xref: /openbsd/gnu/llvm/llvm/lib/TableGen/TGLexer.cpp (revision d415bd75)
1 //===- TGLexer.cpp - Lexer for TableGen -----------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // Implement the Lexer for TableGen.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "TGLexer.h"
14 #include "llvm/ADT/ArrayRef.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/Compiler.h"
19 #include "llvm/Support/MemoryBuffer.h"
20 #include "llvm/Support/SourceMgr.h"
21 #include "llvm/TableGen/Error.h"
22 #include <algorithm>
23 #include <cctype>
24 #include <cerrno>
25 #include <cstdint>
26 #include <cstdio>
27 #include <cstdlib>
28 #include <cstring>
29 
30 using namespace llvm;
31 
32 namespace {
33 // A list of supported preprocessing directives with their
34 // internal token kinds and names.
35 struct {
36   tgtok::TokKind Kind;
37   const char *Word;
38 } PreprocessorDirs[] = {
39   { tgtok::Ifdef, "ifdef" },
40   { tgtok::Ifndef, "ifndef" },
41   { tgtok::Else, "else" },
42   { tgtok::Endif, "endif" },
43   { tgtok::Define, "define" }
44 };
45 } // end anonymous namespace
46 
TGLexer(SourceMgr & SM,ArrayRef<std::string> Macros)47 TGLexer::TGLexer(SourceMgr &SM, ArrayRef<std::string> Macros) : SrcMgr(SM) {
48   CurBuffer = SrcMgr.getMainFileID();
49   CurBuf = SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer();
50   CurPtr = CurBuf.begin();
51   TokStart = nullptr;
52 
53   // Pretend that we enter the "top-level" include file.
54   PrepIncludeStack.push_back(
55       std::make_unique<std::vector<PreprocessorControlDesc>>());
56 
57   // Put all macros defined in the command line into the DefinedMacros set.
58   for (const std::string &MacroName : Macros)
59     DefinedMacros.insert(MacroName);
60 }
61 
getLoc() const62 SMLoc TGLexer::getLoc() const {
63   return SMLoc::getFromPointer(TokStart);
64 }
65 
getLocRange() const66 SMRange TGLexer::getLocRange() const {
67   return {getLoc(), SMLoc::getFromPointer(CurPtr)};
68 }
69 
70 /// ReturnError - Set the error to the specified string at the specified
71 /// location.  This is defined to always return tgtok::Error.
ReturnError(SMLoc Loc,const Twine & Msg)72 tgtok::TokKind TGLexer::ReturnError(SMLoc Loc, const Twine &Msg) {
73   PrintError(Loc, Msg);
74   return tgtok::Error;
75 }
76 
ReturnError(const char * Loc,const Twine & Msg)77 tgtok::TokKind TGLexer::ReturnError(const char *Loc, const Twine &Msg) {
78   return ReturnError(SMLoc::getFromPointer(Loc), Msg);
79 }
80 
processEOF()81 bool TGLexer::processEOF() {
82   SMLoc ParentIncludeLoc = SrcMgr.getParentIncludeLoc(CurBuffer);
83   if (ParentIncludeLoc != SMLoc()) {
84     // If prepExitInclude() detects a problem with the preprocessing
85     // control stack, it will return false.  Pretend that we reached
86     // the final EOF and stop lexing more tokens by returning false
87     // to LexToken().
88     if (!prepExitInclude(false))
89       return false;
90 
91     CurBuffer = SrcMgr.FindBufferContainingLoc(ParentIncludeLoc);
92     CurBuf = SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer();
93     CurPtr = ParentIncludeLoc.getPointer();
94     // Make sure TokStart points into the parent file's buffer.
95     // LexToken() assigns to it before calling getNextChar(),
96     // so it is pointing into the included file now.
97     TokStart = CurPtr;
98     return true;
99   }
100 
101   // Pretend that we exit the "top-level" include file.
102   // Note that in case of an error (e.g. control stack imbalance)
103   // the routine will issue a fatal error.
104   prepExitInclude(true);
105   return false;
106 }
107 
getNextChar()108 int TGLexer::getNextChar() {
109   char CurChar = *CurPtr++;
110   switch (CurChar) {
111   default:
112     return (unsigned char)CurChar;
113 
114   case 0: {
115     // A NUL character in the stream is either the end of the current buffer or
116     // a spurious NUL in the file.  Disambiguate that here.
117     if (CurPtr - 1 == CurBuf.end()) {
118       --CurPtr; // Arrange for another call to return EOF again.
119       return EOF;
120     }
121     PrintError(getLoc(),
122                "NUL character is invalid in source; treated as space");
123     return ' ';
124   }
125 
126   case '\n':
127   case '\r':
128     // Handle the newline character by ignoring it and incrementing the line
129     // count.  However, be careful about 'dos style' files with \n\r in them.
130     // Only treat a \n\r or \r\n as a single line.
131     if ((*CurPtr == '\n' || (*CurPtr == '\r')) &&
132         *CurPtr != CurChar)
133       ++CurPtr;  // Eat the two char newline sequence.
134     return '\n';
135   }
136 }
137 
peekNextChar(int Index) const138 int TGLexer::peekNextChar(int Index) const {
139   return *(CurPtr + Index);
140 }
141 
LexToken(bool FileOrLineStart)142 tgtok::TokKind TGLexer::LexToken(bool FileOrLineStart) {
143   TokStart = CurPtr;
144   // This always consumes at least one character.
145   int CurChar = getNextChar();
146 
147   switch (CurChar) {
148   default:
149     // Handle letters: [a-zA-Z_]
150     if (isalpha(CurChar) || CurChar == '_')
151       return LexIdentifier();
152 
153     // Unknown character, emit an error.
154     return ReturnError(TokStart, "Unexpected character");
155   case EOF:
156     // Lex next token, if we just left an include file.
157     // Note that leaving an include file means that the next
158     // symbol is located at the end of the 'include "..."'
159     // construct, so LexToken() is called with default
160     // false parameter.
161     if (processEOF())
162       return LexToken();
163 
164     // Return EOF denoting the end of lexing.
165     return tgtok::Eof;
166 
167   case ':': return tgtok::colon;
168   case ';': return tgtok::semi;
169   case ',': return tgtok::comma;
170   case '<': return tgtok::less;
171   case '>': return tgtok::greater;
172   case ']': return tgtok::r_square;
173   case '{': return tgtok::l_brace;
174   case '}': return tgtok::r_brace;
175   case '(': return tgtok::l_paren;
176   case ')': return tgtok::r_paren;
177   case '=': return tgtok::equal;
178   case '?': return tgtok::question;
179   case '#':
180     if (FileOrLineStart) {
181       tgtok::TokKind Kind = prepIsDirective();
182       if (Kind != tgtok::Error)
183         return lexPreprocessor(Kind);
184     }
185 
186     return tgtok::paste;
187 
188   // The period is a separate case so we can recognize the "..."
189   // range punctuator.
190   case '.':
191     if (peekNextChar(0) == '.') {
192       ++CurPtr; // Eat second dot.
193       if (peekNextChar(0) == '.') {
194         ++CurPtr; // Eat third dot.
195         return tgtok::dotdotdot;
196       }
197       return ReturnError(TokStart, "Invalid '..' punctuation");
198     }
199     return tgtok::dot;
200 
201   case '\r':
202     PrintFatalError("getNextChar() must never return '\r'");
203     return tgtok::Error;
204 
205   case ' ':
206   case '\t':
207     // Ignore whitespace.
208     return LexToken(FileOrLineStart);
209   case '\n':
210     // Ignore whitespace, and identify the new line.
211     return LexToken(true);
212   case '/':
213     // If this is the start of a // comment, skip until the end of the line or
214     // the end of the buffer.
215     if (*CurPtr == '/')
216       SkipBCPLComment();
217     else if (*CurPtr == '*') {
218       if (SkipCComment())
219         return tgtok::Error;
220     } else // Otherwise, this is an error.
221       return ReturnError(TokStart, "Unexpected character");
222     return LexToken(FileOrLineStart);
223   case '-': case '+':
224   case '0': case '1': case '2': case '3': case '4': case '5': case '6':
225   case '7': case '8': case '9': {
226     int NextChar = 0;
227     if (isdigit(CurChar)) {
228       // Allow identifiers to start with a number if it is followed by
229       // an identifier.  This can happen with paste operations like
230       // foo#8i.
231       int i = 0;
232       do {
233         NextChar = peekNextChar(i++);
234       } while (isdigit(NextChar));
235 
236       if (NextChar == 'x' || NextChar == 'b') {
237         // If this is [0-9]b[01] or [0-9]x[0-9A-fa-f] this is most
238         // likely a number.
239         int NextNextChar = peekNextChar(i);
240         switch (NextNextChar) {
241         default:
242           break;
243         case '0': case '1':
244           if (NextChar == 'b')
245             return LexNumber();
246           [[fallthrough]];
247         case '2': case '3': case '4': case '5':
248         case '6': case '7': case '8': case '9':
249         case 'a': case 'b': case 'c': case 'd': case 'e': case 'f':
250         case 'A': case 'B': case 'C': case 'D': case 'E': case 'F':
251           if (NextChar == 'x')
252             return LexNumber();
253           break;
254         }
255       }
256     }
257 
258     if (isalpha(NextChar) || NextChar == '_')
259       return LexIdentifier();
260 
261     return LexNumber();
262   }
263   case '"': return LexString();
264   case '$': return LexVarName();
265   case '[': return LexBracket();
266   case '!': return LexExclaim();
267   }
268 }
269 
270 /// LexString - Lex "[^"]*"
LexString()271 tgtok::TokKind TGLexer::LexString() {
272   const char *StrStart = CurPtr;
273 
274   CurStrVal = "";
275 
276   while (*CurPtr != '"') {
277     // If we hit the end of the buffer, report an error.
278     if (*CurPtr == 0 && CurPtr == CurBuf.end())
279       return ReturnError(StrStart, "End of file in string literal");
280 
281     if (*CurPtr == '\n' || *CurPtr == '\r')
282       return ReturnError(StrStart, "End of line in string literal");
283 
284     if (*CurPtr != '\\') {
285       CurStrVal += *CurPtr++;
286       continue;
287     }
288 
289     ++CurPtr;
290 
291     switch (*CurPtr) {
292     case '\\': case '\'': case '"':
293       // These turn into their literal character.
294       CurStrVal += *CurPtr++;
295       break;
296     case 't':
297       CurStrVal += '\t';
298       ++CurPtr;
299       break;
300     case 'n':
301       CurStrVal += '\n';
302       ++CurPtr;
303       break;
304 
305     case '\n':
306     case '\r':
307       return ReturnError(CurPtr, "escaped newlines not supported in tblgen");
308 
309     // If we hit the end of the buffer, report an error.
310     case '\0':
311       if (CurPtr == CurBuf.end())
312         return ReturnError(StrStart, "End of file in string literal");
313       [[fallthrough]];
314     default:
315       return ReturnError(CurPtr, "invalid escape in string literal");
316     }
317   }
318 
319   ++CurPtr;
320   return tgtok::StrVal;
321 }
322 
LexVarName()323 tgtok::TokKind TGLexer::LexVarName() {
324   if (!isalpha(CurPtr[0]) && CurPtr[0] != '_')
325     return ReturnError(TokStart, "Invalid variable name");
326 
327   // Otherwise, we're ok, consume the rest of the characters.
328   const char *VarNameStart = CurPtr++;
329 
330   while (isalpha(*CurPtr) || isdigit(*CurPtr) || *CurPtr == '_')
331     ++CurPtr;
332 
333   CurStrVal.assign(VarNameStart, CurPtr);
334   return tgtok::VarName;
335 }
336 
LexIdentifier()337 tgtok::TokKind TGLexer::LexIdentifier() {
338   // The first letter is [a-zA-Z_].
339   const char *IdentStart = TokStart;
340 
341   // Match the rest of the identifier regex: [0-9a-zA-Z_]*
342   while (isalpha(*CurPtr) || isdigit(*CurPtr) || *CurPtr == '_')
343     ++CurPtr;
344 
345   // Check to see if this identifier is a reserved keyword.
346   StringRef Str(IdentStart, CurPtr-IdentStart);
347 
348   tgtok::TokKind Kind = StringSwitch<tgtok::TokKind>(Str)
349     .Case("int", tgtok::Int)
350     .Case("bit", tgtok::Bit)
351     .Case("bits", tgtok::Bits)
352     .Case("string", tgtok::String)
353     .Case("list", tgtok::List)
354     .Case("code", tgtok::Code)
355     .Case("dag", tgtok::Dag)
356     .Case("class", tgtok::Class)
357     .Case("def", tgtok::Def)
358     .Case("true", tgtok::TrueVal)
359     .Case("false", tgtok::FalseVal)
360     .Case("foreach", tgtok::Foreach)
361     .Case("defm", tgtok::Defm)
362     .Case("defset", tgtok::Defset)
363     .Case("multiclass", tgtok::MultiClass)
364     .Case("field", tgtok::Field)
365     .Case("let", tgtok::Let)
366     .Case("in", tgtok::In)
367     .Case("defvar", tgtok::Defvar)
368     .Case("include", tgtok::Include)
369     .Case("if", tgtok::If)
370     .Case("then", tgtok::Then)
371     .Case("else", tgtok::ElseKW)
372     .Case("assert", tgtok::Assert)
373     .Default(tgtok::Id);
374 
375   // A couple of tokens require special processing.
376   switch (Kind) {
377     case tgtok::Include:
378       if (LexInclude()) return tgtok::Error;
379       return Lex();
380     case tgtok::Id:
381       CurStrVal.assign(Str.begin(), Str.end());
382       break;
383     default:
384       break;
385   }
386 
387   return Kind;
388 }
389 
390 /// LexInclude - We just read the "include" token.  Get the string token that
391 /// comes next and enter the include.
LexInclude()392 bool TGLexer::LexInclude() {
393   // The token after the include must be a string.
394   tgtok::TokKind Tok = LexToken();
395   if (Tok == tgtok::Error) return true;
396   if (Tok != tgtok::StrVal) {
397     PrintError(getLoc(), "Expected filename after include");
398     return true;
399   }
400 
401   // Get the string.
402   std::string Filename = CurStrVal;
403   std::string IncludedFile;
404 
405   CurBuffer = SrcMgr.AddIncludeFile(Filename, SMLoc::getFromPointer(CurPtr),
406                                     IncludedFile);
407   if (!CurBuffer) {
408     PrintError(getLoc(), "Could not find include file '" + Filename + "'");
409     return true;
410   }
411 
412   Dependencies.insert(IncludedFile);
413   // Save the line number and lex buffer of the includer.
414   CurBuf = SrcMgr.getMemoryBuffer(CurBuffer)->getBuffer();
415   CurPtr = CurBuf.begin();
416 
417   PrepIncludeStack.push_back(
418       std::make_unique<std::vector<PreprocessorControlDesc>>());
419   return false;
420 }
421 
422 /// SkipBCPLComment - Skip over the comment by finding the next CR or LF.
423 /// Or we may end up at the end of the buffer.
SkipBCPLComment()424 void TGLexer::SkipBCPLComment() {
425   ++CurPtr;  // skip the second slash.
426   auto EOLPos = CurBuf.find_first_of("\r\n", CurPtr - CurBuf.data());
427   CurPtr = (EOLPos == StringRef::npos) ? CurBuf.end() : CurBuf.data() + EOLPos;
428 }
429 
430 /// SkipCComment - This skips C-style /**/ comments.  The only difference from C
431 /// is that we allow nesting.
SkipCComment()432 bool TGLexer::SkipCComment() {
433   ++CurPtr;  // skip the star.
434   unsigned CommentDepth = 1;
435 
436   while (true) {
437     int CurChar = getNextChar();
438     switch (CurChar) {
439     case EOF:
440       PrintError(TokStart, "Unterminated comment!");
441       return true;
442     case '*':
443       // End of the comment?
444       if (CurPtr[0] != '/') break;
445 
446       ++CurPtr;   // End the */.
447       if (--CommentDepth == 0)
448         return false;
449       break;
450     case '/':
451       // Start of a nested comment?
452       if (CurPtr[0] != '*') break;
453       ++CurPtr;
454       ++CommentDepth;
455       break;
456     }
457   }
458 }
459 
460 /// LexNumber - Lex:
461 ///    [-+]?[0-9]+
462 ///    0x[0-9a-fA-F]+
463 ///    0b[01]+
LexNumber()464 tgtok::TokKind TGLexer::LexNumber() {
465   if (CurPtr[-1] == '0') {
466     if (CurPtr[0] == 'x') {
467       ++CurPtr;
468       const char *NumStart = CurPtr;
469       while (isxdigit(CurPtr[0]))
470         ++CurPtr;
471 
472       // Requires at least one hex digit.
473       if (CurPtr == NumStart)
474         return ReturnError(TokStart, "Invalid hexadecimal number");
475 
476       errno = 0;
477       CurIntVal = strtoll(NumStart, nullptr, 16);
478       if (errno == EINVAL)
479         return ReturnError(TokStart, "Invalid hexadecimal number");
480       if (errno == ERANGE) {
481         errno = 0;
482         CurIntVal = (int64_t)strtoull(NumStart, nullptr, 16);
483         if (errno == EINVAL)
484           return ReturnError(TokStart, "Invalid hexadecimal number");
485         if (errno == ERANGE)
486           return ReturnError(TokStart, "Hexadecimal number out of range");
487       }
488       return tgtok::IntVal;
489     } else if (CurPtr[0] == 'b') {
490       ++CurPtr;
491       const char *NumStart = CurPtr;
492       while (CurPtr[0] == '0' || CurPtr[0] == '1')
493         ++CurPtr;
494 
495       // Requires at least one binary digit.
496       if (CurPtr == NumStart)
497         return ReturnError(CurPtr-2, "Invalid binary number");
498       CurIntVal = strtoll(NumStart, nullptr, 2);
499       return tgtok::BinaryIntVal;
500     }
501   }
502 
503   // Check for a sign without a digit.
504   if (!isdigit(CurPtr[0])) {
505     if (CurPtr[-1] == '-')
506       return tgtok::minus;
507     else if (CurPtr[-1] == '+')
508       return tgtok::plus;
509   }
510 
511   while (isdigit(CurPtr[0]))
512     ++CurPtr;
513   CurIntVal = strtoll(TokStart, nullptr, 10);
514   return tgtok::IntVal;
515 }
516 
517 /// LexBracket - We just read '['.  If this is a code block, return it,
518 /// otherwise return the bracket.  Match: '[' and '[{ ( [^}]+ | }[^]] )* }]'
LexBracket()519 tgtok::TokKind TGLexer::LexBracket() {
520   if (CurPtr[0] != '{')
521     return tgtok::l_square;
522   ++CurPtr;
523   const char *CodeStart = CurPtr;
524   while (true) {
525     int Char = getNextChar();
526     if (Char == EOF) break;
527 
528     if (Char != '}') continue;
529 
530     Char = getNextChar();
531     if (Char == EOF) break;
532     if (Char == ']') {
533       CurStrVal.assign(CodeStart, CurPtr-2);
534       return tgtok::CodeFragment;
535     }
536   }
537 
538   return ReturnError(CodeStart - 2, "Unterminated code block");
539 }
540 
541 /// LexExclaim - Lex '!' and '![a-zA-Z]+'.
LexExclaim()542 tgtok::TokKind TGLexer::LexExclaim() {
543   if (!isalpha(*CurPtr))
544     return ReturnError(CurPtr - 1, "Invalid \"!operator\"");
545 
546   const char *Start = CurPtr++;
547   while (isalpha(*CurPtr))
548     ++CurPtr;
549 
550   // Check to see which operator this is.
551   tgtok::TokKind Kind =
552     StringSwitch<tgtok::TokKind>(StringRef(Start, CurPtr - Start))
553     .Case("eq", tgtok::XEq)
554     .Case("ne", tgtok::XNe)
555     .Case("le", tgtok::XLe)
556     .Case("lt", tgtok::XLt)
557     .Case("ge", tgtok::XGe)
558     .Case("gt", tgtok::XGt)
559     .Case("if", tgtok::XIf)
560     .Case("cond", tgtok::XCond)
561     .Case("isa", tgtok::XIsA)
562     .Case("head", tgtok::XHead)
563     .Case("tail", tgtok::XTail)
564     .Case("size", tgtok::XSize)
565     .Case("con", tgtok::XConcat)
566     .Case("dag", tgtok::XDag)
567     .Case("add", tgtok::XADD)
568     .Case("sub", tgtok::XSUB)
569     .Case("mul", tgtok::XMUL)
570     .Case("div", tgtok::XDIV)
571     .Case("not", tgtok::XNOT)
572     .Case("logtwo", tgtok::XLOG2)
573     .Case("and", tgtok::XAND)
574     .Case("or", tgtok::XOR)
575     .Case("xor", tgtok::XXOR)
576     .Case("shl", tgtok::XSHL)
577     .Case("sra", tgtok::XSRA)
578     .Case("srl", tgtok::XSRL)
579     .Case("cast", tgtok::XCast)
580     .Case("empty", tgtok::XEmpty)
581     .Case("subst", tgtok::XSubst)
582     .Case("foldl", tgtok::XFoldl)
583     .Case("foreach", tgtok::XForEach)
584     .Case("filter", tgtok::XFilter)
585     .Case("listconcat", tgtok::XListConcat)
586     .Case("listsplat", tgtok::XListSplat)
587     .Case("listremove", tgtok::XListRemove)
588     .Case("strconcat", tgtok::XStrConcat)
589     .Case("interleave", tgtok::XInterleave)
590     .Case("substr", tgtok::XSubstr)
591     .Case("find", tgtok::XFind)
592     .Cases("setdagop", "setop", tgtok::XSetDagOp) // !setop is deprecated.
593     .Cases("getdagop", "getop", tgtok::XGetDagOp) // !getop is deprecated.
594     .Case("exists", tgtok::XExists)
595     .Default(tgtok::Error);
596 
597   return Kind != tgtok::Error ? Kind : ReturnError(Start-1, "Unknown operator");
598 }
599 
prepExitInclude(bool IncludeStackMustBeEmpty)600 bool TGLexer::prepExitInclude(bool IncludeStackMustBeEmpty) {
601   // Report an error, if preprocessor control stack for the current
602   // file is not empty.
603   if (!PrepIncludeStack.back()->empty()) {
604     prepReportPreprocessorStackError();
605 
606     return false;
607   }
608 
609   // Pop the preprocessing controls from the include stack.
610   if (PrepIncludeStack.empty()) {
611     PrintFatalError("Preprocessor include stack is empty");
612   }
613 
614   PrepIncludeStack.pop_back();
615 
616   if (IncludeStackMustBeEmpty) {
617     if (!PrepIncludeStack.empty())
618       PrintFatalError("Preprocessor include stack is not empty");
619   } else {
620     if (PrepIncludeStack.empty())
621       PrintFatalError("Preprocessor include stack is empty");
622   }
623 
624   return true;
625 }
626 
prepIsDirective() const627 tgtok::TokKind TGLexer::prepIsDirective() const {
628   for (const auto &PD : PreprocessorDirs) {
629     int NextChar = *CurPtr;
630     bool Match = true;
631     unsigned I = 0;
632     for (; I < strlen(PD.Word); ++I) {
633       if (NextChar != PD.Word[I]) {
634         Match = false;
635         break;
636       }
637 
638       NextChar = peekNextChar(I + 1);
639     }
640 
641     // Check for whitespace after the directive.  If there is no whitespace,
642     // then we do not recognize it as a preprocessing directive.
643     if (Match) {
644       tgtok::TokKind Kind = PD.Kind;
645 
646       // New line and EOF may follow only #else/#endif.  It will be reported
647       // as an error for #ifdef/#define after the call to prepLexMacroName().
648       if (NextChar == ' ' || NextChar == '\t' || NextChar == EOF ||
649           NextChar == '\n' ||
650           // It looks like TableGen does not support '\r' as the actual
651           // carriage return, e.g. getNextChar() treats a single '\r'
652           // as '\n'.  So we do the same here.
653           NextChar == '\r')
654         return Kind;
655 
656       // Allow comments after some directives, e.g.:
657       //     #else// OR #else/**/
658       //     #endif// OR #endif/**/
659       //
660       // Note that we do allow comments after #ifdef/#define here, e.g.
661       //     #ifdef/**/ AND #ifdef//
662       //     #define/**/ AND #define//
663       //
664       // These cases will be reported as incorrect after calling
665       // prepLexMacroName().  We could have supported C-style comments
666       // after #ifdef/#define, but this would complicate the code
667       // for little benefit.
668       if (NextChar == '/') {
669         NextChar = peekNextChar(I + 1);
670 
671         if (NextChar == '*' || NextChar == '/')
672           return Kind;
673 
674         // Pretend that we do not recognize the directive.
675       }
676     }
677   }
678 
679   return tgtok::Error;
680 }
681 
prepEatPreprocessorDirective(tgtok::TokKind Kind)682 bool TGLexer::prepEatPreprocessorDirective(tgtok::TokKind Kind) {
683   TokStart = CurPtr;
684 
685   for (const auto &PD : PreprocessorDirs)
686     if (PD.Kind == Kind) {
687       // Advance CurPtr to the end of the preprocessing word.
688       CurPtr += strlen(PD.Word);
689       return true;
690     }
691 
692   PrintFatalError("Unsupported preprocessing token in "
693                   "prepEatPreprocessorDirective()");
694   return false;
695 }
696 
lexPreprocessor(tgtok::TokKind Kind,bool ReturnNextLiveToken)697 tgtok::TokKind TGLexer::lexPreprocessor(
698     tgtok::TokKind Kind, bool ReturnNextLiveToken) {
699 
700   // We must be looking at a preprocessing directive.  Eat it!
701   if (!prepEatPreprocessorDirective(Kind))
702     PrintFatalError("lexPreprocessor() called for unknown "
703                     "preprocessor directive");
704 
705   if (Kind == tgtok::Ifdef || Kind == tgtok::Ifndef) {
706     StringRef MacroName = prepLexMacroName();
707     StringRef IfTokName = Kind == tgtok::Ifdef ? "#ifdef" : "#ifndef";
708     if (MacroName.empty())
709       return ReturnError(TokStart, "Expected macro name after " + IfTokName);
710 
711     bool MacroIsDefined = DefinedMacros.count(MacroName) != 0;
712 
713     // Canonicalize ifndef to ifdef equivalent
714     if (Kind == tgtok::Ifndef) {
715       MacroIsDefined = !MacroIsDefined;
716       Kind = tgtok::Ifdef;
717     }
718 
719     // Regardless of whether we are processing tokens or not,
720     // we put the #ifdef control on stack.
721     PrepIncludeStack.back()->push_back(
722         {Kind, MacroIsDefined, SMLoc::getFromPointer(TokStart)});
723 
724     if (!prepSkipDirectiveEnd())
725       return ReturnError(CurPtr, "Only comments are supported after " +
726                                      IfTokName + " NAME");
727 
728     // If we were not processing tokens before this #ifdef,
729     // then just return back to the lines skipping code.
730     if (!ReturnNextLiveToken)
731       return Kind;
732 
733     // If we were processing tokens before this #ifdef,
734     // and the macro is defined, then just return the next token.
735     if (MacroIsDefined)
736       return LexToken();
737 
738     // We were processing tokens before this #ifdef, and the macro
739     // is not defined, so we have to start skipping the lines.
740     // If the skipping is successful, it will return the token following
741     // either #else or #endif corresponding to this #ifdef.
742     if (prepSkipRegion(ReturnNextLiveToken))
743       return LexToken();
744 
745     return tgtok::Error;
746   } else if (Kind == tgtok::Else) {
747     // Check if this #else is correct before calling prepSkipDirectiveEnd(),
748     // which will move CurPtr away from the beginning of #else.
749     if (PrepIncludeStack.back()->empty())
750       return ReturnError(TokStart, "#else without #ifdef or #ifndef");
751 
752     PreprocessorControlDesc IfdefEntry = PrepIncludeStack.back()->back();
753 
754     if (IfdefEntry.Kind != tgtok::Ifdef) {
755       PrintError(TokStart, "double #else");
756       return ReturnError(IfdefEntry.SrcPos, "Previous #else is here");
757     }
758 
759     // Replace the corresponding #ifdef's control with its negation
760     // on the control stack.
761     PrepIncludeStack.back()->pop_back();
762     PrepIncludeStack.back()->push_back(
763         {Kind, !IfdefEntry.IsDefined, SMLoc::getFromPointer(TokStart)});
764 
765     if (!prepSkipDirectiveEnd())
766       return ReturnError(CurPtr, "Only comments are supported after #else");
767 
768     // If we were processing tokens before this #else,
769     // we have to start skipping lines until the matching #endif.
770     if (ReturnNextLiveToken) {
771       if (prepSkipRegion(ReturnNextLiveToken))
772         return LexToken();
773 
774       return tgtok::Error;
775     }
776 
777     // Return to the lines skipping code.
778     return Kind;
779   } else if (Kind == tgtok::Endif) {
780     // Check if this #endif is correct before calling prepSkipDirectiveEnd(),
781     // which will move CurPtr away from the beginning of #endif.
782     if (PrepIncludeStack.back()->empty())
783       return ReturnError(TokStart, "#endif without #ifdef");
784 
785     auto &IfdefOrElseEntry = PrepIncludeStack.back()->back();
786 
787     if (IfdefOrElseEntry.Kind != tgtok::Ifdef &&
788         IfdefOrElseEntry.Kind != tgtok::Else) {
789       PrintFatalError("Invalid preprocessor control on the stack");
790       return tgtok::Error;
791     }
792 
793     if (!prepSkipDirectiveEnd())
794       return ReturnError(CurPtr, "Only comments are supported after #endif");
795 
796     PrepIncludeStack.back()->pop_back();
797 
798     // If we were processing tokens before this #endif, then
799     // we should continue it.
800     if (ReturnNextLiveToken) {
801       return LexToken();
802     }
803 
804     // Return to the lines skipping code.
805     return Kind;
806   } else if (Kind == tgtok::Define) {
807     StringRef MacroName = prepLexMacroName();
808     if (MacroName.empty())
809       return ReturnError(TokStart, "Expected macro name after #define");
810 
811     if (!DefinedMacros.insert(MacroName).second)
812       PrintWarning(getLoc(),
813                    "Duplicate definition of macro: " + Twine(MacroName));
814 
815     if (!prepSkipDirectiveEnd())
816       return ReturnError(CurPtr,
817                          "Only comments are supported after #define NAME");
818 
819     if (!ReturnNextLiveToken) {
820       PrintFatalError("#define must be ignored during the lines skipping");
821       return tgtok::Error;
822     }
823 
824     return LexToken();
825   }
826 
827   PrintFatalError("Preprocessing directive is not supported");
828   return tgtok::Error;
829 }
830 
prepSkipRegion(bool MustNeverBeFalse)831 bool TGLexer::prepSkipRegion(bool MustNeverBeFalse) {
832   if (!MustNeverBeFalse)
833     PrintFatalError("Invalid recursion.");
834 
835   do {
836     // Skip all symbols to the line end.
837     prepSkipToLineEnd();
838 
839     // Find the first non-whitespace symbol in the next line(s).
840     if (!prepSkipLineBegin())
841       return false;
842 
843     // If the first non-blank/comment symbol on the line is '#',
844     // it may be a start of preprocessing directive.
845     //
846     // If it is not '#' just go to the next line.
847     if (*CurPtr == '#')
848       ++CurPtr;
849     else
850       continue;
851 
852     tgtok::TokKind Kind = prepIsDirective();
853 
854     // If we did not find a preprocessing directive or it is #define,
855     // then just skip to the next line.  We do not have to do anything
856     // for #define in the line-skipping mode.
857     if (Kind == tgtok::Error || Kind == tgtok::Define)
858       continue;
859 
860     tgtok::TokKind ProcessedKind = lexPreprocessor(Kind, false);
861 
862     // If lexPreprocessor() encountered an error during lexing this
863     // preprocessor idiom, then return false to the calling lexPreprocessor().
864     // This will force tgtok::Error to be returned to the tokens processing.
865     if (ProcessedKind == tgtok::Error)
866       return false;
867 
868     if (Kind != ProcessedKind)
869       PrintFatalError("prepIsDirective() and lexPreprocessor() "
870                       "returned different token kinds");
871 
872     // If this preprocessing directive enables tokens processing,
873     // then return to the lexPreprocessor() and get to the next token.
874     // We can move from line-skipping mode to processing tokens only
875     // due to #else or #endif.
876     if (prepIsProcessingEnabled()) {
877       if (Kind != tgtok::Else && Kind != tgtok::Endif) {
878         PrintFatalError("Tokens processing was enabled by an unexpected "
879                         "preprocessing directive");
880         return false;
881       }
882 
883       return true;
884     }
885   } while (CurPtr != CurBuf.end());
886 
887   // We have reached the end of the file, but never left the lines-skipping
888   // mode.  This means there is no matching #endif.
889   prepReportPreprocessorStackError();
890   return false;
891 }
892 
prepLexMacroName()893 StringRef TGLexer::prepLexMacroName() {
894   // Skip whitespaces between the preprocessing directive and the macro name.
895   while (*CurPtr == ' ' || *CurPtr == '\t')
896     ++CurPtr;
897 
898   TokStart = CurPtr;
899   // Macro names start with [a-zA-Z_].
900   if (*CurPtr != '_' && !isalpha(*CurPtr))
901     return "";
902 
903   // Match the rest of the identifier regex: [0-9a-zA-Z_]*
904   while (isalpha(*CurPtr) || isdigit(*CurPtr) || *CurPtr == '_')
905     ++CurPtr;
906 
907   return StringRef(TokStart, CurPtr - TokStart);
908 }
909 
prepSkipLineBegin()910 bool TGLexer::prepSkipLineBegin() {
911   while (CurPtr != CurBuf.end()) {
912     switch (*CurPtr) {
913     case ' ':
914     case '\t':
915     case '\n':
916     case '\r':
917       break;
918 
919     case '/': {
920       int NextChar = peekNextChar(1);
921       if (NextChar == '*') {
922         // Skip C-style comment.
923         // Note that we do not care about skipping the C++-style comments.
924         // If the line contains "//", it may not contain any processable
925         // preprocessing directive.  Just return CurPtr pointing to
926         // the first '/' in this case.  We also do not care about
927         // incorrect symbols after the first '/' - we are in lines-skipping
928         // mode, so incorrect code is allowed to some extent.
929 
930         // Set TokStart to the beginning of the comment to enable proper
931         // diagnostic printing in case of error in SkipCComment().
932         TokStart = CurPtr;
933 
934         // CurPtr must point to '*' before call to SkipCComment().
935         ++CurPtr;
936         if (SkipCComment())
937           return false;
938       } else {
939         // CurPtr points to the non-whitespace '/'.
940         return true;
941       }
942 
943       // We must not increment CurPtr after the comment was lexed.
944       continue;
945     }
946 
947     default:
948       return true;
949     }
950 
951     ++CurPtr;
952   }
953 
954   // We have reached the end of the file.  Return to the lines skipping
955   // code, and allow it to handle the EOF as needed.
956   return true;
957 }
958 
prepSkipDirectiveEnd()959 bool TGLexer::prepSkipDirectiveEnd() {
960   while (CurPtr != CurBuf.end()) {
961     switch (*CurPtr) {
962     case ' ':
963     case '\t':
964       break;
965 
966     case '\n':
967     case '\r':
968       return true;
969 
970     case '/': {
971       int NextChar = peekNextChar(1);
972       if (NextChar == '/') {
973         // Skip C++-style comment.
974         // We may just return true now, but let's skip to the line/buffer end
975         // to simplify the method specification.
976         ++CurPtr;
977         SkipBCPLComment();
978       } else if (NextChar == '*') {
979         // When we are skipping C-style comment at the end of a preprocessing
980         // directive, we can skip several lines.  If any meaningful TD token
981         // follows the end of the C-style comment on the same line, it will
982         // be considered as an invalid usage of TD token.
983         // For example, we want to forbid usages like this one:
984         //     #define MACRO class Class {}
985         // But with C-style comments we also disallow the following:
986         //     #define MACRO /* This macro is used
987         //                      to ... */ class Class {}
988         // One can argue that this should be allowed, but it does not seem
989         // to be worth of the complication.  Moreover, this matches
990         // the C preprocessor behavior.
991 
992         // Set TokStart to the beginning of the comment to enable proper
993         // diagnostic printer in case of error in SkipCComment().
994         TokStart = CurPtr;
995         ++CurPtr;
996         if (SkipCComment())
997           return false;
998       } else {
999         TokStart = CurPtr;
1000         PrintError(CurPtr, "Unexpected character");
1001         return false;
1002       }
1003 
1004       // We must not increment CurPtr after the comment was lexed.
1005       continue;
1006     }
1007 
1008     default:
1009       // Do not allow any non-whitespaces after the directive.
1010       TokStart = CurPtr;
1011       return false;
1012     }
1013 
1014     ++CurPtr;
1015   }
1016 
1017   return true;
1018 }
1019 
prepSkipToLineEnd()1020 void TGLexer::prepSkipToLineEnd() {
1021   while (*CurPtr != '\n' && *CurPtr != '\r' && CurPtr != CurBuf.end())
1022     ++CurPtr;
1023 }
1024 
prepIsProcessingEnabled()1025 bool TGLexer::prepIsProcessingEnabled() {
1026   for (const PreprocessorControlDesc &I :
1027        llvm::reverse(*PrepIncludeStack.back()))
1028     if (!I.IsDefined)
1029       return false;
1030 
1031   return true;
1032 }
1033 
prepReportPreprocessorStackError()1034 void TGLexer::prepReportPreprocessorStackError() {
1035   if (PrepIncludeStack.back()->empty())
1036     PrintFatalError("prepReportPreprocessorStackError() called with "
1037                     "empty control stack");
1038 
1039   auto &PrepControl = PrepIncludeStack.back()->back();
1040   PrintError(CurBuf.end(), "Reached EOF without matching #endif");
1041   PrintError(PrepControl.SrcPos, "The latest preprocessor control is here");
1042 
1043   TokStart = CurPtr;
1044 }
1045