Coverage Report

Created: 2020-09-22 08:39

/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.getBuffer(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
105
                 Tok.is(tok::raw_identifier) &&
63
105
                 Tok.getRawIdentifier() == Name && 
!Lex.LexFromRawLexer(Tok)16
&&
64
16
                 Tok.is(tok::raw_identifier) &&
65
15
                 (!RawIDName || 
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
6
                  Tok.isAtStartOfLine())
106
5
                return SM.getFileOffset(Tok.getLocation());
107
55
            }
108
55
            return 0;
109
55
          }),
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
59
          }));
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
90
  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
95
      Tok.is(tok::raw_identifier) && Tok.getRawIdentifier() == "include") {
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
60
    }
141
26
  }
142
60
  return false;
143
60
}
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
1.76k
StringRef matchingStem(llvm::StringRef Path) {
184
1.76k
  StringRef Name = llvm::sys::path::filename(Path);
185
1.76k
  return Name.substr(0, Name.find('.', 1));
186
1.76k
}
187
188
} // anonymous namespace
189
190
IncludeCategoryManager::IncludeCategoryManager(const IncludeStyle &Style,
191
                                               StringRef FileName)
192
1.01k
    : Style(Style), FileName(FileName) {
193
1.01k
  FileStem = matchingStem(FileName);
194
1.01k
  for (const auto &Category : Style.IncludeCategories)
195
3.08k
    CategoryRegexs.emplace_back(Category.Regex, llvm::Regex::IgnoreCase);
196
1.01k
  IsMainFile = FileName.endswith(".c") || FileName.endswith(".cc") ||
197
442
               FileName.endswith(".cpp") || 
FileName.endswith(".c++")258
||
198
258
               FileName.endswith(".cxx") || FileName.endswith(".m") ||
199
258
               FileName.endswith(".mm");
200
1.01k
  if (!Style.IncludeIsMainSourceRegex.empty()) {
201
4
    llvm::Regex MainFileRegex(Style.IncludeIsMainSourceRegex);
202
4
    IsMainFile |= MainFileRegex.match(FileName);
203
4
  }
204
1.01k
}
205
206
int IncludeCategoryManager::getIncludePriority(StringRef IncludeName,
207
742
                                               bool CheckMainHeader) const {
208
742
  int Ret = INT_MAX;
209
2.05k
  for (unsigned i = 0, e = CategoryRegexs.size(); i != e; 
++i1.30k
)
210
2.05k
    if (CategoryRegexs[i].match(IncludeName)) {
211
742
      Ret = Style.IncludeCategories[i].Priority;
212
742
      break;
213
742
    }
214
742
  if (CheckMainHeader && 
IsMainFile611
&&
Ret > 0588
&&
isMainHeader(IncludeName)586
)
215
29
    Ret = 0;
216
742
  return Ret;
217
742
}
218
219
int IncludeCategoryManager::getSortIncludePriority(StringRef IncludeName,
220
582
                                                   bool CheckMainHeader) const {
221
582
  int Ret = INT_MAX;
222
1.61k
  for (unsigned i = 0, e = CategoryRegexs.size(); i != e; 
++i1.03k
)
223
1.61k
    if (CategoryRegexs[i].match(IncludeName)) {
224
582
      Ret = Style.IncludeCategories[i].SortPriority;
225
582
      if (Ret == 0)
226
565
        Ret = Style.IncludeCategories[i].Priority;
227
582
      break;
228
582
    }
229
582
  if (CheckMainHeader && 
IsMainFile535
&&
Ret > 0512
&&
isMainHeader(IncludeName)510
)
230
17
    Ret = 0;
231
582
  return Ret;
232
582
}
233
1.09k
bool IncludeCategoryManager::isMainHeader(StringRef IncludeName) const {
234
1.09k
  if (!IncludeName.startswith("\""))
235
349
    return false;
236
747
  StringRef HeaderStem = matchingStem(IncludeName.drop_front(1).drop_back(1));
237
747
  if (FileStem.startswith(HeaderStem) ||
238
701
      FileStem.startswith_lower(HeaderStem)) {
239
48
    llvm::Regex MainIncludeRegex(HeaderStem.str() + Style.IncludeIsMainRegex,
240
48
                                 llvm::Regex::IgnoreCase);
241
48
    if (MainIncludeRegex.match(FileStem))
242
46
      return true;
243
701
  }
244
701
  return false;
245
701
}
246
247
HeaderIncludes::HeaderIncludes(StringRef FileName, StringRef Code,
248
                               const IncludeStyle &Style)
249
    : FileName(FileName), Code(Code), FirstIncludeOffset(-1),
250
      MinInsertOffset(
251
          getOffsetAfterHeaderGuardsAndComments(FileName, Code, Style)),
252
      MaxInsertOffset(MinInsertOffset +
253
                      getMaxHeaderInsertionOffset(
254
                          FileName, Code.drop_front(MinInsertOffset), Style)),
255
      Categories(Style, FileName),
256
60
      IncludeRegex(llvm::Regex(IncludeRegexPattern)) {
257
  // Add 0 for main header and INT_MAX for headers that are not in any
258
  // category.
259
60
  Priorities = {0, INT_MAX};
260
60
  for (const auto &Category : Style.IncludeCategories)
261
189
    Priorities.insert(Category.Priority);
262
60
  SmallVector<StringRef, 32> Lines;
263
60
  Code.drop_front(MinInsertOffset).split(Lines, "\n");
264
265
60
  unsigned Offset = MinInsertOffset;
266
60
  unsigned NextLineOffset;
267
60
  SmallVector<StringRef, 4> Matches;
268
226
  for (auto Line : Lines) {
269
226
    NextLineOffset = std::min(Code.size(), Offset + Line.size() + 1);
270
226
    if (IncludeRegex.match(Line, &Matches)) {
271
      // If this is the last line without trailing newline, we need to make
272
      // sure we don't delete across the file boundary.
273
94
      addExistingInclude(
274
94
          Include(Matches[2],
275
94
                  tooling::Range(
276
94
                      Offset, std::min(Line.size() + 1, Code.size() - Offset))),
277
94
          NextLineOffset);
278
94
    }
279
226
    Offset = NextLineOffset;
280
226
  }
281
282
  // Populate CategoryEndOfssets:
283
  // - Ensure that CategoryEndOffset[Highest] is always populated.
284
  // - If CategoryEndOffset[Priority] isn't set, use the next higher value
285
  // that is set, up to CategoryEndOffset[Highest].
286
60
  auto Highest = Priorities.begin();
287
60
  if (CategoryEndOffsets.find(*Highest) == CategoryEndOffsets.end()) {
288
50
    if (FirstIncludeOffset >= 0)
289
31
      CategoryEndOffsets[*Highest] = FirstIncludeOffset;
290
19
    else
291
19
      CategoryEndOffsets[*Highest] = MinInsertOffset;
292
50
  }
293
  // By this point, CategoryEndOffset[Highest] is always set appropriately:
294
  //  - to an appropriate location before/after existing #includes, or
295
  //  - to right after the header guard, or
296
  //  - to the beginning of the file.
297
300
  for (auto I = ++Priorities.begin(), E = Priorities.end(); I != E; 
++I240
)
298
240
    if (CategoryEndOffsets.find(*I) == CategoryEndOffsets.end())
299
185
      CategoryEndOffsets[*I] = CategoryEndOffsets[*std::prev(I)];
300
60
}
301
302
// \p Offset: the start of the line following this include directive.
303
void HeaderIncludes::addExistingInclude(Include IncludeToAdd,
304
94
                                        unsigned NextLineOffset) {
305
94
  auto Iter =
306
94
      ExistingIncludes.try_emplace(trimInclude(IncludeToAdd.Name)).first;
307
94
  Iter->second.push_back(std::move(IncludeToAdd));
308
94
  auto &CurInclude = Iter->second.back();
309
  // The header name with quotes or angle brackets.
310
  // Only record the offset of current #include if we can insert after it.
311
94
  if (CurInclude.R.getOffset() <= MaxInsertOffset) {
312
90
    int Priority = Categories.getIncludePriority(
313
90
        CurInclude.Name, /*CheckMainHeader=*/FirstIncludeOffset < 0);
314
90
    CategoryEndOffsets[Priority] = NextLineOffset;
315
90
    IncludesByPriority[Priority].push_back(&CurInclude);
316
90
    if (FirstIncludeOffset < 0)
317
41
      FirstIncludeOffset = CurInclude.R.getOffset();
318
90
  }
319
94
}
320
321
llvm::Optional<tooling::Replacement>
322
73
HeaderIncludes::insert(llvm::StringRef IncludeName, bool IsAngled) const {
323
73
  assert(IncludeName == trimInclude(IncludeName));
324
  // If a <header> ("header") already exists in code, "header" (<header>) with
325
  // different quotation will still be inserted.
326
  // FIXME: figure out if this is the best behavior.
327
73
  auto It = ExistingIncludes.find(IncludeName);
328
73
  if (It != ExistingIncludes.end())
329
4
    for (const auto &Inc : It->second)
330
4
      if ((IsAngled && 
StringRef(Inc.Name).startswith("<")2
) ||
331
2
          (!IsAngled && StringRef(Inc.Name).startswith("\"")))
332
3
        return llvm::None;
333
70
  std::string Quoted =
334
35
      std::string(llvm::formatv(IsAngled ? "<{0}>" : "\"{0}\"", IncludeName));
335
70
  StringRef QuotedName = Quoted;
336
70
  int Priority = Categories.getIncludePriority(
337
70
      QuotedName, /*CheckMainHeader=*/FirstIncludeOffset < 0);
338
70
  auto CatOffset = CategoryEndOffsets.find(Priority);
339
70
  assert(CatOffset != CategoryEndOffsets.end());
340
70
  unsigned InsertOffset = CatOffset->second; // Fall back offset
341
70
  auto Iter = IncludesByPriority.find(Priority);
342
70
  if (Iter != IncludesByPriority.end()) {
343
32
    for (const auto *Inc : Iter->second) {
344
32
      if (QuotedName < Inc->Name) {
345
11
        InsertOffset = Inc->R.getOffset();
346
11
        break;
347
11
      }
348
32
    }
349
22
  }
350
70
  assert(InsertOffset <= Code.size());
351
70
  std::string NewInclude =
352
70
      std::string(llvm::formatv("#include {0}\n", QuotedName));
353
  // When inserting headers at end of the code, also append '\n' to the code
354
  // if it does not end with '\n'.
355
  // FIXME: when inserting multiple #includes at the end of code, only one
356
  // newline should be added.
357
70
  if (InsertOffset == Code.size() && 
(10
!Code.empty()10
&&
Code.back() != '\n'9
))
358
3
    NewInclude = "\n" + NewInclude;
359
70
  return tooling::Replacement(FileName, InsertOffset, 0, NewInclude);
360
73
}
361
362
tooling::Replacements HeaderIncludes::remove(llvm::StringRef IncludeName,
363
11
                                             bool IsAngled) const {
364
11
  assert(IncludeName == trimInclude(IncludeName));
365
11
  tooling::Replacements Result;
366
11
  auto Iter = ExistingIncludes.find(IncludeName);
367
11
  if (Iter == ExistingIncludes.end())
368
0
    return Result;
369
13
  
for (const auto &Inc : Iter->second)11
{
370
13
    if ((IsAngled && 
StringRef(Inc.Name).startswith("\"")4
) ||
371
12
        (!IsAngled && 
StringRef(Inc.Name).startswith("<")9
))
372
2
      continue;
373
11
    llvm::Error Err = Result.add(tooling::Replacement(
374
11
        FileName, Inc.R.getOffset(), Inc.R.getLength(), ""));
375
11
    if (Err) {
376
0
      auto ErrMsg = "Unexpected conflicts in #include deletions: " +
377
0
                    llvm::toString(std::move(Err));
378
0
      llvm_unreachable(ErrMsg.c_str());
379
0
    }
380
11
  }
381
11
  return Result;
382
11
}
383
384
} // namespace tooling
385
} // namespace clang