Coverage Report

Created: 2021-08-24 07:12

/Users/buildslave/jenkins/workspace/coverage/llvm-project/clang/lib/Tooling/Inclusions/HeaderIncludes.cpp
Line
Count
Source (jump to first uncovered line)
1
//===--- HeaderIncludes.cpp - Insert/Delete #includes --*- C++ -*----------===//
2
//
3
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4
// See https://llvm.org/LICENSE.txt for license information.
5
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6
//
7
//===----------------------------------------------------------------------===//
8
9
#include "clang/Tooling/Inclusions/HeaderIncludes.h"
10
#include "clang/Basic/FileManager.h"
11
#include "clang/Basic/SourceManager.h"
12
#include "clang/Lex/Lexer.h"
13
#include "llvm/ADT/Optional.h"
14
#include "llvm/Support/FormatVariadic.h"
15
#include "llvm/Support/Path.h"
16
17
namespace clang {
18
namespace tooling {
19
namespace {
20
21
180
LangOptions createLangOpts() {
22
180
  LangOptions LangOpts;
23
180
  LangOpts.CPlusPlus = 1;
24
180
  LangOpts.CPlusPlus11 = 1;
25
180
  LangOpts.CPlusPlus14 = 1;
26
180
  LangOpts.LineComment = 1;
27
180
  LangOpts.CXXOperatorNames = 1;
28
180
  LangOpts.Bool = 1;
29
180
  LangOpts.ObjC = 1;
30
180
  LangOpts.MicrosoftExt = 1;    // To get kw___try, kw___finally.
31
180
  LangOpts.DeclSpecKeyword = 1; // To get __declspec.
32
180
  LangOpts.WChar = 1;           // To get wchar_t
33
180
  return LangOpts;
34
180
}
35
36
// Returns the offset after skipping a sequence of tokens, matched by \p
37
// GetOffsetAfterSequence, from the start of the code.
38
// \p GetOffsetAfterSequence should be a function that matches a sequence of
39
// tokens and returns an offset after the sequence.
40
unsigned getOffsetAfterTokenSequence(
41
    StringRef FileName, StringRef Code, const IncludeStyle &Style,
42
    llvm::function_ref<unsigned(const SourceManager &, Lexer &, Token &)>
43
180
        GetOffsetAfterSequence) {
44
180
  SourceManagerForFile VirtualSM(FileName, Code);
45
180
  SourceManager &SM = VirtualSM.get();
46
180
  Lexer Lex(SM.getMainFileID(), SM.getBufferOrFake(SM.getMainFileID()), SM,
47
180
            createLangOpts());
48
180
  Token Tok;
49
  // Get the first token.
50
180
  Lex.LexFromRawLexer(Tok);
51
180
  return GetOffsetAfterSequence(SM, Lex, Tok);
52
180
}
53
54
// Check if a sequence of tokens is like "#<Name> <raw_identifier>". If it is,
55
// \p Tok will be the token after this directive; otherwise, it can be any token
56
// after the given \p Tok (including \p Tok). If \p RawIDName is provided, the
57
// (second) raw_identifier name is checked.
58
bool checkAndConsumeDirectiveWithName(
59
    Lexer &Lex, StringRef Name, Token &Tok,
60
128
    llvm::Optional<StringRef> RawIDName = llvm::None) {
61
128
  bool Matched = Tok.is(tok::hash) && 
!Lex.LexFromRawLexer(Tok)105
&&
62
128
                 
Tok.is(tok::raw_identifier)105
&&
63
128
                 
Tok.getRawIdentifier() == Name105
&&
!Lex.LexFromRawLexer(Tok)16
&&
64
128
                 
Tok.is(tok::raw_identifier)16
&&
65
128
                 
(15
!RawIDName15
||
Tok.getRawIdentifier() == *RawIDName1
);
66
128
  if (Matched)
67
15
    Lex.LexFromRawLexer(Tok);
68
128
  return Matched;
69
128
}
70
71
188
void skipComments(Lexer &Lex, Token &Tok) {
72
188
  while (Tok.is(tok::comment))
73
0
    if (Lex.LexFromRawLexer(Tok))
74
0
      return;
75
188
}
76
77
// Returns the offset after header guard directives and any comments
78
// before/after header guards (e.g. #ifndef/#define pair, #pragma once). If no
79
// header guard is present in the code, this will return the offset after
80
// skipping all comments from the start of the code.
81
unsigned getOffsetAfterHeaderGuardsAndComments(StringRef FileName,
82
                                               StringRef Code,
83
60
                                               const IncludeStyle &Style) {
84
  // \p Consume returns location after header guard or 0 if no header guard is
85
  // found.
86
60
  auto ConsumeHeaderGuardAndComment =
87
60
      [&](std::function<unsigned(const SourceManager &SM, Lexer &Lex,
88
60
                                 Token Tok)>
89
120
              Consume) {
90
120
        return getOffsetAfterTokenSequence(
91
120
            FileName, Code, Style,
92
120
            [&Consume](const SourceManager &SM, Lexer &Lex, Token Tok) {
93
120
              skipComments(Lex, Tok);
94
120
              unsigned InitialOffset = SM.getFileOffset(Tok.getLocation());
95
120
              return std::max(InitialOffset, Consume(SM, Lex, Tok));
96
120
            });
97
120
      };
98
60
  return std::max(
99
      // #ifndef/#define
100
60
      ConsumeHeaderGuardAndComment(
101
60
          [](const SourceManager &SM, Lexer &Lex, Token Tok) -> unsigned {
102
60
            if (checkAndConsumeDirectiveWithName(Lex, "ifndef", Tok)) {
103
8
              skipComments(Lex, Tok);
104
8
              if (checkAndConsumeDirectiveWithName(Lex, "define", Tok) &&
105
8
                  
Tok.isAtStartOfLine()6
)
106
5
                return SM.getFileOffset(Tok.getLocation());
107
8
            }
108
55
            return 0;
109
60
          }),
110
      // #pragma once
111
60
      ConsumeHeaderGuardAndComment(
112
60
          [](const SourceManager &SM, Lexer &Lex, Token Tok) -> unsigned {
113
60
            if (checkAndConsumeDirectiveWithName(Lex, "pragma", Tok,
114
60
                                                 StringRef("once")))
115
1
              return SM.getFileOffset(Tok.getLocation());
116
59
            return 0;
117
60
          }));
118
60
}
119
120
// Check if a sequence of tokens is like
121
//    "#include ("header.h" | <header.h>)".
122
// If it is, \p Tok will be the token after this directive; otherwise, it can be
123
// any token after the given \p Tok (including \p Tok).
124
150
bool checkAndConsumeInclusiveDirective(Lexer &Lex, Token &Tok) {
125
150
  auto Matched = [&]() {
126
90
    Lex.LexFromRawLexer(Tok);
127
90
    return true;
128
90
  };
129
150
  if (Tok.is(tok::hash) && 
!Lex.LexFromRawLexer(Tok)95
&&
130
150
      
Tok.is(tok::raw_identifier)95
&&
Tok.getRawIdentifier() == "include"95
) {
131
90
    if (Lex.LexFromRawLexer(Tok))
132
0
      return false;
133
90
    if (Tok.is(tok::string_literal))
134
64
      return Matched();
135
26
    if (Tok.is(tok::less)) {
136
66
      while (!Lex.LexFromRawLexer(Tok) && 
Tok.isNot(tok::greater)64
) {
137
40
      }
138
26
      if (Tok.is(tok::greater))
139
26
        return Matched();
140
26
    }
141
26
  }
142
60
  return false;
143
150
}
144
145
// Returns the offset of the last #include directive after which a new
146
// #include can be inserted. This ignores #include's after the #include block(s)
147
// in the beginning of a file to avoid inserting headers into code sections
148
// where new #include's should not be added by default.
149
// These code sections include:
150
//      - raw string literals (containing #include).
151
//      - #if blocks.
152
//      - Special #include's among declarations (e.g. functions).
153
//
154
// If no #include after which a new #include can be inserted, this returns the
155
// offset after skipping all comments from the start of the code.
156
// Inserting after an #include is not allowed if it comes after code that is not
157
// #include (e.g. pre-processing directive that is not #include, declarations).
158
unsigned getMaxHeaderInsertionOffset(StringRef FileName, StringRef Code,
159
60
                                     const IncludeStyle &Style) {
160
60
  return getOffsetAfterTokenSequence(
161
60
      FileName, Code, Style,
162
60
      [](const SourceManager &SM, Lexer &Lex, Token Tok) {
163
60
        skipComments(Lex, Tok);
164
60
        unsigned MaxOffset = SM.getFileOffset(Tok.getLocation());
165
150
        while (checkAndConsumeInclusiveDirective(Lex, Tok))
166
90
          MaxOffset = SM.getFileOffset(Tok.getLocation());
167
60
        return MaxOffset;
168
60
      });
169
60
}
170
171
178
inline StringRef trimInclude(StringRef IncludeName) {
172
178
  return IncludeName.trim("\"<>");
173
178
}
174
175
const char IncludeRegexPattern[] =
176
    R"(^[\t\ ]*#[\t\ ]*(import|include)[^"<]*(["<][^">]*[">]))";
177
178
// The filename of Path excluding extension.
179
// Used to match implementation with headers, this differs from sys::path::stem:
180
//  - in names with multiple dots (foo.cu.cc) it terminates at the *first*
181
//  - an empty stem is never returned: /foo/.bar.x => .bar
182
//  - we don't bother to handle . and .. specially
183
831
StringRef matchingStem(llvm::StringRef Path) {
184
831
  StringRef Name = llvm::sys::path::filename(Path);
185
831
  return Name.substr(0, Name.find('.', 1));
186
831
}
187
188
} // anonymous namespace
189
190
IncludeCategoryManager::IncludeCategoryManager(const IncludeStyle &Style,
191
                                               StringRef FileName)
192
1.05k
    : Style(Style), FileName(FileName) {
193
3.21k
  for (const auto &Category : Style.IncludeCategories) {
194
3.21k
    CategoryRegexs.emplace_back(Category.Regex, Category.RegexIsCaseSensitive
195
3.21k
                                                    ? 
llvm::Regex::NoFlags4
196
3.21k
                                                    : 
llvm::Regex::IgnoreCase3.20k
);
197
3.21k
  }
198
1.05k
  IsMainFile = FileName.endswith(".c") || FileName.endswith(".cc") ||
199
1.05k
               
FileName.endswith(".cpp")476
||
FileName.endswith(".c++")262
||
200
1.05k
               
FileName.endswith(".cxx")262
||
FileName.endswith(".m")262
||
201
1.05k
               
FileName.endswith(".mm")262
;
202
1.05k
  if (!Style.IncludeIsMainSourceRegex.empty()) {
203
4
    llvm::Regex MainFileRegex(Style.IncludeIsMainSourceRegex);
204
4
    IsMainFile |= MainFileRegex.match(FileName);
205
4
  }
206
1.05k
}
207
208
int IncludeCategoryManager::getIncludePriority(StringRef IncludeName,
209
813
                                               bool CheckMainHeader) const {
210
813
  int Ret = INT_MAX;
211
2.25k
  for (unsigned i = 0, e = CategoryRegexs.size(); i != e; 
++i1.43k
)
212
2.25k
    if (CategoryRegexs[i].match(IncludeName)) {
213
813
      Ret = Style.IncludeCategories[i].Priority;
214
813
      break;
215
813
    }
216
813
  if (CheckMainHeader && 
IsMainFile679
&&
Ret > 0651
&&
isMainHeader(IncludeName)649
)
217
35
    Ret = 0;
218
813
  return Ret;
219
813
}
220
221
int IncludeCategoryManager::getSortIncludePriority(StringRef IncludeName,
222
653
                                                   bool CheckMainHeader) const {
223
653
  int Ret = INT_MAX;
224
1.81k
  for (unsigned i = 0, e = CategoryRegexs.size(); i != e; 
++i1.15k
)
225
1.81k
    if (CategoryRegexs[i].match(IncludeName)) {
226
653
      Ret = Style.IncludeCategories[i].SortPriority;
227
653
      if (Ret == 0)
228
636
        Ret = Style.IncludeCategories[i].Priority;
229
653
      break;
230
653
    }
231
653
  if (CheckMainHeader && 
IsMainFile603
&&
Ret > 0575
&&
isMainHeader(IncludeName)573
)
232
23
    Ret = 0;
233
653
  return Ret;
234
653
}
235
1.22k
bool IncludeCategoryManager::isMainHeader(StringRef IncludeName) const {
236
1.22k
  if (!IncludeName.startswith("\""))
237
391
    return false;
238
239
831
  IncludeName =
240
831
      IncludeName.drop_front(1).drop_back(1); // remove the surrounding "" or <>
241
  // Not matchingStem: implementation files may have compound extensions but
242
  // headers may not.
243
831
  StringRef HeaderStem = llvm::sys::path::stem(IncludeName);
244
831
  StringRef FileStem = llvm::sys::path::stem(FileName); // foo.cu for foo.cu.cc
245
831
  StringRef MatchingFileStem = matchingStem(FileName);  // foo for foo.cu.cc
246
  // main-header examples:
247
  //  1) foo.h => foo.cc
248
  //  2) foo.h => foo.cu.cc
249
  //  3) foo.proto.h => foo.proto.cc
250
  //
251
  // non-main-header examples:
252
  //  1) foo.h => bar.cc
253
  //  2) foo.proto.h => foo.cc
254
831
  StringRef Matching;
255
831
  if (MatchingFileStem.startswith_insensitive(HeaderStem))
256
58
    Matching = MatchingFileStem; // example 1), 2)
257
773
  else if (FileStem.equals_insensitive(HeaderStem))
258
2
    Matching = FileStem; // example 3)
259
831
  if (!Matching.empty()) {
260
60
    llvm::Regex MainIncludeRegex(HeaderStem.str() + Style.IncludeIsMainRegex,
261
60
                                 llvm::Regex::IgnoreCase);
262
60
    if (MainIncludeRegex.match(Matching))
263
58
      return true;
264
60
  }
265
773
  return false;
266
831
}
267
268
HeaderIncludes::HeaderIncludes(StringRef FileName, StringRef Code,
269
                               const IncludeStyle &Style)
270
    : FileName(FileName), Code(Code), FirstIncludeOffset(-1),
271
      MinInsertOffset(
272
          getOffsetAfterHeaderGuardsAndComments(FileName, Code, Style)),
273
      MaxInsertOffset(MinInsertOffset +
274
                      getMaxHeaderInsertionOffset(
275
                          FileName, Code.drop_front(MinInsertOffset), Style)),
276
      Categories(Style, FileName),
277
60
      IncludeRegex(llvm::Regex(IncludeRegexPattern)) {
278
  // Add 0 for main header and INT_MAX for headers that are not in any
279
  // category.
280
60
  Priorities = {0, INT_MAX};
281
60
  for (const auto &Category : Style.IncludeCategories)
282
189
    Priorities.insert(Category.Priority);
283
60
  SmallVector<StringRef, 32> Lines;
284
60
  Code.drop_front(MinInsertOffset).split(Lines, "\n");
285
286
60
  unsigned Offset = MinInsertOffset;
287
60
  unsigned NextLineOffset;
288
60
  SmallVector<StringRef, 4> Matches;
289
226
  for (auto Line : Lines) {
290
226
    NextLineOffset = std::min(Code.size(), Offset + Line.size() + 1);
291
226
    if (IncludeRegex.match(Line, &Matches)) {
292
      // If this is the last line without trailing newline, we need to make
293
      // sure we don't delete across the file boundary.
294
94
      addExistingInclude(
295
94
          Include(Matches[2],
296
94
                  tooling::Range(
297
94
                      Offset, std::min(Line.size() + 1, Code.size() - Offset))),
298
94
          NextLineOffset);
299
94
    }
300
226
    Offset = NextLineOffset;
301
226
  }
302
303
  // Populate CategoryEndOfssets:
304
  // - Ensure that CategoryEndOffset[Highest] is always populated.
305
  // - If CategoryEndOffset[Priority] isn't set, use the next higher value
306
  // that is set, up to CategoryEndOffset[Highest].
307
60
  auto Highest = Priorities.begin();
308
60
  if (CategoryEndOffsets.find(*Highest) == CategoryEndOffsets.end()) {
309
50
    if (FirstIncludeOffset >= 0)
310
31
      CategoryEndOffsets[*Highest] = FirstIncludeOffset;
311
19
    else
312
19
      CategoryEndOffsets[*Highest] = MinInsertOffset;
313
50
  }
314
  // By this point, CategoryEndOffset[Highest] is always set appropriately:
315
  //  - to an appropriate location before/after existing #includes, or
316
  //  - to right after the header guard, or
317
  //  - to the beginning of the file.
318
300
  for (auto I = ++Priorities.begin(), E = Priorities.end(); I != E; 
++I240
)
319
240
    if (CategoryEndOffsets.find(*I) == CategoryEndOffsets.end())
320
185
      CategoryEndOffsets[*I] = CategoryEndOffsets[*std::prev(I)];
321
60
}
322
323
// \p Offset: the start of the line following this include directive.
324
void HeaderIncludes::addExistingInclude(Include IncludeToAdd,
325
94
                                        unsigned NextLineOffset) {
326
94
  auto Iter =
327
94
      ExistingIncludes.try_emplace(trimInclude(IncludeToAdd.Name)).first;
328
94
  Iter->second.push_back(std::move(IncludeToAdd));
329
94
  auto &CurInclude = Iter->second.back();
330
  // The header name with quotes or angle brackets.
331
  // Only record the offset of current #include if we can insert after it.
332
94
  if (CurInclude.R.getOffset() <= MaxInsertOffset) {
333
90
    int Priority = Categories.getIncludePriority(
334
90
        CurInclude.Name, /*CheckMainHeader=*/FirstIncludeOffset < 0);
335
90
    CategoryEndOffsets[Priority] = NextLineOffset;
336
90
    IncludesByPriority[Priority].push_back(&CurInclude);
337
90
    if (FirstIncludeOffset < 0)
338
41
      FirstIncludeOffset = CurInclude.R.getOffset();
339
90
  }
340
94
}
341
342
llvm::Optional<tooling::Replacement>
343
73
HeaderIncludes::insert(llvm::StringRef IncludeName, bool IsAngled) const {
344
73
  assert(IncludeName == trimInclude(IncludeName));
345
  // If a <header> ("header") already exists in code, "header" (<header>) with
346
  // different quotation will still be inserted.
347
  // FIXME: figure out if this is the best behavior.
348
0
  auto It = ExistingIncludes.find(IncludeName);
349
73
  if (It != ExistingIncludes.end())
350
4
    for (const auto &Inc : It->second)
351
4
      if ((IsAngled && 
StringRef(Inc.Name).startswith("<")2
) ||
352
4
          
(2
!IsAngled2
&&
StringRef(Inc.Name).startswith("\"")2
))
353
3
        return llvm::None;
354
70
  std::string Quoted =
355
70
      std::string(llvm::formatv(IsAngled ? 
"<{0}>"35
:
"\"{0}\""35
, IncludeName));
356
70
  StringRef QuotedName = Quoted;
357
70
  int Priority = Categories.getIncludePriority(
358
70
      QuotedName, /*CheckMainHeader=*/FirstIncludeOffset < 0);
359
70
  auto CatOffset = CategoryEndOffsets.find(Priority);
360
70
  assert(CatOffset != CategoryEndOffsets.end());
361
0
  unsigned InsertOffset = CatOffset->second; // Fall back offset
362
70
  auto Iter = IncludesByPriority.find(Priority);
363
70
  if (Iter != IncludesByPriority.end()) {
364
32
    for (const auto *Inc : Iter->second) {
365
32
      if (QuotedName < Inc->Name) {
366
11
        InsertOffset = Inc->R.getOffset();
367
11
        break;
368
11
      }
369
32
    }
370
22
  }
371
70
  assert(InsertOffset <= Code.size());
372
0
  std::string NewInclude =
373
70
      std::string(llvm::formatv("#include {0}\n", QuotedName));
374
  // When inserting headers at end of the code, also append '\n' to the code
375
  // if it does not end with '\n'.
376
  // FIXME: when inserting multiple #includes at the end of code, only one
377
  // newline should be added.
378
70
  if (InsertOffset == Code.size() && 
(10
!Code.empty()10
&&
Code.back() != '\n'9
))
379
3
    NewInclude = "\n" + NewInclude;
380
70
  return tooling::Replacement(FileName, InsertOffset, 0, NewInclude);
381
73
}
382
383
tooling::Replacements HeaderIncludes::remove(llvm::StringRef IncludeName,
384
11
                                             bool IsAngled) const {
385
11
  assert(IncludeName == trimInclude(IncludeName));
386
0
  tooling::Replacements Result;
387
11
  auto Iter = ExistingIncludes.find(IncludeName);
388
11
  if (Iter == ExistingIncludes.end())
389
0
    return Result;
390
13
  
for (const auto &Inc : Iter->second)11
{
391
13
    if ((IsAngled && 
StringRef(Inc.Name).startswith("\"")4
) ||
392
13
        
(12
!IsAngled12
&&
StringRef(Inc.Name).startswith("<")9
))
393
2
      continue;
394
11
    llvm::Error Err = Result.add(tooling::Replacement(
395
11
        FileName, Inc.R.getOffset(), Inc.R.getLength(), ""));
396
11
    if (Err) {
397
0
      auto ErrMsg = "Unexpected conflicts in #include deletions: " +
398
0
                    llvm::toString(std::move(Err));
399
0
      llvm_unreachable(ErrMsg.c_str());
400
0
    }
401
11
  }
402
11
  return Result;
403
11
}
404
405
} // namespace tooling
406
} // namespace clang