TokenConcatenation.cpp revision 226633 
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/Lex/Preprocessor.h"
16#include "llvm/Support/ErrorHandling.h"
17using namespace clang;
18
19
20/// IsStringPrefix - Return true if Str is a string prefix.
21/// 'L', 'u', 'U', or 'u8'. Including raw versions.
22static bool IsStringPrefix(StringRef Str, bool CPlusPlus0x) {
23
24  if (Str[0] == 'L' ||
25      (CPlusPlus0x && (Str[0] == 'u' || Str[0] == 'U' || Str[0] == 'R'))) {
26
27    if (Str.size() == 1)
28      return true; // "L", "u", "U", and "R"
29
30    // Check for raw flavors. Need to make sure the first character wasn't
31    // already R. Need CPlusPlus0x check for "LR".
32    if (Str[1] == 'R' && Str[0] != 'R' && Str.size() == 2 && CPlusPlus0x)
33      return true; // "LR", "uR", "UR"
34
35    // Check for "u8" and "u8R"
36    if (Str[0] == 'u' && Str[1] == '8') {
37      if (Str.size() == 2) return true; // "u8"
38      if (Str.size() == 3 && Str[2] == 'R') return true; // "u8R"
39    }
40  }
41
42  return false;
43}
44
45/// IsIdentifierStringPrefix - Return true if the spelling of the token
46/// is literally 'L', 'u', 'U', or 'u8'. Including raw versions.
47bool TokenConcatenation::IsIdentifierStringPrefix(const Token &Tok) const {
48  const LangOptions &LangOpts = PP.getLangOptions();
49
50  if (!Tok.needsCleaning()) {
51    if (Tok.getLength() < 1 || Tok.getLength() > 3)
52      return false;
53    SourceManager &SM = PP.getSourceManager();
54    const char *Ptr = SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation()));
55    return IsStringPrefix(StringRef(Ptr, Tok.getLength()),
56                          LangOpts.CPlusPlus0x);
57  }
58
59  if (Tok.getLength() < 256) {
60    char Buffer[256];
61    const char *TokPtr = Buffer;
62    unsigned length = PP.getSpelling(Tok, TokPtr);
63    return IsStringPrefix(StringRef(TokPtr, length), LangOpts.CPlusPlus0x);
64  }
65
66  return IsStringPrefix(StringRef(PP.getSpelling(Tok)), LangOpts.CPlusPlus0x);
67}
68
69TokenConcatenation::TokenConcatenation(Preprocessor &pp) : PP(pp) {
70  memset(TokenInfo, 0, sizeof(TokenInfo));
71
72  // These tokens have custom code in AvoidConcat.
73  TokenInfo[tok::identifier      ] |= aci_custom;
74  TokenInfo[tok::numeric_constant] |= aci_custom_firstchar;
75  TokenInfo[tok::period          ] |= aci_custom_firstchar;
76  TokenInfo[tok::amp             ] |= aci_custom_firstchar;
77  TokenInfo[tok::plus            ] |= aci_custom_firstchar;
78  TokenInfo[tok::minus           ] |= aci_custom_firstchar;
79  TokenInfo[tok::slash           ] |= aci_custom_firstchar;
80  TokenInfo[tok::less            ] |= aci_custom_firstchar;
81  TokenInfo[tok::greater         ] |= aci_custom_firstchar;
82  TokenInfo[tok::pipe            ] |= aci_custom_firstchar;
83  TokenInfo[tok::percent         ] |= aci_custom_firstchar;
84  TokenInfo[tok::colon           ] |= aci_custom_firstchar;
85  TokenInfo[tok::hash            ] |= aci_custom_firstchar;
86  TokenInfo[tok::arrow           ] |= aci_custom_firstchar;
87
88  // These tokens change behavior if followed by an '='.
89  TokenInfo[tok::amp         ] |= aci_avoid_equal;           // &=
90  TokenInfo[tok::plus        ] |= aci_avoid_equal;           // +=
91  TokenInfo[tok::minus       ] |= aci_avoid_equal;           // -=
92  TokenInfo[tok::slash       ] |= aci_avoid_equal;           // /=
93  TokenInfo[tok::less        ] |= aci_avoid_equal;           // <=
94  TokenInfo[tok::greater     ] |= aci_avoid_equal;           // >=
95  TokenInfo[tok::pipe        ] |= aci_avoid_equal;           // |=
96  TokenInfo[tok::percent     ] |= aci_avoid_equal;           // %=
97  TokenInfo[tok::star        ] |= aci_avoid_equal;           // *=
98  TokenInfo[tok::exclaim     ] |= aci_avoid_equal;           // !=
99  TokenInfo[tok::lessless    ] |= aci_avoid_equal;           // <<=
100  TokenInfo[tok::greatergreater] |= aci_avoid_equal;         // >>=
101  TokenInfo[tok::caret       ] |= aci_avoid_equal;           // ^=
102  TokenInfo[tok::equal       ] |= aci_avoid_equal;           // ==
103}
104
105/// GetFirstChar - Get the first character of the token \arg Tok,
106/// avoiding calls to getSpelling where possible.
107static char GetFirstChar(Preprocessor &PP, const Token &Tok) {
108  if (IdentifierInfo *II = Tok.getIdentifierInfo()) {
109    // Avoid spelling identifiers, the most common form of token.
110    return II->getNameStart()[0];
111  } else if (!Tok.needsCleaning()) {
112    if (Tok.isLiteral() && Tok.getLiteralData()) {
113      return *Tok.getLiteralData();
114    } else {
115      SourceManager &SM = PP.getSourceManager();
116      return *SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation()));
117    }
118  } else if (Tok.getLength() < 256) {
119    char Buffer[256];
120    const char *TokPtr = Buffer;
121    PP.getSpelling(Tok, TokPtr);
122    return TokPtr[0];
123  } else {
124    return PP.getSpelling(Tok)[0];
125  }
126}
127
128/// AvoidConcat - If printing PrevTok immediately followed by Tok would cause
129/// the two individual tokens to be lexed as a single token, return true
130/// (which causes a space to be printed between them).  This allows the output
131/// of -E mode to be lexed to the same token stream as lexing the input
132/// directly would.
133///
134/// This code must conservatively return true if it doesn't want to be 100%
135/// accurate.  This will cause the output to include extra space characters,
136/// but the resulting output won't have incorrect concatenations going on.
137/// Examples include "..", which we print with a space between, because we
138/// don't want to track enough to tell "x.." from "...".
139bool TokenConcatenation::AvoidConcat(const Token &PrevPrevTok,
140                                     const Token &PrevTok,
141                                     const Token &Tok) const {
142  // First, check to see if the tokens were directly adjacent in the original
143  // source.  If they were, it must be okay to stick them together: if there
144  // were an issue, the tokens would have been lexed differently.
145  if (PrevTok.getLocation().isFileID() && Tok.getLocation().isFileID() &&
146      PrevTok.getLocation().getLocWithOffset(PrevTok.getLength()) ==
147        Tok.getLocation())
148    return false;
149
150  tok::TokenKind PrevKind = PrevTok.getKind();
151  if (PrevTok.getIdentifierInfo())  // Language keyword or named operator.
152    PrevKind = tok::identifier;
153
154  // Look up information on when we should avoid concatenation with prevtok.
155  unsigned ConcatInfo = TokenInfo[PrevKind];
156
157  // If prevtok never causes a problem for anything after it, return quickly.
158  if (ConcatInfo == 0) return false;
159
160  if (ConcatInfo & aci_avoid_equal) {
161    // If the next token is '=' or '==', avoid concatenation.
162    if (Tok.is(tok::equal) || Tok.is(tok::equalequal))
163      return true;
164    ConcatInfo &= ~aci_avoid_equal;
165  }
166
167  if (ConcatInfo == 0) return false;
168
169  // Basic algorithm: we look at the first character of the second token, and
170  // determine whether it, if appended to the first token, would form (or
171  // would contribute) to a larger token if concatenated.
172  char FirstChar = 0;
173  if (ConcatInfo & aci_custom) {
174    // If the token does not need to know the first character, don't get it.
175  } else {
176    FirstChar = GetFirstChar(PP, Tok);
177  }
178
179  switch (PrevKind) {
180  default:
181    llvm_unreachable("InitAvoidConcatTokenInfo built wrong");
182    return true;
183
184  case tok::raw_identifier:
185    llvm_unreachable("tok::raw_identifier in non-raw lexing mode!");
186    return true;
187
188  case tok::identifier:   // id+id or id+number or id+L"foo".
189    // id+'.'... will not append.
190    if (Tok.is(tok::numeric_constant))
191      return GetFirstChar(PP, Tok) != '.';
192
193    if (Tok.getIdentifierInfo() || Tok.is(tok::wide_string_literal) ||
194        Tok.is(tok::utf8_string_literal) || Tok.is(tok::utf16_string_literal) ||
195        Tok.is(tok::utf32_string_literal) || Tok.is(tok::wide_char_constant) ||
196        Tok.is(tok::utf16_char_constant) || Tok.is(tok::utf32_char_constant))
197      return true;
198
199    // If this isn't identifier + string, we're done.
200    if (Tok.isNot(tok::char_constant) && Tok.isNot(tok::string_literal))
201      return false;
202
203    // Otherwise, this is a narrow character or string.  If the *identifier*
204    // is a literal 'L', 'u8', 'u' or 'U', avoid pasting L "foo" -> L"foo".
205    return IsIdentifierStringPrefix(PrevTok);
206  case tok::numeric_constant:
207    return isalnum(FirstChar) || Tok.is(tok::numeric_constant) ||
208    FirstChar == '+' || FirstChar == '-' || FirstChar == '.';
209  case tok::period:          // ..., .*, .1234
210    return (FirstChar == '.' && PrevPrevTok.is(tok::period)) ||
211    isdigit(FirstChar) ||
212    (PP.getLangOptions().CPlusPlus && FirstChar == '*');
213  case tok::amp:             // &&
214    return FirstChar == '&';
215  case tok::plus:            // ++
216    return FirstChar == '+';
217  case tok::minus:           // --, ->, ->*
218    return FirstChar == '-' || FirstChar == '>';
219  case tok::slash:           //, /*, //
220    return FirstChar == '*' || FirstChar == '/';
221  case tok::less:            // <<, <<=, <:, <%
222    return FirstChar == '<' || FirstChar == ':' || FirstChar == '%';
223  case tok::greater:         // >>, >>=
224    return FirstChar == '>';
225  case tok::pipe:            // ||
226    return FirstChar == '|';
227  case tok::percent:         // %>, %:
228    return FirstChar == '>' || FirstChar == ':';
229  case tok::colon:           // ::, :>
230    return FirstChar == '>' ||
231    (PP.getLangOptions().CPlusPlus && FirstChar == ':');
232  case tok::hash:            // ##, #@, %:%:
233    return FirstChar == '#' || FirstChar == '@' || FirstChar == '%';
234  case tok::arrow:           // ->*
235    return PP.getLangOptions().CPlusPlus && FirstChar == '*';
236  }
237}
238