MacroArgs.cpp revision 200583 
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"
18using namespace clang;
19
20/// MacroArgs ctor function - This destroys the vector passed in.
21MacroArgs *MacroArgs::create(const MacroInfo *MI,
22                             const Token *UnexpArgTokens,
23                             unsigned NumToks, bool VarargsElided,
24                             Preprocessor &PP) {
25  assert(MI->isFunctionLike() &&
26         "Can't have args for an object-like macro!");
27
28  // Allocate memory for the MacroArgs object with the lexer tokens at the end.
29  MacroArgs *Result = (MacroArgs*)malloc(sizeof(MacroArgs) +
30                                         NumToks*sizeof(Token));
31  // Construct the macroargs object.
32  new (Result) MacroArgs(NumToks, VarargsElided);
33
34  // Copy the actual unexpanded tokens to immediately after the result ptr.
35  if (NumToks)
36    memcpy(const_cast<Token*>(Result->getUnexpArgument(0)),
37           UnexpArgTokens, NumToks*sizeof(Token));
38
39  return Result;
40}
41
42/// destroy - Destroy and deallocate the memory for this object.
43///
44void MacroArgs::destroy(Preprocessor &PP) {
45  // Run the dtor to deallocate the vectors.
46  this->~MacroArgs();
47  // Release the memory for the object.
48  free(this);
49}
50
51/// deallocate - This should only be called by the Preprocessor when managing
52/// its freelist.
53MacroArgs *MacroArgs::deallocate() {
54  MacroArgs *Next = ArgCache;
55
56  // Run the dtor to deallocate the vectors.
57  this->~MacroArgs();
58  // Release the memory for the object.
59  free(this);
60
61  return Next;
62}
63
64
65/// getArgLength - Given a pointer to an expanded or unexpanded argument,
66/// return the number of tokens, not counting the EOF, that make up the
67/// argument.
68unsigned MacroArgs::getArgLength(const Token *ArgPtr) {
69  unsigned NumArgTokens = 0;
70  for (; ArgPtr->isNot(tok::eof); ++ArgPtr)
71    ++NumArgTokens;
72  return NumArgTokens;
73}
74
75
76/// getUnexpArgument - Return the unexpanded tokens for the specified formal.
77///
78const Token *MacroArgs::getUnexpArgument(unsigned Arg) const {
79  // The unexpanded argument tokens start immediately after the MacroArgs object
80  // in memory.
81  const Token *Start = (const Token *)(this+1);
82  const Token *Result = Start;
83  // Scan to find Arg.
84  for (; Arg; ++Result) {
85    assert(Result < Start+NumUnexpArgTokens && "Invalid arg #");
86    if (Result->is(tok::eof))
87      --Arg;
88  }
89  assert(Result < Start+NumUnexpArgTokens && "Invalid arg #");
90  return Result;
91}
92
93
94/// ArgNeedsPreexpansion - If we can prove that the argument won't be affected
95/// by pre-expansion, return false.  Otherwise, conservatively return true.
96bool MacroArgs::ArgNeedsPreexpansion(const Token *ArgTok,
97                                     Preprocessor &PP) const {
98  // If there are no identifiers in the argument list, or if the identifiers are
99  // known to not be macros, pre-expansion won't modify it.
100  for (; ArgTok->isNot(tok::eof); ++ArgTok)
101    if (IdentifierInfo *II = ArgTok->getIdentifierInfo()) {
102      if (II->hasMacroDefinition() && PP.getMacroInfo(II)->isEnabled())
103        // Return true even though the macro could be a function-like macro
104        // without a following '(' token.
105        return true;
106    }
107  return false;
108}
109
110/// getPreExpArgument - Return the pre-expanded form of the specified
111/// argument.
112const std::vector<Token> &
113MacroArgs::getPreExpArgument(unsigned Arg, Preprocessor &PP) {
114  assert(Arg < NumUnexpArgTokens && "Invalid argument number!");
115
116  // If we have already computed this, return it.
117  if (PreExpArgTokens.empty())
118    PreExpArgTokens.resize(NumUnexpArgTokens);
119
120  std::vector<Token> &Result = PreExpArgTokens[Arg];
121  if (!Result.empty()) return Result;
122
123  const Token *AT = getUnexpArgument(Arg);
124  unsigned NumToks = getArgLength(AT)+1;  // Include the EOF.
125
126  // Otherwise, we have to pre-expand this argument, populating Result.  To do
127  // this, we set up a fake TokenLexer to lex from the unexpanded argument
128  // list.  With this installed, we lex expanded tokens until we hit the EOF
129  // token at the end of the unexp list.
130  PP.EnterTokenStream(AT, NumToks, false /*disable expand*/,
131                      false /*owns tokens*/);
132
133  // Lex all of the macro-expanded tokens into Result.
134  do {
135    Result.push_back(Token());
136    Token &Tok = Result.back();
137    PP.Lex(Tok);
138  } while (Result.back().isNot(tok::eof));
139
140  // Pop the token stream off the top of the stack.  We know that the internal
141  // pointer inside of it is to the "end" of the token stream, but the stack
142  // will not otherwise be popped until the next token is lexed.  The problem is
143  // that the token may be lexed sometime after the vector of tokens itself is
144  // destroyed, which would be badness.
145  PP.RemoveTopOfLexerStack();
146  return Result;
147}
148
149
150/// StringifyArgument - Implement C99 6.10.3.2p2, converting a sequence of
151/// tokens into the literal string token that should be produced by the C #
152/// preprocessor operator.  If Charify is true, then it should be turned into
153/// a character literal for the Microsoft charize (#@) extension.
154///
155Token MacroArgs::StringifyArgument(const Token *ArgToks,
156                                   Preprocessor &PP, bool Charify) {
157  Token Tok;
158  Tok.startToken();
159  Tok.setKind(tok::string_literal);
160
161  const Token *ArgTokStart = ArgToks;
162
163  // Stringify all the tokens.
164  llvm::SmallString<128> Result;
165  Result += "\"";
166
167  bool isFirst = true;
168  for (; ArgToks->isNot(tok::eof); ++ArgToks) {
169    const Token &Tok = *ArgToks;
170    if (!isFirst && (Tok.hasLeadingSpace() || Tok.isAtStartOfLine()))
171      Result += ' ';
172    isFirst = false;
173
174    // If this is a string or character constant, escape the token as specified
175    // by 6.10.3.2p2.
176    if (Tok.is(tok::string_literal) ||       // "foo"
177        Tok.is(tok::wide_string_literal) ||  // L"foo"
178        Tok.is(tok::char_constant)) {        // 'x' and L'x'.
179      std::string Str = Lexer::Stringify(PP.getSpelling(Tok));
180      Result.append(Str.begin(), Str.end());
181    } else {
182      // Otherwise, just append the token.  Do some gymnastics to get the token
183      // in place and avoid copies where possible.
184      unsigned CurStrLen = Result.size();
185      Result.resize(CurStrLen+Tok.getLength());
186      const char *BufPtr = &Result[CurStrLen];
187      unsigned ActualTokLen = PP.getSpelling(Tok, BufPtr);
188
189      // If getSpelling returned a pointer to an already uniqued version of the
190      // string instead of filling in BufPtr, memcpy it onto our string.
191      if (BufPtr != &Result[CurStrLen])
192        memcpy(&Result[CurStrLen], BufPtr, ActualTokLen);
193
194      // If the token was dirty, the spelling may be shorter than the token.
195      if (ActualTokLen != Tok.getLength())
196        Result.resize(CurStrLen+ActualTokLen);
197    }
198  }
199
200  // If the last character of the string is a \, and if it isn't escaped, this
201  // is an invalid string literal, diagnose it as specified in C99.
202  if (Result.back() == '\\') {
203    // Count the number of consequtive \ characters.  If even, then they are
204    // just escaped backslashes, otherwise it's an error.
205    unsigned FirstNonSlash = Result.size()-2;
206    // Guaranteed to find the starting " if nothing else.
207    while (Result[FirstNonSlash] == '\\')
208      --FirstNonSlash;
209    if ((Result.size()-1-FirstNonSlash) & 1) {
210      // Diagnose errors for things like: #define F(X) #X   /   F(\)
211      PP.Diag(ArgToks[-1], diag::pp_invalid_string_literal);
212      Result.pop_back();  // remove one of the \'s.
213    }
214  }
215  Result += '"';
216
217  // If this is the charify operation and the result is not a legal character
218  // constant, diagnose it.
219  if (Charify) {
220    // First step, turn double quotes into single quotes:
221    Result[0] = '\'';
222    Result[Result.size()-1] = '\'';
223
224    // Check for bogus character.
225    bool isBad = false;
226    if (Result.size() == 3)
227      isBad = Result[1] == '\'';   // ''' is not legal. '\' already fixed above.
228    else
229      isBad = (Result.size() != 4 || Result[1] != '\\');  // Not '\x'
230
231    if (isBad) {
232      PP.Diag(ArgTokStart[0], diag::err_invalid_character_to_charify);
233      Result = "' '";  // Use something arbitrary, but legal.
234    }
235  }
236
237  PP.CreateString(&Result[0], Result.size(), Tok);
238  return Tok;
239}
240
241/// getStringifiedArgument - Compute, cache, and return the specified argument
242/// that has been 'stringified' as required by the # operator.
243const Token &MacroArgs::getStringifiedArgument(unsigned ArgNo,
244                                               Preprocessor &PP) {
245  assert(ArgNo < NumUnexpArgTokens && "Invalid argument number!");
246  if (StringifiedArgs.empty()) {
247    StringifiedArgs.resize(getNumArguments());
248    memset(&StringifiedArgs[0], 0,
249           sizeof(StringifiedArgs[0])*getNumArguments());
250  }
251  if (StringifiedArgs[ArgNo].isNot(tok::string_literal))
252    StringifiedArgs[ArgNo] = StringifyArgument(getUnexpArgument(ArgNo), PP);
253  return StringifiedArgs[ArgNo];
254}
255