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"
17 using namespace clang;
18
19
20 /// IsStringPrefix - Return true if Str is a string prefix.
21 /// 'L', 'u', 'U', or 'u8'. Including raw versions.
IsStringPrefix(StringRef Str,bool CPlusPlus0x)22 static 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.
IsIdentifierStringPrefix(const Token & Tok) const47 bool TokenConcatenation::IsIdentifierStringPrefix(const Token &Tok) const {
48 const LangOptions &LangOpts = PP.getLangOpts();
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
TokenConcatenation(Preprocessor & pp)69 TokenConcatenation::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 have custom code in C++11 mode.
89 if (PP.getLangOpts().CPlusPlus0x) {
90 TokenInfo[tok::string_literal ] |= aci_custom;
91 TokenInfo[tok::wide_string_literal ] |= aci_custom;
92 TokenInfo[tok::utf8_string_literal ] |= aci_custom;
93 TokenInfo[tok::utf16_string_literal] |= aci_custom;
94 TokenInfo[tok::utf32_string_literal] |= aci_custom;
95 TokenInfo[tok::char_constant ] |= aci_custom;
96 TokenInfo[tok::wide_char_constant ] |= aci_custom;
97 TokenInfo[tok::utf16_char_constant ] |= aci_custom;
98 TokenInfo[tok::utf32_char_constant ] |= aci_custom;
99 }
100
101 // These tokens change behavior if followed by an '='.
102 TokenInfo[tok::amp ] |= aci_avoid_equal; // &=
103 TokenInfo[tok::plus ] |= aci_avoid_equal; // +=
104 TokenInfo[tok::minus ] |= aci_avoid_equal; // -=
105 TokenInfo[tok::slash ] |= aci_avoid_equal; // /=
106 TokenInfo[tok::less ] |= aci_avoid_equal; // <=
107 TokenInfo[tok::greater ] |= aci_avoid_equal; // >=
108 TokenInfo[tok::pipe ] |= aci_avoid_equal; // |=
109 TokenInfo[tok::percent ] |= aci_avoid_equal; // %=
110 TokenInfo[tok::star ] |= aci_avoid_equal; // *=
111 TokenInfo[tok::exclaim ] |= aci_avoid_equal; // !=
112 TokenInfo[tok::lessless ] |= aci_avoid_equal; // <<=
113 TokenInfo[tok::greatergreater] |= aci_avoid_equal; // >>=
114 TokenInfo[tok::caret ] |= aci_avoid_equal; // ^=
115 TokenInfo[tok::equal ] |= aci_avoid_equal; // ==
116 }
117
118 /// GetFirstChar - Get the first character of the token \arg Tok,
119 /// avoiding calls to getSpelling where possible.
GetFirstChar(Preprocessor & PP,const Token & Tok)120 static char GetFirstChar(Preprocessor &PP, const Token &Tok) {
121 if (IdentifierInfo *II = Tok.getIdentifierInfo()) {
122 // Avoid spelling identifiers, the most common form of token.
123 return II->getNameStart()[0];
124 } else if (!Tok.needsCleaning()) {
125 if (Tok.isLiteral() && Tok.getLiteralData()) {
126 return *Tok.getLiteralData();
127 } else {
128 SourceManager &SM = PP.getSourceManager();
129 return *SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation()));
130 }
131 } else if (Tok.getLength() < 256) {
132 char Buffer[256];
133 const char *TokPtr = Buffer;
134 PP.getSpelling(Tok, TokPtr);
135 return TokPtr[0];
136 } else {
137 return PP.getSpelling(Tok)[0];
138 }
139 }
140
141 /// AvoidConcat - If printing PrevTok immediately followed by Tok would cause
142 /// the two individual tokens to be lexed as a single token, return true
143 /// (which causes a space to be printed between them). This allows the output
144 /// of -E mode to be lexed to the same token stream as lexing the input
145 /// directly would.
146 ///
147 /// This code must conservatively return true if it doesn't want to be 100%
148 /// accurate. This will cause the output to include extra space characters,
149 /// but the resulting output won't have incorrect concatenations going on.
150 /// Examples include "..", which we print with a space between, because we
151 /// don't want to track enough to tell "x.." from "...".
AvoidConcat(const Token & PrevPrevTok,const Token & PrevTok,const Token & Tok) const152 bool TokenConcatenation::AvoidConcat(const Token &PrevPrevTok,
153 const Token &PrevTok,
154 const Token &Tok) const {
155 // First, check to see if the tokens were directly adjacent in the original
156 // source. If they were, it must be okay to stick them together: if there
157 // were an issue, the tokens would have been lexed differently.
158 if (PrevTok.getLocation().isFileID() && Tok.getLocation().isFileID() &&
159 PrevTok.getLocation().getLocWithOffset(PrevTok.getLength()) ==
160 Tok.getLocation())
161 return false;
162
163 tok::TokenKind PrevKind = PrevTok.getKind();
164 if (PrevTok.getIdentifierInfo()) // Language keyword or named operator.
165 PrevKind = tok::identifier;
166
167 // Look up information on when we should avoid concatenation with prevtok.
168 unsigned ConcatInfo = TokenInfo[PrevKind];
169
170 // If prevtok never causes a problem for anything after it, return quickly.
171 if (ConcatInfo == 0) return false;
172
173 if (ConcatInfo & aci_avoid_equal) {
174 // If the next token is '=' or '==', avoid concatenation.
175 if (Tok.is(tok::equal) || Tok.is(tok::equalequal))
176 return true;
177 ConcatInfo &= ~aci_avoid_equal;
178 }
179
180 if (ConcatInfo == 0) return false;
181
182 // Basic algorithm: we look at the first character of the second token, and
183 // determine whether it, if appended to the first token, would form (or
184 // would contribute) to a larger token if concatenated.
185 char FirstChar = 0;
186 if (ConcatInfo & aci_custom) {
187 // If the token does not need to know the first character, don't get it.
188 } else {
189 FirstChar = GetFirstChar(PP, Tok);
190 }
191
192 switch (PrevKind) {
193 default:
194 llvm_unreachable("InitAvoidConcatTokenInfo built wrong");
195
196 case tok::raw_identifier:
197 llvm_unreachable("tok::raw_identifier in non-raw lexing mode!");
198
199 case tok::string_literal:
200 case tok::wide_string_literal:
201 case tok::utf8_string_literal:
202 case tok::utf16_string_literal:
203 case tok::utf32_string_literal:
204 case tok::char_constant:
205 case tok::wide_char_constant:
206 case tok::utf16_char_constant:
207 case tok::utf32_char_constant:
208 if (!PP.getLangOpts().CPlusPlus0x)
209 return false;
210
211 // In C++11, a string or character literal followed by an identifier is a
212 // single token.
213 if (Tok.getIdentifierInfo())
214 return true;
215
216 // A ud-suffix is an identifier. If the previous token ends with one, treat
217 // it as an identifier.
218 if (!PrevTok.hasUDSuffix())
219 return false;
220 // FALL THROUGH.
221 case tok::identifier: // id+id or id+number or id+L"foo".
222 // id+'.'... will not append.
223 if (Tok.is(tok::numeric_constant))
224 return GetFirstChar(PP, Tok) != '.';
225
226 if (Tok.getIdentifierInfo() || Tok.is(tok::wide_string_literal) ||
227 Tok.is(tok::utf8_string_literal) || Tok.is(tok::utf16_string_literal) ||
228 Tok.is(tok::utf32_string_literal) || Tok.is(tok::wide_char_constant) ||
229 Tok.is(tok::utf16_char_constant) || Tok.is(tok::utf32_char_constant))
230 return true;
231
232 // If this isn't identifier + string, we're done.
233 if (Tok.isNot(tok::char_constant) && Tok.isNot(tok::string_literal))
234 return false;
235
236 // Otherwise, this is a narrow character or string. If the *identifier*
237 // is a literal 'L', 'u8', 'u' or 'U', avoid pasting L "foo" -> L"foo".
238 return IsIdentifierStringPrefix(PrevTok);
239
240 case tok::numeric_constant:
241 return isalnum(FirstChar) || Tok.is(tok::numeric_constant) ||
242 FirstChar == '+' || FirstChar == '-' || FirstChar == '.' ||
243 (PP.getLangOpts().CPlusPlus0x && FirstChar == '_');
244 case tok::period: // ..., .*, .1234
245 return (FirstChar == '.' && PrevPrevTok.is(tok::period)) ||
246 isdigit(FirstChar) ||
247 (PP.getLangOpts().CPlusPlus && FirstChar == '*');
248 case tok::amp: // &&
249 return FirstChar == '&';
250 case tok::plus: // ++
251 return FirstChar == '+';
252 case tok::minus: // --, ->, ->*
253 return FirstChar == '-' || FirstChar == '>';
254 case tok::slash: //, /*, //
255 return FirstChar == '*' || FirstChar == '/';
256 case tok::less: // <<, <<=, <:, <%
257 return FirstChar == '<' || FirstChar == ':' || FirstChar == '%';
258 case tok::greater: // >>, >>=
259 return FirstChar == '>';
260 case tok::pipe: // ||
261 return FirstChar == '|';
262 case tok::percent: // %>, %:
263 return FirstChar == '>' || FirstChar == ':';
264 case tok::colon: // ::, :>
265 return FirstChar == '>' ||
266 (PP.getLangOpts().CPlusPlus && FirstChar == ':');
267 case tok::hash: // ##, #@, %:%:
268 return FirstChar == '#' || FirstChar == '@' || FirstChar == '%';
269 case tok::arrow: // ->*
270 return PP.getLangOpts().CPlusPlus && FirstChar == '*';
271 }
272 }
273