TokenConcatenation.cpp revision 218893 
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/// StartsWithL - Return true if the spelling of this token starts with 'L'.
21bool TokenConcatenation::StartsWithL(const Token &Tok) const {
22  if (!Tok.needsCleaning()) {
23    SourceManager &SM = PP.getSourceManager();
24    return *SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation())) == 'L';
25  }
26
27  if (Tok.getLength() < 256) {
28    char Buffer[256];
29    const char *TokPtr = Buffer;
30    PP.getSpelling(Tok, TokPtr);
31    return TokPtr[0] == 'L';
32  }
33
34  return PP.getSpelling(Tok)[0] == 'L';
35}
36
37/// IsIdentifierL - Return true if the spelling of this token is literally
38/// 'L'.
39bool TokenConcatenation::IsIdentifierL(const Token &Tok) const {
40  if (!Tok.needsCleaning()) {
41    if (Tok.getLength() != 1)
42      return false;
43    SourceManager &SM = PP.getSourceManager();
44    return *SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation())) == 'L';
45  }
46
47  if (Tok.getLength() < 256) {
48    char Buffer[256];
49    const char *TokPtr = Buffer;
50    if (PP.getSpelling(Tok, TokPtr) != 1)
51      return false;
52    return TokPtr[0] == 'L';
53  }
54
55  return PP.getSpelling(Tok) == "L";
56}
57
58TokenConcatenation::TokenConcatenation(Preprocessor &pp) : PP(pp) {
59  memset(TokenInfo, 0, sizeof(TokenInfo));
60
61  // These tokens have custom code in AvoidConcat.
62  TokenInfo[tok::identifier      ] |= aci_custom;
63  TokenInfo[tok::numeric_constant] |= aci_custom_firstchar;
64  TokenInfo[tok::period          ] |= aci_custom_firstchar;
65  TokenInfo[tok::amp             ] |= aci_custom_firstchar;
66  TokenInfo[tok::plus            ] |= aci_custom_firstchar;
67  TokenInfo[tok::minus           ] |= aci_custom_firstchar;
68  TokenInfo[tok::slash           ] |= aci_custom_firstchar;
69  TokenInfo[tok::less            ] |= aci_custom_firstchar;
70  TokenInfo[tok::greater         ] |= aci_custom_firstchar;
71  TokenInfo[tok::pipe            ] |= aci_custom_firstchar;
72  TokenInfo[tok::percent         ] |= aci_custom_firstchar;
73  TokenInfo[tok::colon           ] |= aci_custom_firstchar;
74  TokenInfo[tok::hash            ] |= aci_custom_firstchar;
75  TokenInfo[tok::arrow           ] |= aci_custom_firstchar;
76
77  // These tokens change behavior if followed by an '='.
78  TokenInfo[tok::amp         ] |= aci_avoid_equal;           // &=
79  TokenInfo[tok::plus        ] |= aci_avoid_equal;           // +=
80  TokenInfo[tok::minus       ] |= aci_avoid_equal;           // -=
81  TokenInfo[tok::slash       ] |= aci_avoid_equal;           // /=
82  TokenInfo[tok::less        ] |= aci_avoid_equal;           // <=
83  TokenInfo[tok::greater     ] |= aci_avoid_equal;           // >=
84  TokenInfo[tok::pipe        ] |= aci_avoid_equal;           // |=
85  TokenInfo[tok::percent     ] |= aci_avoid_equal;           // %=
86  TokenInfo[tok::star        ] |= aci_avoid_equal;           // *=
87  TokenInfo[tok::exclaim     ] |= aci_avoid_equal;           // !=
88  TokenInfo[tok::lessless    ] |= aci_avoid_equal;           // <<=
89  TokenInfo[tok::greatergreater] |= aci_avoid_equal;         // >>=
90  TokenInfo[tok::caret       ] |= aci_avoid_equal;           // ^=
91  TokenInfo[tok::equal       ] |= aci_avoid_equal;           // ==
92}
93
94/// GetFirstChar - Get the first character of the token \arg Tok,
95/// avoiding calls to getSpelling where possible.
96static char GetFirstChar(Preprocessor &PP, const Token &Tok) {
97  if (IdentifierInfo *II = Tok.getIdentifierInfo()) {
98    // Avoid spelling identifiers, the most common form of token.
99    return II->getNameStart()[0];
100  } else if (!Tok.needsCleaning()) {
101    if (Tok.isLiteral() && Tok.getLiteralData()) {
102      return *Tok.getLiteralData();
103    } else {
104      SourceManager &SM = PP.getSourceManager();
105      return *SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation()));
106    }
107  } else if (Tok.getLength() < 256) {
108    char Buffer[256];
109    const char *TokPtr = Buffer;
110    PP.getSpelling(Tok, TokPtr);
111    return TokPtr[0];
112  } else {
113    return PP.getSpelling(Tok)[0];
114  }
115}
116
117/// AvoidConcat - If printing PrevTok immediately followed by Tok would cause
118/// the two individual tokens to be lexed as a single token, return true
119/// (which causes a space to be printed between them).  This allows the output
120/// of -E mode to be lexed to the same token stream as lexing the input
121/// directly would.
122///
123/// This code must conservatively return true if it doesn't want to be 100%
124/// accurate.  This will cause the output to include extra space characters,
125/// but the resulting output won't have incorrect concatenations going on.
126/// Examples include "..", which we print with a space between, because we
127/// don't want to track enough to tell "x.." from "...".
128bool TokenConcatenation::AvoidConcat(const Token &PrevPrevTok,
129                                     const Token &PrevTok,
130                                     const Token &Tok) const {
131  // First, check to see if the tokens were directly adjacent in the original
132  // source.  If they were, it must be okay to stick them together: if there
133  // were an issue, the tokens would have been lexed differently.
134  if (PrevTok.getLocation().isFileID() && Tok.getLocation().isFileID() &&
135      PrevTok.getLocation().getFileLocWithOffset(PrevTok.getLength()) ==
136        Tok.getLocation())
137    return false;
138
139  tok::TokenKind PrevKind = PrevTok.getKind();
140  if (PrevTok.getIdentifierInfo())  // Language keyword or named operator.
141    PrevKind = tok::identifier;
142
143  // Look up information on when we should avoid concatenation with prevtok.
144  unsigned ConcatInfo = TokenInfo[PrevKind];
145
146  // If prevtok never causes a problem for anything after it, return quickly.
147  if (ConcatInfo == 0) return false;
148
149  if (ConcatInfo & aci_avoid_equal) {
150    // If the next token is '=' or '==', avoid concatenation.
151    if (Tok.is(tok::equal) || Tok.is(tok::equalequal))
152      return true;
153    ConcatInfo &= ~aci_avoid_equal;
154  }
155
156  if (ConcatInfo == 0) return false;
157
158  // Basic algorithm: we look at the first character of the second token, and
159  // determine whether it, if appended to the first token, would form (or
160  // would contribute) to a larger token if concatenated.
161  char FirstChar = 0;
162  if (ConcatInfo & aci_custom) {
163    // If the token does not need to know the first character, don't get it.
164  } else {
165    FirstChar = GetFirstChar(PP, Tok);
166  }
167
168  switch (PrevKind) {
169  default:
170    llvm_unreachable("InitAvoidConcatTokenInfo built wrong");
171    return true;
172
173  case tok::raw_identifier:
174    llvm_unreachable("tok::raw_identifier in non-raw lexing mode!");
175    return true;
176
177  case tok::identifier:   // id+id or id+number or id+L"foo".
178    // id+'.'... will not append.
179    if (Tok.is(tok::numeric_constant))
180      return GetFirstChar(PP, Tok) != '.';
181
182    if (Tok.getIdentifierInfo() || Tok.is(tok::wide_string_literal) /* ||
183     Tok.is(tok::wide_char_literal)*/)
184      return true;
185
186    // If this isn't identifier + string, we're done.
187    if (Tok.isNot(tok::char_constant) && Tok.isNot(tok::string_literal))
188      return false;
189
190    // FIXME: need a wide_char_constant!
191
192    // If the string was a wide string L"foo" or wide char L'f', it would
193    // concat with the previous identifier into fooL"bar".  Avoid this.
194    if (StartsWithL(Tok))
195      return true;
196
197    // Otherwise, this is a narrow character or string.  If the *identifier*
198    // is a literal 'L', avoid pasting L "foo" -> L"foo".
199    return IsIdentifierL(PrevTok);
200  case tok::numeric_constant:
201    return isalnum(FirstChar) || Tok.is(tok::numeric_constant) ||
202    FirstChar == '+' || FirstChar == '-' || FirstChar == '.';
203  case tok::period:          // ..., .*, .1234
204    return (FirstChar == '.' && PrevPrevTok.is(tok::period)) ||
205    isdigit(FirstChar) ||
206    (PP.getLangOptions().CPlusPlus && FirstChar == '*');
207  case tok::amp:             // &&
208    return FirstChar == '&';
209  case tok::plus:            // ++
210    return FirstChar == '+';
211  case tok::minus:           // --, ->, ->*
212    return FirstChar == '-' || FirstChar == '>';
213  case tok::slash:           //, /*, //
214    return FirstChar == '*' || FirstChar == '/';
215  case tok::less:            // <<, <<=, <:, <%
216    return FirstChar == '<' || FirstChar == ':' || FirstChar == '%';
217  case tok::greater:         // >>, >>=
218    return FirstChar == '>';
219  case tok::pipe:            // ||
220    return FirstChar == '|';
221  case tok::percent:         // %>, %:
222    return FirstChar == '>' || FirstChar == ':';
223  case tok::colon:           // ::, :>
224    return FirstChar == '>' ||
225    (PP.getLangOptions().CPlusPlus && FirstChar == ':');
226  case tok::hash:            // ##, #@, %:%:
227    return FirstChar == '#' || FirstChar == '@' || FirstChar == '%';
228  case tok::arrow:           // ->*
229    return PP.getLangOptions().CPlusPlus && FirstChar == '*';
230  }
231}
232