1//===--- TokenConcatenation.cpp - Token Concatenation Avoidance -----------===//
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 TokenConcatenation class.
11//
12//===----------------------------------------------------------------------===//
13
14#include "clang/Lex/TokenConcatenation.h"
15#include "clang/Basic/CharInfo.h"
16#include "clang/Lex/Preprocessor.h"
17#include "llvm/Support/ErrorHandling.h"
18using namespace clang;
19
20
21/// IsStringPrefix - Return true if Str is a string prefix.
22/// 'L', 'u', 'U', or 'u8'. Including raw versions.
23static bool IsStringPrefix(StringRef Str, bool CPlusPlus11) {
24
25  if (Str[0] == 'L' ||
26      (CPlusPlus11 && (Str[0] == 'u' || Str[0] == 'U' || Str[0] == 'R'))) {
27
28    if (Str.size() == 1)
29      return true; // "L", "u", "U", and "R"
30
31    // Check for raw flavors. Need to make sure the first character wasn't
32    // already R. Need CPlusPlus11 check for "LR".
33    if (Str[1] == 'R' && Str[0] != 'R' && Str.size() == 2 && CPlusPlus11)
34      return true; // "LR", "uR", "UR"
35
36    // Check for "u8" and "u8R"
37    if (Str[0] == 'u' && Str[1] == '8') {
38      if (Str.size() == 2) return true; // "u8"
39      if (Str.size() == 3 && Str[2] == 'R') return true; // "u8R"
40    }
41  }
42
43  return false;
44}
45
46/// IsIdentifierStringPrefix - Return true if the spelling of the token
47/// is literally 'L', 'u', 'U', or 'u8'. Including raw versions.
48bool TokenConcatenation::IsIdentifierStringPrefix(const Token &Tok) const {
49  const LangOptions &LangOpts = PP.getLangOpts();
50
51  if (!Tok.needsCleaning()) {
52    if (Tok.getLength() < 1 || Tok.getLength() > 3)
53      return false;
54    SourceManager &SM = PP.getSourceManager();
55    const char *Ptr = SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation()));
56    return IsStringPrefix(StringRef(Ptr, Tok.getLength()),
57                          LangOpts.CPlusPlus11);
58  }
59
60  if (Tok.getLength() < 256) {
61    char Buffer[256];
62    const char *TokPtr = Buffer;
63    unsigned length = PP.getSpelling(Tok, TokPtr);
64    return IsStringPrefix(StringRef(TokPtr, length), LangOpts.CPlusPlus11);
65  }
66
67  return IsStringPrefix(StringRef(PP.getSpelling(Tok)), LangOpts.CPlusPlus11);
68}
69
70TokenConcatenation::TokenConcatenation(Preprocessor &pp) : PP(pp) {
71  memset(TokenInfo, 0, sizeof(TokenInfo));
72
73  // These tokens have custom code in AvoidConcat.
74  TokenInfo[tok::identifier      ] |= aci_custom;
75  TokenInfo[tok::numeric_constant] |= aci_custom_firstchar;
76  TokenInfo[tok::period          ] |= aci_custom_firstchar;
77  TokenInfo[tok::amp             ] |= aci_custom_firstchar;
78  TokenInfo[tok::plus            ] |= aci_custom_firstchar;
79  TokenInfo[tok::minus           ] |= aci_custom_firstchar;
80  TokenInfo[tok::slash           ] |= aci_custom_firstchar;
81  TokenInfo[tok::less            ] |= aci_custom_firstchar;
82  TokenInfo[tok::greater         ] |= aci_custom_firstchar;
83  TokenInfo[tok::pipe            ] |= aci_custom_firstchar;
84  TokenInfo[tok::percent         ] |= aci_custom_firstchar;
85  TokenInfo[tok::colon           ] |= aci_custom_firstchar;
86  TokenInfo[tok::hash            ] |= aci_custom_firstchar;
87  TokenInfo[tok::arrow           ] |= aci_custom_firstchar;
88
89  // These tokens have custom code in C++11 mode.
90  if (PP.getLangOpts().CPlusPlus11) {
91    TokenInfo[tok::string_literal      ] |= aci_custom;
92    TokenInfo[tok::wide_string_literal ] |= aci_custom;
93    TokenInfo[tok::utf8_string_literal ] |= aci_custom;
94    TokenInfo[tok::utf16_string_literal] |= aci_custom;
95    TokenInfo[tok::utf32_string_literal] |= aci_custom;
96    TokenInfo[tok::char_constant       ] |= aci_custom;
97    TokenInfo[tok::wide_char_constant  ] |= aci_custom;
98    TokenInfo[tok::utf16_char_constant ] |= aci_custom;
99    TokenInfo[tok::utf32_char_constant ] |= aci_custom;
100  }
101
102  // These tokens change behavior if followed by an '='.
103  TokenInfo[tok::amp         ] |= aci_avoid_equal;           // &=
104  TokenInfo[tok::plus        ] |= aci_avoid_equal;           // +=
105  TokenInfo[tok::minus       ] |= aci_avoid_equal;           // -=
106  TokenInfo[tok::slash       ] |= aci_avoid_equal;           // /=
107  TokenInfo[tok::less        ] |= aci_avoid_equal;           // <=
108  TokenInfo[tok::greater     ] |= aci_avoid_equal;           // >=
109  TokenInfo[tok::pipe        ] |= aci_avoid_equal;           // |=
110  TokenInfo[tok::percent     ] |= aci_avoid_equal;           // %=
111  TokenInfo[tok::star        ] |= aci_avoid_equal;           // *=
112  TokenInfo[tok::exclaim     ] |= aci_avoid_equal;           // !=
113  TokenInfo[tok::lessless    ] |= aci_avoid_equal;           // <<=
114  TokenInfo[tok::greatergreater] |= aci_avoid_equal;         // >>=
115  TokenInfo[tok::caret       ] |= aci_avoid_equal;           // ^=
116  TokenInfo[tok::equal       ] |= aci_avoid_equal;           // ==
117}
118
119/// GetFirstChar - Get the first character of the token \arg Tok,
120/// avoiding calls to getSpelling where possible.
121static char GetFirstChar(Preprocessor &PP, const Token &Tok) {
122  if (IdentifierInfo *II = Tok.getIdentifierInfo()) {
123    // Avoid spelling identifiers, the most common form of token.
124    return II->getNameStart()[0];
125  } else if (!Tok.needsCleaning()) {
126    if (Tok.isLiteral() && Tok.getLiteralData()) {
127      return *Tok.getLiteralData();
128    } else {
129      SourceManager &SM = PP.getSourceManager();
130      return *SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation()));
131    }
132  } else if (Tok.getLength() < 256) {
133    char Buffer[256];
134    const char *TokPtr = Buffer;
135    PP.getSpelling(Tok, TokPtr);
136    return TokPtr[0];
137  } else {
138    return PP.getSpelling(Tok)[0];
139  }
140}
141
142/// AvoidConcat - If printing PrevTok immediately followed by Tok would cause
143/// the two individual tokens to be lexed as a single token, return true
144/// (which causes a space to be printed between them).  This allows the output
145/// of -E mode to be lexed to the same token stream as lexing the input
146/// directly would.
147///
148/// This code must conservatively return true if it doesn't want to be 100%
149/// accurate.  This will cause the output to include extra space characters,
150/// but the resulting output won't have incorrect concatenations going on.
151/// Examples include "..", which we print with a space between, because we
152/// don't want to track enough to tell "x.." from "...".
153bool TokenConcatenation::AvoidConcat(const Token &PrevPrevTok,
154                                     const Token &PrevTok,
155                                     const Token &Tok) const {
156  // First, check to see if the tokens were directly adjacent in the original
157  // source.  If they were, it must be okay to stick them together: if there
158  // were an issue, the tokens would have been lexed differently.
159  SourceManager &SM = PP.getSourceManager();
160  SourceLocation PrevSpellLoc = SM.getSpellingLoc(PrevTok.getLocation());
161  SourceLocation SpellLoc = SM.getSpellingLoc(Tok.getLocation());
162  if (PrevSpellLoc.getLocWithOffset(PrevTok.getLength()) == SpellLoc)
163    return false;
164
165  tok::TokenKind PrevKind = PrevTok.getKind();
166  if (PrevTok.getIdentifierInfo())  // Language keyword or named operator.
167    PrevKind = tok::identifier;
168
169  // Look up information on when we should avoid concatenation with prevtok.
170  unsigned ConcatInfo = TokenInfo[PrevKind];
171
172  // If prevtok never causes a problem for anything after it, return quickly.
173  if (ConcatInfo == 0) return false;
174
175  if (ConcatInfo & aci_avoid_equal) {
176    // If the next token is '=' or '==', avoid concatenation.
177    if (Tok.is(tok::equal) || Tok.is(tok::equalequal))
178      return true;
179    ConcatInfo &= ~aci_avoid_equal;
180  }
181
182  if (ConcatInfo == 0) return false;
183
184  // Basic algorithm: we look at the first character of the second token, and
185  // determine whether it, if appended to the first token, would form (or
186  // would contribute) to a larger token if concatenated.
187  char FirstChar = 0;
188  if (ConcatInfo & aci_custom) {
189    // If the token does not need to know the first character, don't get it.
190  } else {
191    FirstChar = GetFirstChar(PP, Tok);
192  }
193
194  switch (PrevKind) {
195  default:
196    llvm_unreachable("InitAvoidConcatTokenInfo built wrong");
197
198  case tok::raw_identifier:
199    llvm_unreachable("tok::raw_identifier in non-raw lexing mode!");
200
201  case tok::string_literal:
202  case tok::wide_string_literal:
203  case tok::utf8_string_literal:
204  case tok::utf16_string_literal:
205  case tok::utf32_string_literal:
206  case tok::char_constant:
207  case tok::wide_char_constant:
208  case tok::utf16_char_constant:
209  case tok::utf32_char_constant:
210    if (!PP.getLangOpts().CPlusPlus11)
211      return false;
212
213    // In C++11, a string or character literal followed by an identifier is a
214    // single token.
215    if (Tok.getIdentifierInfo())
216      return true;
217
218    // A ud-suffix is an identifier. If the previous token ends with one, treat
219    // it as an identifier.
220    if (!PrevTok.hasUDSuffix())
221      return false;
222    // FALL THROUGH.
223  case tok::identifier:   // id+id or id+number or id+L"foo".
224    // id+'.'... will not append.
225    if (Tok.is(tok::numeric_constant))
226      return GetFirstChar(PP, Tok) != '.';
227
228    if (Tok.getIdentifierInfo() || Tok.is(tok::wide_string_literal) ||
229        Tok.is(tok::utf8_string_literal) || Tok.is(tok::utf16_string_literal) ||
230        Tok.is(tok::utf32_string_literal) || Tok.is(tok::wide_char_constant) ||
231        Tok.is(tok::utf16_char_constant) || Tok.is(tok::utf32_char_constant))
232      return true;
233
234    // If this isn't identifier + string, we're done.
235    if (Tok.isNot(tok::char_constant) && Tok.isNot(tok::string_literal))
236      return false;
237
238    // Otherwise, this is a narrow character or string.  If the *identifier*
239    // is a literal 'L', 'u8', 'u' or 'U', avoid pasting L "foo" -> L"foo".
240    return IsIdentifierStringPrefix(PrevTok);
241
242  case tok::numeric_constant:
243    return isPreprocessingNumberBody(FirstChar) ||
244           FirstChar == '+' || FirstChar == '-';
245  case tok::period:          // ..., .*, .1234
246    return (FirstChar == '.' && PrevPrevTok.is(tok::period)) ||
247           isDigit(FirstChar) ||
248           (PP.getLangOpts().CPlusPlus && FirstChar == '*');
249  case tok::amp:             // &&
250    return FirstChar == '&';
251  case tok::plus:            // ++
252    return FirstChar == '+';
253  case tok::minus:           // --, ->, ->*
254    return FirstChar == '-' || FirstChar == '>';
255  case tok::slash:           //, /*, //
256    return FirstChar == '*' || FirstChar == '/';
257  case tok::less:            // <<, <<=, <:, <%
258    return FirstChar == '<' || FirstChar == ':' || FirstChar == '%';
259  case tok::greater:         // >>, >>=
260    return FirstChar == '>';
261  case tok::pipe:            // ||
262    return FirstChar == '|';
263  case tok::percent:         // %>, %:
264    return FirstChar == '>' || FirstChar == ':';
265  case tok::colon:           // ::, :>
266    return FirstChar == '>' ||
267    (PP.getLangOpts().CPlusPlus && FirstChar == ':');
268  case tok::hash:            // ##, #@, %:%:
269    return FirstChar == '#' || FirstChar == '@' || FirstChar == '%';
270  case tok::arrow:           // ->*
271    return PP.getLangOpts().CPlusPlus && FirstChar == '*';
272  }
273}
274