtr++;
 462   while (isalpha(*CurPtr))
 463     ++CurPtr;
 464
 465   // Check to see which operator this is.
 466   tgtok::TokKind Kind =
 467     StringSwitch<tgtok::TokKind>(StringRef(Start, CurPtr - Start))
 468     .Case("eq", tgtok::XEq)
 469     .Case("if", tgtok::XIf)
 470     .Case("head", tgtok::XHead)
 471     .Case("tail", tgtok::XTail)
 472     .Case("con", tgtok::XConcat)
 473     .Case("add", tgtok::XADD)
 474     .Case("and", tgtok::XAND)
 475     .Case("shl", tgtok::XSHL)
 476     .Case("sra", tgtok::XSRA)
 477     .Case("srl", tgtok::XSRL)
 478     .Case("cast", tgtok::XCast)
 479     .Case("empty", tgtok::XEmpty)
 480     .Case("subst", tgtok::XSubst)
 481     .Case("foreach", tgtok::XForEach)
 482     .Case("listconcat", tgtok::XListConcat)
 483     .Case("strconcat", tgtok::XStrConcat)
 484     .Default(tgtok::Error);
 485
 486   return Kind != tgtok::Error ? Kind : ReturnError(Start-1, "Unknown operator");
 487 }
 488