1 //===- VerifyDiagnosticConsumer.cpp - Verifying Diagnostic Client ---------===//
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 is a concrete diagnostic client, which buffers the diagnostic messages.
11 //
12 //===----------------------------------------------------------------------===//
13
14 #include "clang/Frontend/VerifyDiagnosticConsumer.h"
15 #include "clang/Basic/CharInfo.h"
16 #include "clang/Basic/Diagnostic.h"
17 #include "clang/Basic/DiagnosticOptions.h"
18 #include "clang/Basic/FileManager.h"
19 #include "clang/Basic/LLVM.h"
20 #include "clang/Basic/SourceLocation.h"
21 #include "clang/Basic/SourceManager.h"
22 #include "clang/Basic/TokenKinds.h"
23 #include "clang/Frontend/FrontendDiagnostic.h"
24 #include "clang/Frontend/TextDiagnosticBuffer.h"
25 #include "clang/Lex/HeaderSearch.h"
26 #include "clang/Lex/Lexer.h"
27 #include "clang/Lex/PPCallbacks.h"
28 #include "clang/Lex/Preprocessor.h"
29 #include "clang/Lex/Token.h"
30 #include "llvm/ADT/STLExtras.h"
31 #include "llvm/ADT/SmallPtrSet.h"
32 #include "llvm/ADT/SmallString.h"
33 #include "llvm/ADT/StringRef.h"
34 #include "llvm/ADT/Twine.h"
35 #include "llvm/Support/ErrorHandling.h"
36 #include "llvm/Support/Regex.h"
37 #include "llvm/Support/raw_ostream.h"
38 #include <algorithm>
39 #include <cassert>
40 #include <cstddef>
41 #include <cstring>
42 #include <iterator>
43 #include <memory>
44 #include <string>
45 #include <utility>
46 #include <vector>
47
48 using namespace clang;
49
50 using Directive = VerifyDiagnosticConsumer::Directive;
51 using DirectiveList = VerifyDiagnosticConsumer::DirectiveList;
52 using ExpectedData = VerifyDiagnosticConsumer::ExpectedData;
53
VerifyDiagnosticConsumer(DiagnosticsEngine & Diags_)54 VerifyDiagnosticConsumer::VerifyDiagnosticConsumer(DiagnosticsEngine &Diags_)
55 : Diags(Diags_), PrimaryClient(Diags.getClient()),
56 PrimaryClientOwner(Diags.takeClient()),
57 Buffer(new TextDiagnosticBuffer()), Status(HasNoDirectives) {
58 if (Diags.hasSourceManager())
59 setSourceManager(Diags.getSourceManager());
60 }
61
~VerifyDiagnosticConsumer()62 VerifyDiagnosticConsumer::~VerifyDiagnosticConsumer() {
63 assert(!ActiveSourceFiles && "Incomplete parsing of source files!");
64 assert(!CurrentPreprocessor && "CurrentPreprocessor should be invalid!");
65 SrcManager = nullptr;
66 CheckDiagnostics();
67 assert(!Diags.ownsClient() &&
68 "The VerifyDiagnosticConsumer takes over ownership of the client!");
69 }
70
71 #ifndef NDEBUG
72
73 namespace {
74
75 class VerifyFileTracker : public PPCallbacks {
76 VerifyDiagnosticConsumer &Verify;
77 SourceManager &SM;
78
79 public:
VerifyFileTracker(VerifyDiagnosticConsumer & Verify,SourceManager & SM)80 VerifyFileTracker(VerifyDiagnosticConsumer &Verify, SourceManager &SM)
81 : Verify(Verify), SM(SM) {}
82
83 /// Hook into the preprocessor and update the list of parsed
84 /// files when the preprocessor indicates a new file is entered.
FileChanged(SourceLocation Loc,FileChangeReason Reason,SrcMgr::CharacteristicKind FileType,FileID PrevFID)85 void FileChanged(SourceLocation Loc, FileChangeReason Reason,
86 SrcMgr::CharacteristicKind FileType,
87 FileID PrevFID) override {
88 Verify.UpdateParsedFileStatus(SM, SM.getFileID(Loc),
89 VerifyDiagnosticConsumer::IsParsed);
90 }
91 };
92
93 } // namespace
94
95 #endif
96
97 // DiagnosticConsumer interface.
98
BeginSourceFile(const LangOptions & LangOpts,const Preprocessor * PP)99 void VerifyDiagnosticConsumer::BeginSourceFile(const LangOptions &LangOpts,
100 const Preprocessor *PP) {
101 // Attach comment handler on first invocation.
102 if (++ActiveSourceFiles == 1) {
103 if (PP) {
104 CurrentPreprocessor = PP;
105 this->LangOpts = &LangOpts;
106 setSourceManager(PP->getSourceManager());
107 const_cast<Preprocessor *>(PP)->addCommentHandler(this);
108 #ifndef NDEBUG
109 // Debug build tracks parsed files.
110 const_cast<Preprocessor *>(PP)->addPPCallbacks(
111 llvm::make_unique<VerifyFileTracker>(*this, *SrcManager));
112 #endif
113 }
114 }
115
116 assert((!PP || CurrentPreprocessor == PP) && "Preprocessor changed!");
117 PrimaryClient->BeginSourceFile(LangOpts, PP);
118 }
119
EndSourceFile()120 void VerifyDiagnosticConsumer::EndSourceFile() {
121 assert(ActiveSourceFiles && "No active source files!");
122 PrimaryClient->EndSourceFile();
123
124 // Detach comment handler once last active source file completed.
125 if (--ActiveSourceFiles == 0) {
126 if (CurrentPreprocessor)
127 const_cast<Preprocessor *>(CurrentPreprocessor)->
128 removeCommentHandler(this);
129
130 // Check diagnostics once last file completed.
131 CheckDiagnostics();
132 CurrentPreprocessor = nullptr;
133 LangOpts = nullptr;
134 }
135 }
136
HandleDiagnostic(DiagnosticsEngine::Level DiagLevel,const Diagnostic & Info)137 void VerifyDiagnosticConsumer::HandleDiagnostic(
138 DiagnosticsEngine::Level DiagLevel, const Diagnostic &Info) {
139 if (Info.hasSourceManager()) {
140 // If this diagnostic is for a different source manager, ignore it.
141 if (SrcManager && &Info.getSourceManager() != SrcManager)
142 return;
143
144 setSourceManager(Info.getSourceManager());
145 }
146
147 #ifndef NDEBUG
148 // Debug build tracks unparsed files for possible
149 // unparsed expected-* directives.
150 if (SrcManager) {
151 SourceLocation Loc = Info.getLocation();
152 if (Loc.isValid()) {
153 ParsedStatus PS = IsUnparsed;
154
155 Loc = SrcManager->getExpansionLoc(Loc);
156 FileID FID = SrcManager->getFileID(Loc);
157
158 const FileEntry *FE = SrcManager->getFileEntryForID(FID);
159 if (FE && CurrentPreprocessor && SrcManager->isLoadedFileID(FID)) {
160 // If the file is a modules header file it shall not be parsed
161 // for expected-* directives.
162 HeaderSearch &HS = CurrentPreprocessor->getHeaderSearchInfo();
163 if (HS.findModuleForHeader(FE))
164 PS = IsUnparsedNoDirectives;
165 }
166
167 UpdateParsedFileStatus(*SrcManager, FID, PS);
168 }
169 }
170 #endif
171
172 // Send the diagnostic to the buffer, we will check it once we reach the end
173 // of the source file (or are destructed).
174 Buffer->HandleDiagnostic(DiagLevel, Info);
175 }
176
177 //===----------------------------------------------------------------------===//
178 // Checking diagnostics implementation.
179 //===----------------------------------------------------------------------===//
180
181 using DiagList = TextDiagnosticBuffer::DiagList;
182 using const_diag_iterator = TextDiagnosticBuffer::const_iterator;
183
184 namespace {
185
186 /// StandardDirective - Directive with string matching.
187 class StandardDirective : public Directive {
188 public:
StandardDirective(SourceLocation DirectiveLoc,SourceLocation DiagnosticLoc,bool MatchAnyLine,StringRef Text,unsigned Min,unsigned Max)189 StandardDirective(SourceLocation DirectiveLoc, SourceLocation DiagnosticLoc,
190 bool MatchAnyLine, StringRef Text, unsigned Min,
191 unsigned Max)
192 : Directive(DirectiveLoc, DiagnosticLoc, MatchAnyLine, Text, Min, Max) {}
193
isValid(std::string & Error)194 bool isValid(std::string &Error) override {
195 // all strings are considered valid; even empty ones
196 return true;
197 }
198
match(StringRef S)199 bool match(StringRef S) override {
200 return S.find(Text) != StringRef::npos;
201 }
202 };
203
204 /// RegexDirective - Directive with regular-expression matching.
205 class RegexDirective : public Directive {
206 public:
RegexDirective(SourceLocation DirectiveLoc,SourceLocation DiagnosticLoc,bool MatchAnyLine,StringRef Text,unsigned Min,unsigned Max,StringRef RegexStr)207 RegexDirective(SourceLocation DirectiveLoc, SourceLocation DiagnosticLoc,
208 bool MatchAnyLine, StringRef Text, unsigned Min, unsigned Max,
209 StringRef RegexStr)
210 : Directive(DirectiveLoc, DiagnosticLoc, MatchAnyLine, Text, Min, Max),
211 Regex(RegexStr) {}
212
isValid(std::string & Error)213 bool isValid(std::string &Error) override {
214 return Regex.isValid(Error);
215 }
216
match(StringRef S)217 bool match(StringRef S) override {
218 return Regex.match(S);
219 }
220
221 private:
222 llvm::Regex Regex;
223 };
224
225 class ParseHelper
226 {
227 public:
ParseHelper(StringRef S)228 ParseHelper(StringRef S)
229 : Begin(S.begin()), End(S.end()), C(Begin), P(Begin) {}
230
231 // Return true if string literal is next.
Next(StringRef S)232 bool Next(StringRef S) {
233 P = C;
234 PEnd = C + S.size();
235 if (PEnd > End)
236 return false;
237 return memcmp(P, S.data(), S.size()) == 0;
238 }
239
240 // Return true if number is next.
241 // Output N only if number is next.
Next(unsigned & N)242 bool Next(unsigned &N) {
243 unsigned TMP = 0;
244 P = C;
245 for (; P < End && P[0] >= '0' && P[0] <= '9'; ++P) {
246 TMP *= 10;
247 TMP += P[0] - '0';
248 }
249 if (P == C)
250 return false;
251 PEnd = P;
252 N = TMP;
253 return true;
254 }
255
256 // Return true if string literal S is matched in content.
257 // When true, P marks begin-position of the match, and calling Advance sets C
258 // to end-position of the match.
259 // If S is the empty string, then search for any letter instead (makes sense
260 // with FinishDirectiveToken=true).
261 // If EnsureStartOfWord, then skip matches that don't start a new word.
262 // If FinishDirectiveToken, then assume the match is the start of a comment
263 // directive for -verify, and extend the match to include the entire first
264 // token of that directive.
Search(StringRef S,bool EnsureStartOfWord=false,bool FinishDirectiveToken=false)265 bool Search(StringRef S, bool EnsureStartOfWord = false,
266 bool FinishDirectiveToken = false) {
267 do {
268 if (!S.empty()) {
269 P = std::search(C, End, S.begin(), S.end());
270 PEnd = P + S.size();
271 }
272 else {
273 P = C;
274 while (P != End && !isLetter(*P))
275 ++P;
276 PEnd = P + 1;
277 }
278 if (P == End)
279 break;
280 // If not start of word but required, skip and search again.
281 if (EnsureStartOfWord
282 // Check if string literal starts a new word.
283 && !(P == Begin || isWhitespace(P[-1])
284 // Or it could be preceded by the start of a comment.
285 || (P > (Begin + 1) && (P[-1] == '/' || P[-1] == '*')
286 && P[-2] == '/')))
287 continue;
288 if (FinishDirectiveToken) {
289 while (PEnd != End && (isAlphanumeric(*PEnd)
290 || *PEnd == '-' || *PEnd == '_'))
291 ++PEnd;
292 // Put back trailing digits and hyphens to be parsed later as a count
293 // or count range. Because -verify prefixes must start with letters,
294 // we know the actual directive we found starts with a letter, so
295 // we won't put back the entire directive word and thus record an empty
296 // string.
297 assert(isLetter(*P) && "-verify prefix must start with a letter");
298 while (isDigit(PEnd[-1]) || PEnd[-1] == '-')
299 --PEnd;
300 }
301 return true;
302 } while (Advance());
303 return false;
304 }
305
306 // Return true if a CloseBrace that closes the OpenBrace at the current nest
307 // level is found. When true, P marks begin-position of CloseBrace.
SearchClosingBrace(StringRef OpenBrace,StringRef CloseBrace)308 bool SearchClosingBrace(StringRef OpenBrace, StringRef CloseBrace) {
309 unsigned Depth = 1;
310 P = C;
311 while (P < End) {
312 StringRef S(P, End - P);
313 if (S.startswith(OpenBrace)) {
314 ++Depth;
315 P += OpenBrace.size();
316 } else if (S.startswith(CloseBrace)) {
317 --Depth;
318 if (Depth == 0) {
319 PEnd = P + CloseBrace.size();
320 return true;
321 }
322 P += CloseBrace.size();
323 } else {
324 ++P;
325 }
326 }
327 return false;
328 }
329
330 // Advance 1-past previous next/search.
331 // Behavior is undefined if previous next/search failed.
Advance()332 bool Advance() {
333 C = PEnd;
334 return C < End;
335 }
336
337 // Skip zero or more whitespace.
SkipWhitespace()338 void SkipWhitespace() {
339 for (; C < End && isWhitespace(*C); ++C)
340 ;
341 }
342
343 // Return true if EOF reached.
Done()344 bool Done() {
345 return !(C < End);
346 }
347
348 // Beginning of expected content.
349 const char * const Begin;
350
351 // End of expected content (1-past).
352 const char * const End;
353
354 // Position of next char in content.
355 const char *C;
356
357 const char *P;
358
359 private:
360 // Previous next/search subject end (1-past).
361 const char *PEnd = nullptr;
362 };
363
364 } // anonymous
365
366 /// ParseDirective - Go through the comment and see if it indicates expected
367 /// diagnostics. If so, then put them in the appropriate directive list.
368 ///
369 /// Returns true if any valid directives were found.
ParseDirective(StringRef S,ExpectedData * ED,SourceManager & SM,Preprocessor * PP,SourceLocation Pos,VerifyDiagnosticConsumer::DirectiveStatus & Status)370 static bool ParseDirective(StringRef S, ExpectedData *ED, SourceManager &SM,
371 Preprocessor *PP, SourceLocation Pos,
372 VerifyDiagnosticConsumer::DirectiveStatus &Status) {
373 DiagnosticsEngine &Diags = PP ? PP->getDiagnostics() : SM.getDiagnostics();
374
375 // A single comment may contain multiple directives.
376 bool FoundDirective = false;
377 for (ParseHelper PH(S); !PH.Done();) {
378 // Search for the initial directive token.
379 // If one prefix, save time by searching only for its directives.
380 // Otherwise, search for any potential directive token and check it later.
381 const auto &Prefixes = Diags.getDiagnosticOptions().VerifyPrefixes;
382 if (!(Prefixes.size() == 1 ? PH.Search(*Prefixes.begin(), true, true)
383 : PH.Search("", true, true)))
384 break;
385 PH.Advance();
386
387 // Default directive kind.
388 bool RegexKind = false;
389 const char* KindStr = "string";
390
391 // Parse the initial directive token in reverse so we can easily determine
392 // its exact actual prefix. If we were to parse it from the front instead,
393 // it would be harder to determine where the prefix ends because there
394 // might be multiple matching -verify prefixes because some might prefix
395 // others.
396 StringRef DToken(PH.P, PH.C - PH.P);
397
398 // Regex in initial directive token: -re
399 if (DToken.endswith("-re")) {
400 RegexKind = true;
401 KindStr = "regex";
402 DToken = DToken.substr(0, DToken.size()-3);
403 }
404
405 // Type in initial directive token: -{error|warning|note|no-diagnostics}
406 DirectiveList *DL = nullptr;
407 bool NoDiag = false;
408 StringRef DType;
409 if (DToken.endswith(DType="-error"))
410 DL = ED ? &ED->Errors : nullptr;
411 else if (DToken.endswith(DType="-warning"))
412 DL = ED ? &ED->Warnings : nullptr;
413 else if (DToken.endswith(DType="-remark"))
414 DL = ED ? &ED->Remarks : nullptr;
415 else if (DToken.endswith(DType="-note"))
416 DL = ED ? &ED->Notes : nullptr;
417 else if (DToken.endswith(DType="-no-diagnostics")) {
418 NoDiag = true;
419 if (RegexKind)
420 continue;
421 }
422 else
423 continue;
424 DToken = DToken.substr(0, DToken.size()-DType.size());
425
426 // What's left in DToken is the actual prefix. That might not be a -verify
427 // prefix even if there is only one -verify prefix (for example, the full
428 // DToken is foo-bar-warning, but foo is the only -verify prefix).
429 if (!std::binary_search(Prefixes.begin(), Prefixes.end(), DToken))
430 continue;
431
432 if (NoDiag) {
433 if (Status == VerifyDiagnosticConsumer::HasOtherExpectedDirectives)
434 Diags.Report(Pos, diag::err_verify_invalid_no_diags)
435 << /*IsExpectedNoDiagnostics=*/true;
436 else
437 Status = VerifyDiagnosticConsumer::HasExpectedNoDiagnostics;
438 continue;
439 }
440 if (Status == VerifyDiagnosticConsumer::HasExpectedNoDiagnostics) {
441 Diags.Report(Pos, diag::err_verify_invalid_no_diags)
442 << /*IsExpectedNoDiagnostics=*/false;
443 continue;
444 }
445 Status = VerifyDiagnosticConsumer::HasOtherExpectedDirectives;
446
447 // If a directive has been found but we're not interested
448 // in storing the directive information, return now.
449 if (!DL)
450 return true;
451
452 // Next optional token: @
453 SourceLocation ExpectedLoc;
454 bool MatchAnyLine = false;
455 if (!PH.Next("@")) {
456 ExpectedLoc = Pos;
457 } else {
458 PH.Advance();
459 unsigned Line = 0;
460 bool FoundPlus = PH.Next("+");
461 if (FoundPlus || PH.Next("-")) {
462 // Relative to current line.
463 PH.Advance();
464 bool Invalid = false;
465 unsigned ExpectedLine = SM.getSpellingLineNumber(Pos, &Invalid);
466 if (!Invalid && PH.Next(Line) && (FoundPlus || Line < ExpectedLine)) {
467 if (FoundPlus) ExpectedLine += Line;
468 else ExpectedLine -= Line;
469 ExpectedLoc = SM.translateLineCol(SM.getFileID(Pos), ExpectedLine, 1);
470 }
471 } else if (PH.Next(Line)) {
472 // Absolute line number.
473 if (Line > 0)
474 ExpectedLoc = SM.translateLineCol(SM.getFileID(Pos), Line, 1);
475 } else if (PP && PH.Search(":")) {
476 // Specific source file.
477 StringRef Filename(PH.C, PH.P-PH.C);
478 PH.Advance();
479
480 // Lookup file via Preprocessor, like a #include.
481 const DirectoryLookup *CurDir;
482 const FileEntry *FE =
483 PP->LookupFile(Pos, Filename, false, nullptr, nullptr, CurDir,
484 nullptr, nullptr, nullptr, nullptr);
485 if (!FE) {
486 Diags.Report(Pos.getLocWithOffset(PH.C-PH.Begin),
487 diag::err_verify_missing_file) << Filename << KindStr;
488 continue;
489 }
490
491 if (SM.translateFile(FE).isInvalid())
492 SM.createFileID(FE, Pos, SrcMgr::C_User);
493
494 if (PH.Next(Line) && Line > 0)
495 ExpectedLoc = SM.translateFileLineCol(FE, Line, 1);
496 else if (PH.Next("*")) {
497 MatchAnyLine = true;
498 ExpectedLoc = SM.translateFileLineCol(FE, 1, 1);
499 }
500 } else if (PH.Next("*")) {
501 MatchAnyLine = true;
502 ExpectedLoc = SourceLocation();
503 }
504
505 if (ExpectedLoc.isInvalid() && !MatchAnyLine) {
506 Diags.Report(Pos.getLocWithOffset(PH.C-PH.Begin),
507 diag::err_verify_missing_line) << KindStr;
508 continue;
509 }
510 PH.Advance();
511 }
512
513 // Skip optional whitespace.
514 PH.SkipWhitespace();
515
516 // Next optional token: positive integer or a '+'.
517 unsigned Min = 1;
518 unsigned Max = 1;
519 if (PH.Next(Min)) {
520 PH.Advance();
521 // A positive integer can be followed by a '+' meaning min
522 // or more, or by a '-' meaning a range from min to max.
523 if (PH.Next("+")) {
524 Max = Directive::MaxCount;
525 PH.Advance();
526 } else if (PH.Next("-")) {
527 PH.Advance();
528 if (!PH.Next(Max) || Max < Min) {
529 Diags.Report(Pos.getLocWithOffset(PH.C-PH.Begin),
530 diag::err_verify_invalid_range) << KindStr;
531 continue;
532 }
533 PH.Advance();
534 } else {
535 Max = Min;
536 }
537 } else if (PH.Next("+")) {
538 // '+' on its own means "1 or more".
539 Max = Directive::MaxCount;
540 PH.Advance();
541 }
542
543 // Skip optional whitespace.
544 PH.SkipWhitespace();
545
546 // Next token: {{
547 if (!PH.Next("{{")) {
548 Diags.Report(Pos.getLocWithOffset(PH.C-PH.Begin),
549 diag::err_verify_missing_start) << KindStr;
550 continue;
551 }
552 PH.Advance();
553 const char* const ContentBegin = PH.C; // mark content begin
554
555 // Search for token: }}
556 if (!PH.SearchClosingBrace("{{", "}}")) {
557 Diags.Report(Pos.getLocWithOffset(PH.C-PH.Begin),
558 diag::err_verify_missing_end) << KindStr;
559 continue;
560 }
561 const char* const ContentEnd = PH.P; // mark content end
562 PH.Advance();
563
564 // Build directive text; convert \n to newlines.
565 std::string Text;
566 StringRef NewlineStr = "\\n";
567 StringRef Content(ContentBegin, ContentEnd-ContentBegin);
568 size_t CPos = 0;
569 size_t FPos;
570 while ((FPos = Content.find(NewlineStr, CPos)) != StringRef::npos) {
571 Text += Content.substr(CPos, FPos-CPos);
572 Text += '\n';
573 CPos = FPos + NewlineStr.size();
574 }
575 if (Text.empty())
576 Text.assign(ContentBegin, ContentEnd);
577
578 // Check that regex directives contain at least one regex.
579 if (RegexKind && Text.find("{{") == StringRef::npos) {
580 Diags.Report(Pos.getLocWithOffset(ContentBegin-PH.Begin),
581 diag::err_verify_missing_regex) << Text;
582 return false;
583 }
584
585 // Construct new directive.
586 std::unique_ptr<Directive> D = Directive::create(
587 RegexKind, Pos, ExpectedLoc, MatchAnyLine, Text, Min, Max);
588
589 std::string Error;
590 if (D->isValid(Error)) {
591 DL->push_back(std::move(D));
592 FoundDirective = true;
593 } else {
594 Diags.Report(Pos.getLocWithOffset(ContentBegin-PH.Begin),
595 diag::err_verify_invalid_content)
596 << KindStr << Error;
597 }
598 }
599
600 return FoundDirective;
601 }
602
603 /// HandleComment - Hook into the preprocessor and extract comments containing
604 /// expected errors and warnings.
HandleComment(Preprocessor & PP,SourceRange Comment)605 bool VerifyDiagnosticConsumer::HandleComment(Preprocessor &PP,
606 SourceRange Comment) {
607 SourceManager &SM = PP.getSourceManager();
608
609 // If this comment is for a different source manager, ignore it.
610 if (SrcManager && &SM != SrcManager)
611 return false;
612
613 SourceLocation CommentBegin = Comment.getBegin();
614
615 const char *CommentRaw = SM.getCharacterData(CommentBegin);
616 StringRef C(CommentRaw, SM.getCharacterData(Comment.getEnd()) - CommentRaw);
617
618 if (C.empty())
619 return false;
620
621 // Fold any "\<EOL>" sequences
622 size_t loc = C.find('\\');
623 if (loc == StringRef::npos) {
624 ParseDirective(C, &ED, SM, &PP, CommentBegin, Status);
625 return false;
626 }
627
628 std::string C2;
629 C2.reserve(C.size());
630
631 for (size_t last = 0;; loc = C.find('\\', last)) {
632 if (loc == StringRef::npos || loc == C.size()) {
633 C2 += C.substr(last);
634 break;
635 }
636 C2 += C.substr(last, loc-last);
637 last = loc + 1;
638
639 if (C[last] == '\n' || C[last] == '\r') {
640 ++last;
641
642 // Escape \r\n or \n\r, but not \n\n.
643 if (last < C.size())
644 if (C[last] == '\n' || C[last] == '\r')
645 if (C[last] != C[last-1])
646 ++last;
647 } else {
648 // This was just a normal backslash.
649 C2 += '\\';
650 }
651 }
652
653 if (!C2.empty())
654 ParseDirective(C2, &ED, SM, &PP, CommentBegin, Status);
655 return false;
656 }
657
658 #ifndef NDEBUG
659 /// Lex the specified source file to determine whether it contains
660 /// any expected-* directives. As a Lexer is used rather than a full-blown
661 /// Preprocessor, directives inside skipped #if blocks will still be found.
662 ///
663 /// \return true if any directives were found.
findDirectives(SourceManager & SM,FileID FID,const LangOptions & LangOpts)664 static bool findDirectives(SourceManager &SM, FileID FID,
665 const LangOptions &LangOpts) {
666 // Create a raw lexer to pull all the comments out of FID.
667 if (FID.isInvalid())
668 return false;
669
670 // Create a lexer to lex all the tokens of the main file in raw mode.
671 const llvm::MemoryBuffer *FromFile = SM.getBuffer(FID);
672 Lexer RawLex(FID, FromFile, SM, LangOpts);
673
674 // Return comments as tokens, this is how we find expected diagnostics.
675 RawLex.SetCommentRetentionState(true);
676
677 Token Tok;
678 Tok.setKind(tok::comment);
679 VerifyDiagnosticConsumer::DirectiveStatus Status =
680 VerifyDiagnosticConsumer::HasNoDirectives;
681 while (Tok.isNot(tok::eof)) {
682 RawLex.LexFromRawLexer(Tok);
683 if (!Tok.is(tok::comment)) continue;
684
685 std::string Comment = RawLex.getSpelling(Tok, SM, LangOpts);
686 if (Comment.empty()) continue;
687
688 // Find first directive.
689 if (ParseDirective(Comment, nullptr, SM, nullptr, Tok.getLocation(),
690 Status))
691 return true;
692 }
693 return false;
694 }
695 #endif // !NDEBUG
696
697 /// Takes a list of diagnostics that have been generated but not matched
698 /// by an expected-* directive and produces a diagnostic to the user from this.
PrintUnexpected(DiagnosticsEngine & Diags,SourceManager * SourceMgr,const_diag_iterator diag_begin,const_diag_iterator diag_end,const char * Kind)699 static unsigned PrintUnexpected(DiagnosticsEngine &Diags, SourceManager *SourceMgr,
700 const_diag_iterator diag_begin,
701 const_diag_iterator diag_end,
702 const char *Kind) {
703 if (diag_begin == diag_end) return 0;
704
705 SmallString<256> Fmt;
706 llvm::raw_svector_ostream OS(Fmt);
707 for (const_diag_iterator I = diag_begin, E = diag_end; I != E; ++I) {
708 if (I->first.isInvalid() || !SourceMgr)
709 OS << "\n (frontend)";
710 else {
711 OS << "\n ";
712 if (const FileEntry *File = SourceMgr->getFileEntryForID(
713 SourceMgr->getFileID(I->first)))
714 OS << " File " << File->getName();
715 OS << " Line " << SourceMgr->getPresumedLineNumber(I->first);
716 }
717 OS << ": " << I->second;
718 }
719
720 Diags.Report(diag::err_verify_inconsistent_diags).setForceEmit()
721 << Kind << /*Unexpected=*/true << OS.str();
722 return std::distance(diag_begin, diag_end);
723 }
724
725 /// Takes a list of diagnostics that were expected to have been generated
726 /// but were not and produces a diagnostic to the user from this.
PrintExpected(DiagnosticsEngine & Diags,SourceManager & SourceMgr,std::vector<Directive * > & DL,const char * Kind)727 static unsigned PrintExpected(DiagnosticsEngine &Diags,
728 SourceManager &SourceMgr,
729 std::vector<Directive *> &DL, const char *Kind) {
730 if (DL.empty())
731 return 0;
732
733 SmallString<256> Fmt;
734 llvm::raw_svector_ostream OS(Fmt);
735 for (const auto *D : DL) {
736 if (D->DiagnosticLoc.isInvalid())
737 OS << "\n File *";
738 else
739 OS << "\n File " << SourceMgr.getFilename(D->DiagnosticLoc);
740 if (D->MatchAnyLine)
741 OS << " Line *";
742 else
743 OS << " Line " << SourceMgr.getPresumedLineNumber(D->DiagnosticLoc);
744 if (D->DirectiveLoc != D->DiagnosticLoc)
745 OS << " (directive at "
746 << SourceMgr.getFilename(D->DirectiveLoc) << ':'
747 << SourceMgr.getPresumedLineNumber(D->DirectiveLoc) << ')';
748 OS << ": " << D->Text;
749 }
750
751 Diags.Report(diag::err_verify_inconsistent_diags).setForceEmit()
752 << Kind << /*Unexpected=*/false << OS.str();
753 return DL.size();
754 }
755
756 /// Determine whether two source locations come from the same file.
IsFromSameFile(SourceManager & SM,SourceLocation DirectiveLoc,SourceLocation DiagnosticLoc)757 static bool IsFromSameFile(SourceManager &SM, SourceLocation DirectiveLoc,
758 SourceLocation DiagnosticLoc) {
759 while (DiagnosticLoc.isMacroID())
760 DiagnosticLoc = SM.getImmediateMacroCallerLoc(DiagnosticLoc);
761
762 if (SM.isWrittenInSameFile(DirectiveLoc, DiagnosticLoc))
763 return true;
764
765 const FileEntry *DiagFile = SM.getFileEntryForID(SM.getFileID(DiagnosticLoc));
766 if (!DiagFile && SM.isWrittenInMainFile(DirectiveLoc))
767 return true;
768
769 return (DiagFile == SM.getFileEntryForID(SM.getFileID(DirectiveLoc)));
770 }
771
772 /// CheckLists - Compare expected to seen diagnostic lists and return the
773 /// the difference between them.
CheckLists(DiagnosticsEngine & Diags,SourceManager & SourceMgr,const char * Label,DirectiveList & Left,const_diag_iterator d2_begin,const_diag_iterator d2_end,bool IgnoreUnexpected)774 static unsigned CheckLists(DiagnosticsEngine &Diags, SourceManager &SourceMgr,
775 const char *Label,
776 DirectiveList &Left,
777 const_diag_iterator d2_begin,
778 const_diag_iterator d2_end,
779 bool IgnoreUnexpected) {
780 std::vector<Directive *> LeftOnly;
781 DiagList Right(d2_begin, d2_end);
782
783 for (auto &Owner : Left) {
784 Directive &D = *Owner;
785 unsigned LineNo1 = SourceMgr.getPresumedLineNumber(D.DiagnosticLoc);
786
787 for (unsigned i = 0; i < D.Max; ++i) {
788 DiagList::iterator II, IE;
789 for (II = Right.begin(), IE = Right.end(); II != IE; ++II) {
790 if (!D.MatchAnyLine) {
791 unsigned LineNo2 = SourceMgr.getPresumedLineNumber(II->first);
792 if (LineNo1 != LineNo2)
793 continue;
794 }
795
796 if (!D.DiagnosticLoc.isInvalid() &&
797 !IsFromSameFile(SourceMgr, D.DiagnosticLoc, II->first))
798 continue;
799
800 const std::string &RightText = II->second;
801 if (D.match(RightText))
802 break;
803 }
804 if (II == IE) {
805 // Not found.
806 if (i >= D.Min) break;
807 LeftOnly.push_back(&D);
808 } else {
809 // Found. The same cannot be found twice.
810 Right.erase(II);
811 }
812 }
813 }
814 // Now all that's left in Right are those that were not matched.
815 unsigned num = PrintExpected(Diags, SourceMgr, LeftOnly, Label);
816 if (!IgnoreUnexpected)
817 num += PrintUnexpected(Diags, &SourceMgr, Right.begin(), Right.end(), Label);
818 return num;
819 }
820
821 /// CheckResults - This compares the expected results to those that
822 /// were actually reported. It emits any discrepencies. Return "true" if there
823 /// were problems. Return "false" otherwise.
CheckResults(DiagnosticsEngine & Diags,SourceManager & SourceMgr,const TextDiagnosticBuffer & Buffer,ExpectedData & ED)824 static unsigned CheckResults(DiagnosticsEngine &Diags, SourceManager &SourceMgr,
825 const TextDiagnosticBuffer &Buffer,
826 ExpectedData &ED) {
827 // We want to capture the delta between what was expected and what was
828 // seen.
829 //
830 // Expected \ Seen - set expected but not seen
831 // Seen \ Expected - set seen but not expected
832 unsigned NumProblems = 0;
833
834 const DiagnosticLevelMask DiagMask =
835 Diags.getDiagnosticOptions().getVerifyIgnoreUnexpected();
836
837 // See if there are error mismatches.
838 NumProblems += CheckLists(Diags, SourceMgr, "error", ED.Errors,
839 Buffer.err_begin(), Buffer.err_end(),
840 bool(DiagnosticLevelMask::Error & DiagMask));
841
842 // See if there are warning mismatches.
843 NumProblems += CheckLists(Diags, SourceMgr, "warning", ED.Warnings,
844 Buffer.warn_begin(), Buffer.warn_end(),
845 bool(DiagnosticLevelMask::Warning & DiagMask));
846
847 // See if there are remark mismatches.
848 NumProblems += CheckLists(Diags, SourceMgr, "remark", ED.Remarks,
849 Buffer.remark_begin(), Buffer.remark_end(),
850 bool(DiagnosticLevelMask::Remark & DiagMask));
851
852 // See if there are note mismatches.
853 NumProblems += CheckLists(Diags, SourceMgr, "note", ED.Notes,
854 Buffer.note_begin(), Buffer.note_end(),
855 bool(DiagnosticLevelMask::Note & DiagMask));
856
857 return NumProblems;
858 }
859
UpdateParsedFileStatus(SourceManager & SM,FileID FID,ParsedStatus PS)860 void VerifyDiagnosticConsumer::UpdateParsedFileStatus(SourceManager &SM,
861 FileID FID,
862 ParsedStatus PS) {
863 // Check SourceManager hasn't changed.
864 setSourceManager(SM);
865
866 #ifndef NDEBUG
867 if (FID.isInvalid())
868 return;
869
870 const FileEntry *FE = SM.getFileEntryForID(FID);
871
872 if (PS == IsParsed) {
873 // Move the FileID from the unparsed set to the parsed set.
874 UnparsedFiles.erase(FID);
875 ParsedFiles.insert(std::make_pair(FID, FE));
876 } else if (!ParsedFiles.count(FID) && !UnparsedFiles.count(FID)) {
877 // Add the FileID to the unparsed set if we haven't seen it before.
878
879 // Check for directives.
880 bool FoundDirectives;
881 if (PS == IsUnparsedNoDirectives)
882 FoundDirectives = false;
883 else
884 FoundDirectives = !LangOpts || findDirectives(SM, FID, *LangOpts);
885
886 // Add the FileID to the unparsed set.
887 UnparsedFiles.insert(std::make_pair(FID,
888 UnparsedFileStatus(FE, FoundDirectives)));
889 }
890 #endif
891 }
892
CheckDiagnostics()893 void VerifyDiagnosticConsumer::CheckDiagnostics() {
894 // Ensure any diagnostics go to the primary client.
895 DiagnosticConsumer *CurClient = Diags.getClient();
896 std::unique_ptr<DiagnosticConsumer> Owner = Diags.takeClient();
897 Diags.setClient(PrimaryClient, false);
898
899 #ifndef NDEBUG
900 // In a debug build, scan through any files that may have been missed
901 // during parsing and issue a fatal error if directives are contained
902 // within these files. If a fatal error occurs, this suggests that
903 // this file is being parsed separately from the main file, in which
904 // case consider moving the directives to the correct place, if this
905 // is applicable.
906 if (!UnparsedFiles.empty()) {
907 // Generate a cache of parsed FileEntry pointers for alias lookups.
908 llvm::SmallPtrSet<const FileEntry *, 8> ParsedFileCache;
909 for (const auto &I : ParsedFiles)
910 if (const FileEntry *FE = I.second)
911 ParsedFileCache.insert(FE);
912
913 // Iterate through list of unparsed files.
914 for (const auto &I : UnparsedFiles) {
915 const UnparsedFileStatus &Status = I.second;
916 const FileEntry *FE = Status.getFile();
917
918 // Skip files that have been parsed via an alias.
919 if (FE && ParsedFileCache.count(FE))
920 continue;
921
922 // Report a fatal error if this file contained directives.
923 if (Status.foundDirectives()) {
924 llvm::report_fatal_error(Twine("-verify directives found after rather"
925 " than during normal parsing of ",
926 StringRef(FE ? FE->getName() : "(unknown)")));
927 }
928 }
929
930 // UnparsedFiles has been processed now, so clear it.
931 UnparsedFiles.clear();
932 }
933 #endif // !NDEBUG
934
935 if (SrcManager) {
936 // Produce an error if no expected-* directives could be found in the
937 // source file(s) processed.
938 if (Status == HasNoDirectives) {
939 Diags.Report(diag::err_verify_no_directives).setForceEmit();
940 ++NumErrors;
941 Status = HasNoDirectivesReported;
942 }
943
944 // Check that the expected diagnostics occurred.
945 NumErrors += CheckResults(Diags, *SrcManager, *Buffer, ED);
946 } else {
947 const DiagnosticLevelMask DiagMask =
948 ~Diags.getDiagnosticOptions().getVerifyIgnoreUnexpected();
949 if (bool(DiagnosticLevelMask::Error & DiagMask))
950 NumErrors += PrintUnexpected(Diags, nullptr, Buffer->err_begin(),
951 Buffer->err_end(), "error");
952 if (bool(DiagnosticLevelMask::Warning & DiagMask))
953 NumErrors += PrintUnexpected(Diags, nullptr, Buffer->warn_begin(),
954 Buffer->warn_end(), "warn");
955 if (bool(DiagnosticLevelMask::Remark & DiagMask))
956 NumErrors += PrintUnexpected(Diags, nullptr, Buffer->remark_begin(),
957 Buffer->remark_end(), "remark");
958 if (bool(DiagnosticLevelMask::Note & DiagMask))
959 NumErrors += PrintUnexpected(Diags, nullptr, Buffer->note_begin(),
960 Buffer->note_end(), "note");
961 }
962
963 Diags.setClient(CurClient, Owner.release() != nullptr);
964
965 // Reset the buffer, we have processed all the diagnostics in it.
966 Buffer.reset(new TextDiagnosticBuffer());
967 ED.Reset();
968 }
969
create(bool RegexKind,SourceLocation DirectiveLoc,SourceLocation DiagnosticLoc,bool MatchAnyLine,StringRef Text,unsigned Min,unsigned Max)970 std::unique_ptr<Directive> Directive::create(bool RegexKind,
971 SourceLocation DirectiveLoc,
972 SourceLocation DiagnosticLoc,
973 bool MatchAnyLine, StringRef Text,
974 unsigned Min, unsigned Max) {
975 if (!RegexKind)
976 return llvm::make_unique<StandardDirective>(DirectiveLoc, DiagnosticLoc,
977 MatchAnyLine, Text, Min, Max);
978
979 // Parse the directive into a regular expression.
980 std::string RegexStr;
981 StringRef S = Text;
982 while (!S.empty()) {
983 if (S.startswith("{{")) {
984 S = S.drop_front(2);
985 size_t RegexMatchLength = S.find("}}");
986 assert(RegexMatchLength != StringRef::npos);
987 // Append the regex, enclosed in parentheses.
988 RegexStr += "(";
989 RegexStr.append(S.data(), RegexMatchLength);
990 RegexStr += ")";
991 S = S.drop_front(RegexMatchLength + 2);
992 } else {
993 size_t VerbatimMatchLength = S.find("{{");
994 if (VerbatimMatchLength == StringRef::npos)
995 VerbatimMatchLength = S.size();
996 // Escape and append the fixed string.
997 RegexStr += llvm::Regex::escape(S.substr(0, VerbatimMatchLength));
998 S = S.drop_front(VerbatimMatchLength);
999 }
1000 }
1001
1002 return llvm::make_unique<RegexDirective>(
1003 DirectiveLoc, DiagnosticLoc, MatchAnyLine, Text, Min, Max, RegexStr);
1004 }
1005