MacroArgs.cpp revision 341825 
1//===--- MacroArgs.cpp - Formal argument info for Macros ------------------===//
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 MacroArgs interface.
11//
12//===----------------------------------------------------------------------===//
13
14#include "clang/Lex/MacroArgs.h"
15#include "clang/Lex/LexDiagnostic.h"
16#include "clang/Lex/MacroInfo.h"
17#include "clang/Lex/Preprocessor.h"
18#include "llvm/ADT/SmallString.h"
19#include "llvm/Support/SaveAndRestore.h"
20#include <algorithm>
21
22using namespace clang;
23
24/// MacroArgs ctor function - This destroys the vector passed in.
25MacroArgs *MacroArgs::create(const MacroInfo *MI,
26                             ArrayRef<Token> UnexpArgTokens,
27                             bool VarargsElided, Preprocessor &PP) {
28  assert(MI->isFunctionLike() &&
29         "Can't have args for an object-like macro!");
30  MacroArgs **ResultEnt = nullptr;
31  unsigned ClosestMatch = ~0U;
32
33  // See if we have an entry with a big enough argument list to reuse on the
34  // free list.  If so, reuse it.
35  for (MacroArgs **Entry = &PP.MacroArgCache; *Entry;
36       Entry = &(*Entry)->ArgCache) {
37    if ((*Entry)->NumUnexpArgTokens >= UnexpArgTokens.size() &&
38        (*Entry)->NumUnexpArgTokens < ClosestMatch) {
39      ResultEnt = Entry;
40
41      // If we have an exact match, use it.
42      if ((*Entry)->NumUnexpArgTokens == UnexpArgTokens.size())
43        break;
44      // Otherwise, use the best fit.
45      ClosestMatch = (*Entry)->NumUnexpArgTokens;
46    }
47  }
48  MacroArgs *Result;
49  if (!ResultEnt) {
50    // Allocate memory for a MacroArgs object with the lexer tokens at the end,
51    // and construct the MacroArgs object.
52    Result = new (
53        llvm::safe_malloc(totalSizeToAlloc<Token>(UnexpArgTokens.size())))
54        MacroArgs(UnexpArgTokens.size(), VarargsElided, MI->getNumParams());
55  } else {
56    Result = *ResultEnt;
57    // Unlink this node from the preprocessors singly linked list.
58    *ResultEnt = Result->ArgCache;
59    Result->NumUnexpArgTokens = UnexpArgTokens.size();
60    Result->VarargsElided = VarargsElided;
61    Result->NumMacroArgs = MI->getNumParams();
62  }
63
64  // Copy the actual unexpanded tokens to immediately after the result ptr.
65  if (!UnexpArgTokens.empty()) {
66    static_assert(std::is_trivial<Token>::value,
67                  "assume trivial copyability if copying into the "
68                  "uninitialized array (as opposed to reusing a cached "
69                  "MacroArgs)");
70    std::copy(UnexpArgTokens.begin(), UnexpArgTokens.end(),
71              Result->getTrailingObjects<Token>());
72  }
73
74  return Result;
75}
76
77/// destroy - Destroy and deallocate the memory for this object.
78///
79void MacroArgs::destroy(Preprocessor &PP) {
80  StringifiedArgs.clear();
81
82  // Don't clear PreExpArgTokens, just clear the entries.  Clearing the entries
83  // would deallocate the element vectors.
84  for (unsigned i = 0, e = PreExpArgTokens.size(); i != e; ++i)
85    PreExpArgTokens[i].clear();
86
87  // Add this to the preprocessor's free list.
88  ArgCache = PP.MacroArgCache;
89  PP.MacroArgCache = this;
90}
91
92/// deallocate - This should only be called by the Preprocessor when managing
93/// its freelist.
94MacroArgs *MacroArgs::deallocate() {
95  MacroArgs *Next = ArgCache;
96
97  // Run the dtor to deallocate the vectors.
98  this->~MacroArgs();
99  // Release the memory for the object.
100  static_assert(std::is_trivially_destructible<Token>::value,
101                "assume trivially destructible and forego destructors");
102  free(this);
103
104  return Next;
105}
106
107
108/// getArgLength - Given a pointer to an expanded or unexpanded argument,
109/// return the number of tokens, not counting the EOF, that make up the
110/// argument.
111unsigned MacroArgs::getArgLength(const Token *ArgPtr) {
112  unsigned NumArgTokens = 0;
113  for (; ArgPtr->isNot(tok::eof); ++ArgPtr)
114    ++NumArgTokens;
115  return NumArgTokens;
116}
117
118
119/// getUnexpArgument - Return the unexpanded tokens for the specified formal.
120///
121const Token *MacroArgs::getUnexpArgument(unsigned Arg) const {
122
123  assert(Arg < getNumMacroArguments() && "Invalid arg #");
124  // The unexpanded argument tokens start immediately after the MacroArgs object
125  // in memory.
126  const Token *Start = getTrailingObjects<Token>();
127  const Token *Result = Start;
128
129  // Scan to find Arg.
130  for (; Arg; ++Result) {
131    assert(Result < Start+NumUnexpArgTokens && "Invalid arg #");
132    if (Result->is(tok::eof))
133      --Arg;
134  }
135  assert(Result < Start+NumUnexpArgTokens && "Invalid arg #");
136  return Result;
137}
138
139// This function assumes that the variadic arguments are the tokens
140// corresponding to the last parameter (ellipsis) - and since tokens are
141// separated by the 'eof' token, if that is the only token corresponding to that
142// last parameter, we know no variadic arguments were supplied.
143bool MacroArgs::invokedWithVariadicArgument(const MacroInfo *const MI) const {
144  if (!MI->isVariadic())
145    return false;
146  const int VariadicArgIndex = getNumMacroArguments() - 1;
147  return getUnexpArgument(VariadicArgIndex)->isNot(tok::eof);
148}
149
150/// ArgNeedsPreexpansion - If we can prove that the argument won't be affected
151/// by pre-expansion, return false.  Otherwise, conservatively return true.
152bool MacroArgs::ArgNeedsPreexpansion(const Token *ArgTok,
153                                     Preprocessor &PP) const {
154  // If there are no identifiers in the argument list, or if the identifiers are
155  // known to not be macros, pre-expansion won't modify it.
156  for (; ArgTok->isNot(tok::eof); ++ArgTok)
157    if (IdentifierInfo *II = ArgTok->getIdentifierInfo())
158      if (II->hasMacroDefinition())
159        // Return true even though the macro could be a function-like macro
160        // without a following '(' token, or could be disabled, or not visible.
161        return true;
162  return false;
163}
164
165/// getPreExpArgument - Return the pre-expanded form of the specified
166/// argument.
167const std::vector<Token> &MacroArgs::getPreExpArgument(unsigned Arg,
168                                                       Preprocessor &PP) {
169  assert(Arg < getNumMacroArguments() && "Invalid argument number!");
170
171  // If we have already computed this, return it.
172  if (PreExpArgTokens.size() < getNumMacroArguments())
173    PreExpArgTokens.resize(getNumMacroArguments());
174
175  std::vector<Token> &Result = PreExpArgTokens[Arg];
176  if (!Result.empty()) return Result;
177
178  SaveAndRestore<bool> PreExpandingMacroArgs(PP.InMacroArgPreExpansion, true);
179
180  const Token *AT = getUnexpArgument(Arg);
181  unsigned NumToks = getArgLength(AT)+1;  // Include the EOF.
182
183  // Otherwise, we have to pre-expand this argument, populating Result.  To do
184  // this, we set up a fake TokenLexer to lex from the unexpanded argument
185  // list.  With this installed, we lex expanded tokens until we hit the EOF
186  // token at the end of the unexp list.
187  PP.EnterTokenStream(AT, NumToks, false /*disable expand*/,
188                      false /*owns tokens*/);
189
190  // Lex all of the macro-expanded tokens into Result.
191  do {
192    Result.push_back(Token());
193    Token &Tok = Result.back();
194    PP.Lex(Tok);
195  } while (Result.back().isNot(tok::eof));
196
197  // Pop the token stream off the top of the stack.  We know that the internal
198  // pointer inside of it is to the "end" of the token stream, but the stack
199  // will not otherwise be popped until the next token is lexed.  The problem is
200  // that the token may be lexed sometime after the vector of tokens itself is
201  // destroyed, which would be badness.
202  if (PP.InCachingLexMode())
203    PP.ExitCachingLexMode();
204  PP.RemoveTopOfLexerStack();
205  return Result;
206}
207
208
209/// StringifyArgument - Implement C99 6.10.3.2p2, converting a sequence of
210/// tokens into the literal string token that should be produced by the C #
211/// preprocessor operator.  If Charify is true, then it should be turned into
212/// a character literal for the Microsoft charize (#@) extension.
213///
214Token MacroArgs::StringifyArgument(const Token *ArgToks,
215                                   Preprocessor &PP, bool Charify,
216                                   SourceLocation ExpansionLocStart,
217                                   SourceLocation ExpansionLocEnd) {
218  Token Tok;
219  Tok.startToken();
220  Tok.setKind(Charify ? tok::char_constant : tok::string_literal);
221
222  const Token *ArgTokStart = ArgToks;
223
224  // Stringify all the tokens.
225  SmallString<128> Result;
226  Result += "\"";
227
228  bool isFirst = true;
229  for (; ArgToks->isNot(tok::eof); ++ArgToks) {
230    const Token &Tok = *ArgToks;
231    if (!isFirst && (Tok.hasLeadingSpace() || Tok.isAtStartOfLine()))
232      Result += ' ';
233    isFirst = false;
234
235    // If this is a string or character constant, escape the token as specified
236    // by 6.10.3.2p2.
237    if (tok::isStringLiteral(Tok.getKind()) || // "foo", u8R"x(foo)x"_bar, etc.
238        Tok.is(tok::char_constant) ||          // 'x'
239        Tok.is(tok::wide_char_constant) ||     // L'x'.
240        Tok.is(tok::utf8_char_constant) ||     // u8'x'.
241        Tok.is(tok::utf16_char_constant) ||    // u'x'.
242        Tok.is(tok::utf32_char_constant)) {    // U'x'.
243      bool Invalid = false;
244      std::string TokStr = PP.getSpelling(Tok, &Invalid);
245      if (!Invalid) {
246        std::string Str = Lexer::Stringify(TokStr);
247        Result.append(Str.begin(), Str.end());
248      }
249    } else if (Tok.is(tok::code_completion)) {
250      PP.CodeCompleteNaturalLanguage();
251    } else {
252      // Otherwise, just append the token.  Do some gymnastics to get the token
253      // in place and avoid copies where possible.
254      unsigned CurStrLen = Result.size();
255      Result.resize(CurStrLen+Tok.getLength());
256      const char *BufPtr = Result.data() + CurStrLen;
257      bool Invalid = false;
258      unsigned ActualTokLen = PP.getSpelling(Tok, BufPtr, &Invalid);
259
260      if (!Invalid) {
261        // If getSpelling returned a pointer to an already uniqued version of
262        // the string instead of filling in BufPtr, memcpy it onto our string.
263        if (ActualTokLen && BufPtr != &Result[CurStrLen])
264          memcpy(&Result[CurStrLen], BufPtr, ActualTokLen);
265
266        // If the token was dirty, the spelling may be shorter than the token.
267        if (ActualTokLen != Tok.getLength())
268          Result.resize(CurStrLen+ActualTokLen);
269      }
270    }
271  }
272
273  // If the last character of the string is a \, and if it isn't escaped, this
274  // is an invalid string literal, diagnose it as specified in C99.
275  if (Result.back() == '\\') {
276    // Count the number of consecutive \ characters.  If even, then they are
277    // just escaped backslashes, otherwise it's an error.
278    unsigned FirstNonSlash = Result.size()-2;
279    // Guaranteed to find the starting " if nothing else.
280    while (Result[FirstNonSlash] == '\\')
281      --FirstNonSlash;
282    if ((Result.size()-1-FirstNonSlash) & 1) {
283      // Diagnose errors for things like: #define F(X) #X   /   F(\)
284      PP.Diag(ArgToks[-1], diag::pp_invalid_string_literal);
285      Result.pop_back();  // remove one of the \'s.
286    }
287  }
288  Result += '"';
289
290  // If this is the charify operation and the result is not a legal character
291  // constant, diagnose it.
292  if (Charify) {
293    // First step, turn double quotes into single quotes:
294    Result[0] = '\'';
295    Result[Result.size()-1] = '\'';
296
297    // Check for bogus character.
298    bool isBad = false;
299    if (Result.size() == 3)
300      isBad = Result[1] == '\'';   // ''' is not legal. '\' already fixed above.
301    else
302      isBad = (Result.size() != 4 || Result[1] != '\\');  // Not '\x'
303
304    if (isBad) {
305      PP.Diag(ArgTokStart[0], diag::err_invalid_character_to_charify);
306      Result = "' '";  // Use something arbitrary, but legal.
307    }
308  }
309
310  PP.CreateString(Result, Tok,
311                  ExpansionLocStart, ExpansionLocEnd);
312  return Tok;
313}
314
315/// getStringifiedArgument - Compute, cache, and return the specified argument
316/// that has been 'stringified' as required by the # operator.
317const Token &MacroArgs::getStringifiedArgument(unsigned ArgNo,
318                                               Preprocessor &PP,
319                                               SourceLocation ExpansionLocStart,
320                                               SourceLocation ExpansionLocEnd) {
321  assert(ArgNo < getNumMacroArguments() && "Invalid argument number!");
322  if (StringifiedArgs.empty())
323    StringifiedArgs.resize(getNumMacroArguments(), {});
324
325  if (StringifiedArgs[ArgNo].isNot(tok::string_literal))
326    StringifiedArgs[ArgNo] = StringifyArgument(getUnexpArgument(ArgNo), PP,
327                                               /*Charify=*/false,
328                                               ExpansionLocStart,
329                                               ExpansionLocEnd);
330  return StringifiedArgs[ArgNo];
331}
332