Home | History | Annotate | Download | only in Lex 
      1 //===--- TokenLexer.cpp - Lex from a token stream -------------------------===//
      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 // This file implements the TokenLexer interface.
     11 //
     12 //===----------------------------------------------------------------------===//
     13 
     14 #include "MacroArgs.h"
     15 #include "clang/Lex/MacroInfo.h"
     16 #include "clang/Lex/Preprocessor.h"
     17 #include "clang/Lex/LexDiagnostic.h"
     18 
     19 #include <algorithm>
     20 
     21 using namespace clang;
     22 
     23 /// MacroArgs ctor function - This destroys the vector passed in.
     24 MacroArgs *MacroArgs::create(const MacroInfo *MI,
     25                              llvm::ArrayRef<Token> UnexpArgTokens,
     26                              bool VarargsElided, Preprocessor &PP) {
     27   assert(MI->isFunctionLike() &&
     28          "Can't have args for an object-like macro!");
     29   MacroArgs **ResultEnt = 0;
     30   unsigned ClosestMatch = ~0U;
     31 
     32   // See if we have an entry with a big enough argument list to reuse on the
     33   // free list.  If so, reuse it.
     34   for (MacroArgs **Entry = &PP.MacroArgCache; *Entry;
     35        Entry = &(*Entry)->ArgCache)
     36     if ((*Entry)->NumUnexpArgTokens >= UnexpArgTokens.size() &&
     37         (*Entry)->NumUnexpArgTokens < ClosestMatch) {
     38       ResultEnt = Entry;
     39 
     40       // If we have an exact match, use it.
     41       if ((*Entry)->NumUnexpArgTokens == UnexpArgTokens.size())
     42         break;
     43       // Otherwise, use the best fit.
     44       ClosestMatch = (*Entry)->NumUnexpArgTokens;
     45     }
     46 
     47   MacroArgs *Result;
     48   if (ResultEnt == 0) {
     49     // Allocate memory for a MacroArgs object with the lexer tokens at the end.
     50     Result = (MacroArgs*)malloc(sizeof(MacroArgs) +
     51                                 UnexpArgTokens.size() * sizeof(Token));
     52     // Construct the MacroArgs object.
     53     new (Result) MacroArgs(UnexpArgTokens.size(), VarargsElided);
     54   } else {
     55     Result = *ResultEnt;
     56     // Unlink this node from the preprocessors singly linked list.
     57     *ResultEnt = Result->ArgCache;
     58     Result->NumUnexpArgTokens = UnexpArgTokens.size();
     59     Result->VarargsElided = VarargsElided;
     60   }
     61 
     62   // Copy the actual unexpanded tokens to immediately after the result ptr.
     63   if (!UnexpArgTokens.empty())
     64     std::copy(UnexpArgTokens.begin(), UnexpArgTokens.end(),
     65               const_cast<Token*>(Result->getUnexpArgument(0)));
     66 
     67   return Result;
     68 }
     69 
     70 /// destroy - Destroy and deallocate the memory for this object.
     71 ///
     72 void MacroArgs::destroy(Preprocessor &PP) {
     73   StringifiedArgs.clear();
     74 
     75   // Don't clear PreExpArgTokens, just clear the entries.  Clearing the entries
     76   // would deallocate the element vectors.
     77   for (unsigned i = 0, e = PreExpArgTokens.size(); i != e; ++i)
     78     PreExpArgTokens[i].clear();
     79 
     80   // Add this to the preprocessor's free list.
     81   ArgCache = PP.MacroArgCache;
     82   PP.MacroArgCache = this;
     83 }
     84 
     85 /// deallocate - This should only be called by the Preprocessor when managing
     86 /// its freelist.
     87 MacroArgs *MacroArgs::deallocate() {
     88   MacroArgs *Next = ArgCache;
     89 
     90   // Run the dtor to deallocate the vectors.
     91   this->~MacroArgs();
     92   // Release the memory for the object.
     93   free(this);
     94 
     95   return Next;
     96 }
     97 
     98 
     99 /// getArgLength - Given a pointer to an expanded or unexpanded argument,
    100 /// return the number of tokens, not counting the EOF, that make up the
    101 /// argument.
    102 unsigned MacroArgs::getArgLength(const Token *ArgPtr) {
    103   unsigned NumArgTokens = 0;
    104   for (; ArgPtr->isNot(tok::eof); ++ArgPtr)
    105     ++NumArgTokens;
    106   return NumArgTokens;
    107 }
    108 
    109 
    110 /// getUnexpArgument - Return the unexpanded tokens for the specified formal.
    111 ///
    112 const Token *MacroArgs::getUnexpArgument(unsigned Arg) const {
    113   // The unexpanded argument tokens start immediately after the MacroArgs object
    114   // in memory.
    115   const Token *Start = (const Token *)(this+1);
    116   const Token *Result = Start;
    117   // Scan to find Arg.
    118   for (; Arg; ++Result) {
    119     assert(Result < Start+NumUnexpArgTokens && "Invalid arg #");
    120     if (Result->is(tok::eof))
    121       --Arg;
    122   }
    123   assert(Result < Start+NumUnexpArgTokens && "Invalid arg #");
    124   return Result;
    125 }
    126 
    127 
    128 /// ArgNeedsPreexpansion - If we can prove that the argument won't be affected
    129 /// by pre-expansion, return false.  Otherwise, conservatively return true.
    130 bool MacroArgs::ArgNeedsPreexpansion(const Token *ArgTok,
    131                                      Preprocessor &PP) const {
    132   // If there are no identifiers in the argument list, or if the identifiers are
    133   // known to not be macros, pre-expansion won't modify it.
    134   for (; ArgTok->isNot(tok::eof); ++ArgTok)
    135     if (IdentifierInfo *II = ArgTok->getIdentifierInfo()) {
    136       if (II->hasMacroDefinition() && PP.getMacroInfo(II)->isEnabled())
    137         // Return true even though the macro could be a function-like macro
    138         // without a following '(' token.
    139         return true;
    140     }
    141   return false;
    142 }
    143 
    144 /// getPreExpArgument - Return the pre-expanded form of the specified
    145 /// argument.
    146 const std::vector<Token> &
    147 MacroArgs::getPreExpArgument(unsigned Arg, const MacroInfo *MI,
    148                              Preprocessor &PP) {
    149   assert(Arg < MI->getNumArgs() && "Invalid argument number!");
    150 
    151   // If we have already computed this, return it.
    152   if (PreExpArgTokens.size() < MI->getNumArgs())
    153     PreExpArgTokens.resize(MI->getNumArgs());
    154 
    155   std::vector<Token> &Result = PreExpArgTokens[Arg];
    156   if (!Result.empty()) return Result;
    157 
    158   const Token *AT = getUnexpArgument(Arg);
    159   unsigned NumToks = getArgLength(AT)+1;  // Include the EOF.
    160 
    161   // Otherwise, we have to pre-expand this argument, populating Result.  To do
    162   // this, we set up a fake TokenLexer to lex from the unexpanded argument
    163   // list.  With this installed, we lex expanded tokens until we hit the EOF
    164   // token at the end of the unexp list.
    165   PP.EnterTokenStream(AT, NumToks, false /*disable expand*/,
    166                       false /*owns tokens*/);
    167 
    168   // Lex all of the macro-expanded tokens into Result.
    169   do {
    170     Result.push_back(Token());
    171     Token &Tok = Result.back();
    172     PP.Lex(Tok);
    173   } while (Result.back().isNot(tok::eof));
    174 
    175   // Pop the token stream off the top of the stack.  We know that the internal
    176   // pointer inside of it is to the "end" of the token stream, but the stack
    177   // will not otherwise be popped until the next token is lexed.  The problem is
    178   // that the token may be lexed sometime after the vector of tokens itself is
    179   // destroyed, which would be badness.
    180   PP.RemoveTopOfLexerStack();
    181   return Result;
    182 }
    183 
    184 
    185 /// StringifyArgument - Implement C99 6.10.3.2p2, converting a sequence of
    186 /// tokens into the literal string token that should be produced by the C #
    187 /// preprocessor operator.  If Charify is true, then it should be turned into
    188 /// a character literal for the Microsoft charize (#@) extension.
    189 ///
    190 Token MacroArgs::StringifyArgument(const Token *ArgToks,
    191                                    Preprocessor &PP, bool Charify,
    192                                    SourceLocation ExpansionLocStart,
    193                                    SourceLocation ExpansionLocEnd) {
    194   Token Tok;
    195   Tok.startToken();
    196   Tok.setKind(Charify ? tok::char_constant : tok::string_literal);
    197 
    198   const Token *ArgTokStart = ArgToks;
    199 
    200   // Stringify all the tokens.
    201   llvm::SmallString<128> Result;
    202   Result += "\"";
    203 
    204   bool isFirst = true;
    205   for (; ArgToks->isNot(tok::eof); ++ArgToks) {
    206     const Token &Tok = *ArgToks;
    207     if (!isFirst && (Tok.hasLeadingSpace() || Tok.isAtStartOfLine()))
    208       Result += ' ';
    209     isFirst = false;
    210 
    211     // If this is a string or character constant, escape the token as specified
    212     // by 6.10.3.2p2.
    213     if (Tok.is(tok::string_literal) ||       // "foo"
    214         Tok.is(tok::wide_string_literal) ||  // L"foo"
    215         Tok.is(tok::utf8_string_literal) ||  // u8"foo"
    216         Tok.is(tok::utf16_string_literal) || // u"foo"
    217         Tok.is(tok::utf32_string_literal) || // U"foo"
    218         Tok.is(tok::char_constant) ||        // 'x'
    219         Tok.is(tok::wide_char_constant) ||   // L'x'.
    220         Tok.is(tok::utf16_char_constant) ||  // u'x'.
    221         Tok.is(tok::utf32_char_constant)) {  // U'x'.
    222       bool Invalid = false;
    223       std::string TokStr = PP.getSpelling(Tok, &Invalid);
    224       if (!Invalid) {
    225         std::string Str = Lexer::Stringify(TokStr);
    226         Result.append(Str.begin(), Str.end());
    227       }
    228     } else if (Tok.is(tok::code_completion)) {
    229       PP.CodeCompleteNaturalLanguage();
    230     } else {
    231       // Otherwise, just append the token.  Do some gymnastics to get the token
    232       // in place and avoid copies where possible.
    233       unsigned CurStrLen = Result.size();
    234       Result.resize(CurStrLen+Tok.getLength());
    235       const char *BufPtr = &Result[CurStrLen];
    236       bool Invalid = false;
    237       unsigned ActualTokLen = PP.getSpelling(Tok, BufPtr, &Invalid);
    238 
    239       if (!Invalid) {
    240         // If getSpelling returned a pointer to an already uniqued version of
    241         // the string instead of filling in BufPtr, memcpy it onto our string.
    242         if (BufPtr != &Result[CurStrLen])
    243           memcpy(&Result[CurStrLen], BufPtr, ActualTokLen);
    244 
    245         // If the token was dirty, the spelling may be shorter than the token.
    246         if (ActualTokLen != Tok.getLength())
    247           Result.resize(CurStrLen+ActualTokLen);
    248       }
    249     }
    250   }
    251 
    252   // If the last character of the string is a \, and if it isn't escaped, this
    253   // is an invalid string literal, diagnose it as specified in C99.
    254   if (Result.back() == '\\') {
    255     // Count the number of consequtive \ characters.  If even, then they are
    256     // just escaped backslashes, otherwise it's an error.
    257     unsigned FirstNonSlash = Result.size()-2;
    258     // Guaranteed to find the starting " if nothing else.
    259     while (Result[FirstNonSlash] == '\\')
    260       --FirstNonSlash;
    261     if ((Result.size()-1-FirstNonSlash) & 1) {
    262       // Diagnose errors for things like: #define F(X) #X   /   F(\)
    263       PP.Diag(ArgToks[-1], diag::pp_invalid_string_literal);
    264       Result.pop_back();  // remove one of the \'s.
    265     }
    266   }
    267   Result += '"';
    268 
    269   // If this is the charify operation and the result is not a legal character
    270   // constant, diagnose it.
    271   if (Charify) {
    272     // First step, turn double quotes into single quotes:
    273     Result[0] = '\'';
    274     Result[Result.size()-1] = '\'';
    275 
    276     // Check for bogus character.
    277     bool isBad = false;
    278     if (Result.size() == 3)
    279       isBad = Result[1] == '\'';   // ''' is not legal. '\' already fixed above.
    280     else
    281       isBad = (Result.size() != 4 || Result[1] != '\\');  // Not '\x'
    282 
    283     if (isBad) {
    284       PP.Diag(ArgTokStart[0], diag::err_invalid_character_to_charify);
    285       Result = "' '";  // Use something arbitrary, but legal.
    286     }
    287   }
    288 
    289   PP.CreateString(&Result[0], Result.size(), Tok,
    290                   ExpansionLocStart, ExpansionLocEnd);
    291   return Tok;
    292 }
    293 
    294 /// getStringifiedArgument - Compute, cache, and return the specified argument
    295 /// that has been 'stringified' as required by the # operator.
    296 const Token &MacroArgs::getStringifiedArgument(unsigned ArgNo,
    297                                                Preprocessor &PP,
    298                                                SourceLocation ExpansionLocStart,
    299                                                SourceLocation ExpansionLocEnd) {
    300   assert(ArgNo < NumUnexpArgTokens && "Invalid argument number!");
    301   if (StringifiedArgs.empty()) {
    302     StringifiedArgs.resize(getNumArguments());
    303     memset((void*)&StringifiedArgs[0], 0,
    304            sizeof(StringifiedArgs[0])*getNumArguments());
    305   }
    306   if (StringifiedArgs[ArgNo].isNot(tok::string_literal))
    307     StringifiedArgs[ArgNo] = StringifyArgument(getUnexpArgument(ArgNo), PP,
    308                                                /*Charify=*/false,
    309                                                ExpansionLocStart,
    310                                                ExpansionLocEnd);
    311   return StringifiedArgs[ArgNo];
    312 }
    313