TokenConcatenation.cpp revision 206084 
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"
16using namespace clang;
17
18
19/// StartsWithL - Return true if the spelling of this token starts with 'L'.
20bool TokenConcatenation::StartsWithL(const Token &Tok) const {
21  if (!Tok.needsCleaning()) {
22    SourceManager &SM = PP.getSourceManager();
23    return *SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation())) == 'L';
24  }
25
26  if (Tok.getLength() < 256) {
27    char Buffer[256];
28    const char *TokPtr = Buffer;
29    PP.getSpelling(Tok, TokPtr);
30    return TokPtr[0] == 'L';
31  }
32
33  return PP.getSpelling(Tok)[0] == 'L';
34}
35
36/// IsIdentifierL - Return true if the spelling of this token is literally
37/// 'L'.
38bool TokenConcatenation::IsIdentifierL(const Token &Tok) const {
39  if (!Tok.needsCleaning()) {
40    if (Tok.getLength() != 1)
41      return false;
42    SourceManager &SM = PP.getSourceManager();
43    return *SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation())) == 'L';
44  }
45
46  if (Tok.getLength() < 256) {
47    char Buffer[256];
48    const char *TokPtr = Buffer;
49    if (PP.getSpelling(Tok, TokPtr) != 1)
50      return false;
51    return TokPtr[0] == 'L';
52  }
53
54  return PP.getSpelling(Tok) == "L";
55}
56
57TokenConcatenation::TokenConcatenation(Preprocessor &pp) : PP(pp) {
58  memset(TokenInfo, 0, sizeof(TokenInfo));
59
60  // These tokens have custom code in AvoidConcat.
61  TokenInfo[tok::identifier      ] |= aci_custom;
62  TokenInfo[tok::numeric_constant] |= aci_custom_firstchar;
63  TokenInfo[tok::period          ] |= aci_custom_firstchar;
64  TokenInfo[tok::amp             ] |= aci_custom_firstchar;
65  TokenInfo[tok::plus            ] |= aci_custom_firstchar;
66  TokenInfo[tok::minus           ] |= aci_custom_firstchar;
67  TokenInfo[tok::slash           ] |= aci_custom_firstchar;
68  TokenInfo[tok::less            ] |= aci_custom_firstchar;
69  TokenInfo[tok::greater         ] |= aci_custom_firstchar;
70  TokenInfo[tok::pipe            ] |= aci_custom_firstchar;
71  TokenInfo[tok::percent         ] |= aci_custom_firstchar;
72  TokenInfo[tok::colon           ] |= aci_custom_firstchar;
73  TokenInfo[tok::hash            ] |= aci_custom_firstchar;
74  TokenInfo[tok::arrow           ] |= aci_custom_firstchar;
75
76  // These tokens change behavior if followed by an '='.
77  TokenInfo[tok::amp         ] |= aci_avoid_equal;           // &=
78  TokenInfo[tok::plus        ] |= aci_avoid_equal;           // +=
79  TokenInfo[tok::minus       ] |= aci_avoid_equal;           // -=
80  TokenInfo[tok::slash       ] |= aci_avoid_equal;           // /=
81  TokenInfo[tok::less        ] |= aci_avoid_equal;           // <=
82  TokenInfo[tok::greater     ] |= aci_avoid_equal;           // >=
83  TokenInfo[tok::pipe        ] |= aci_avoid_equal;           // |=
84  TokenInfo[tok::percent     ] |= aci_avoid_equal;           // %=
85  TokenInfo[tok::star        ] |= aci_avoid_equal;           // *=
86  TokenInfo[tok::exclaim     ] |= aci_avoid_equal;           // !=
87  TokenInfo[tok::lessless    ] |= aci_avoid_equal;           // <<=
88  TokenInfo[tok::greatergreater] |= aci_avoid_equal;         // >>=
89  TokenInfo[tok::caret       ] |= aci_avoid_equal;           // ^=
90  TokenInfo[tok::equal       ] |= aci_avoid_equal;           // ==
91}
92
93/// GetFirstChar - Get the first character of the token \arg Tok,
94/// avoiding calls to getSpelling where possible.
95static char GetFirstChar(Preprocessor &PP, const Token &Tok) {
96  if (IdentifierInfo *II = Tok.getIdentifierInfo()) {
97    // Avoid spelling identifiers, the most common form of token.
98    return II->getNameStart()[0];
99  } else if (!Tok.needsCleaning()) {
100    if (Tok.isLiteral() && Tok.getLiteralData()) {
101      return *Tok.getLiteralData();
102    } else {
103      SourceManager &SM = PP.getSourceManager();
104      return *SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation()));
105    }
106  } else if (Tok.getLength() < 256) {
107    char Buffer[256];
108    const char *TokPtr = Buffer;
109    PP.getSpelling(Tok, TokPtr);
110    return TokPtr[0];
111  } else {
112    return PP.getSpelling(Tok)[0];
113  }
114}
115
116/// AvoidConcat - If printing PrevTok immediately followed by Tok would cause
117/// the two individual tokens to be lexed as a single token, return true
118/// (which causes a space to be printed between them).  This allows the output
119/// of -E mode to be lexed to the same token stream as lexing the input
120/// directly would.
121///
122/// This code must conservatively return true if it doesn't want to be 100%
123/// accurate.  This will cause the output to include extra space characters,
124/// but the resulting output won't have incorrect concatenations going on.
125/// Examples include "..", which we print with a space between, because we
126/// don't want to track enough to tell "x.." from "...".
127bool TokenConcatenation::AvoidConcat(const Token &PrevTok,
128                                     const Token &Tok) const {
129  // First, check to see if the tokens were directly adjacent in the original
130  // source.  If they were, it must be okay to stick them together: if there
131  // were an issue, the tokens would have been lexed differently.
132  if (PrevTok.getLocation().isFileID() && Tok.getLocation().isFileID() &&
133      PrevTok.getLocation().getFileLocWithOffset(PrevTok.getLength()) ==
134        Tok.getLocation())
135    return false;
136
137  tok::TokenKind PrevKind = PrevTok.getKind();
138  if (PrevTok.getIdentifierInfo())  // Language keyword or named operator.
139    PrevKind = tok::identifier;
140
141  // Look up information on when we should avoid concatenation with prevtok.
142  unsigned ConcatInfo = TokenInfo[PrevKind];
143
144  // If prevtok never causes a problem for anything after it, return quickly.
145  if (ConcatInfo == 0) return false;
146
147  if (ConcatInfo & aci_avoid_equal) {
148    // If the next token is '=' or '==', avoid concatenation.
149    if (Tok.is(tok::equal) || Tok.is(tok::equalequal))
150      return true;
151    ConcatInfo &= ~aci_avoid_equal;
152  }
153
154  if (ConcatInfo == 0) return false;
155
156  // Basic algorithm: we look at the first character of the second token, and
157  // determine whether it, if appended to the first token, would form (or
158  // would contribute) to a larger token if concatenated.
159  char FirstChar = 0;
160  if (ConcatInfo & aci_custom) {
161    // If the token does not need to know the first character, don't get it.
162  } else {
163    FirstChar = GetFirstChar(PP, Tok);
164  }
165
166  switch (PrevKind) {
167  default: assert(0 && "InitAvoidConcatTokenInfo built wrong");
168  case tok::identifier:   // id+id or id+number or id+L"foo".
169    // id+'.'... will not append.
170    if (Tok.is(tok::numeric_constant))
171      return GetFirstChar(PP, Tok) != '.';
172
173    if (Tok.getIdentifierInfo() || Tok.is(tok::wide_string_literal) /* ||
174     Tok.is(tok::wide_char_literal)*/)
175      return true;
176
177    // If this isn't identifier + string, we're done.
178    if (Tok.isNot(tok::char_constant) && Tok.isNot(tok::string_literal))
179      return false;
180
181    // FIXME: need a wide_char_constant!
182
183    // If the string was a wide string L"foo" or wide char L'f', it would
184    // concat with the previous identifier into fooL"bar".  Avoid this.
185    if (StartsWithL(Tok))
186      return true;
187
188    // Otherwise, this is a narrow character or string.  If the *identifier*
189    // is a literal 'L', avoid pasting L "foo" -> L"foo".
190    return IsIdentifierL(PrevTok);
191  case tok::numeric_constant:
192    return isalnum(FirstChar) || Tok.is(tok::numeric_constant) ||
193    FirstChar == '+' || FirstChar == '-' || FirstChar == '.';
194  case tok::period:          // ..., .*, .1234
195    return FirstChar == '.' || isdigit(FirstChar) ||
196    (PP.getLangOptions().CPlusPlus && FirstChar == '*');
197  case tok::amp:             // &&
198    return FirstChar == '&';
199  case tok::plus:            // ++
200    return FirstChar == '+';
201  case tok::minus:           // --, ->, ->*
202    return FirstChar == '-' || FirstChar == '>';
203  case tok::slash:           //, /*, //
204    return FirstChar == '*' || FirstChar == '/';
205  case tok::less:            // <<, <<=, <:, <%
206    return FirstChar == '<' || FirstChar == ':' || FirstChar == '%';
207  case tok::greater:         // >>, >>=
208    return FirstChar == '>';
209  case tok::pipe:            // ||
210    return FirstChar == '|';
211  case tok::percent:         // %>, %:
212    return FirstChar == '>' || FirstChar == ':';
213  case tok::colon:           // ::, :>
214    return FirstChar == '>' ||
215    (PP.getLangOptions().CPlusPlus && FirstChar == ':');
216  case tok::hash:            // ##, #@, %:%:
217    return FirstChar == '#' || FirstChar == '@' || FirstChar == '%';
218  case tok::arrow:           // ->*
219    return PP.getLangOptions().CPlusPlus && FirstChar == '*';
220  }
221}
222