Extend SpecialCaseList to allow users to blame matches on entries in the file.

[hctim]

Summary

Extends SCL functionality to allow users to find the line number in the file the SCL is built from through SpecialCaseList::inSectionBlame(...).

Also removes the need to compile the SCL before use. As the matcher now contains a list of regexes to test against instead of a single regex, the regexes can be individually built on each insertion rather than one large compilation at the end of construction.

This change also fixes a bug where blank lines would cause the parser to become out-of-sync with the line number. An error on line k was being reported as being on line k - num_blank_lines_before_k.

Note: This change has a cyclical dependency on D39486. Both these changes must be submitted at the same time to avoid a build breakage.

[pcc]

I'd be curious about what sort of perf impact (positive or negative) comes from splitting the special case list into multiple regexes, as this part of the compiler has historically been perf sensitive. It may be that the trigram filter (which is relatively new) makes the impact relatively insignificant.

Perhaps a good way to measure would be to compile a relatively large Chromium TU with/without this patch.

[vlad.tsyrklevich]

I share the same concern as Peter, the TrigramIndex might save us from the performance hit but we are potentially blowing up the amount of regex comparisons we have to perform. With the very wide new cfi-icall blacklist in Chrome we will pass the TrigramIndex for significant parts of Chrome and all those entries are at the bottom of the list so we have to iteratively go through the entire blacklist before hitting those entries. I'd be curious to measure this for a Chrome build with use_cfi_icall=true specifically.

lib/Support/SpecialCaseList.cpp

141

Doesn't the loop already handle this?

[hctim]

Fixed line parsing to not ignore blank lines, line number is correctly generated.

lib/Support/SpecialCaseList.cpp

141

Too right. There is still a latent bug however - llvm::SplitString culls blank lines from the output vector, so it was generating the wrong line numbers for anything following a blank line.

I've fixed this up and made sure that a unit test is there that catches both these cases.

[hctim]

Accidently made last modifications on an old patch, merged the changes across onto the most recent patch and fixed up the regressions.

[vlad.tsyrklevich]

include/llvm/Support/SpecialCaseList.h

115

This is no longer true.

116

s/Set/Map/

[hctim]

Updated with Vlad's comments and rebased over master in preparation for submission.

include/llvm/Support/SpecialCaseList.h

public:
   /// Returns true, if special case list contains a line
   /// \code
   ///   @Prefix:<E>=@Category
   /// \endcode
   /// where @Query satisfies wildcard expression <E> in a given @Section.
   bool inSection(StringRef Section, StringRef Prefix, StringRef Query,
                  StringRef Category = StringRef()) const;
 
+  /// Returns the line number corresponding to the special case list entry if
+  /// the special case list contains a line
+  /// \code
+  ///   @Prefix:<E>=@Category
+  /// \endcode
+  /// where @Query satisfies wildcard expression <E> in a given @Section.
+  /// Returns zero if there is no blacklist entry corresponding to this
+  /// expression.
+  unsigned inSectionBlame(StringRef Section, StringRef Prefix, StringRef Query,
+                          StringRef Category = StringRef()) const;
+
 protected:
   // Implementations of the create*() functions that can also be used by derived
   // classes.
   bool createInternal(const std::vector<std::string> &Paths,
                       std::string &Error);
   bool createInternal(const MemoryBuffer *MB, std::string &Error);
 
   SpecialCaseList(SpecialCaseList const &) = delete;
   SpecialCaseList &operator=(SpecialCaseList const &) = delete;
 
   /// Represents a set of regular expressions.  Regular expressions which are
   /// "literal" (i.e. no regex metacharacters) are stored in Strings, while all
   /// others are represented as a single pipe-separated regex in RegEx.  The
   /// reason for doing so is efficiency; StringSet is much faster at matching
   /// literal strings than Regex.
   class Matcher {
   public:
-    bool insert(std::string Regexp, std::string &REError);
+    bool insert(std::string Regexp, unsigned LineNumber, std::string &REError);
     void compile();
-    bool match(StringRef Query) const;
+    // Returns the line number in the source file that this query matches to.
+    // Returns zero if no match is found.
+    unsigned match(StringRef Query) const;
 
   private:
-    StringSet<> Strings;
+    StringMap<unsigned> Strings;
     TrigramIndex Trigrams;
-    std::unique_ptr<Regex> RegEx;
-    std::string UncompiledRegEx;
+    std::vector<std::pair<std::unique_ptr<Regex>, unsigned>> RegExes;
   };
 
   using SectionEntries = StringMap<StringMap<Matcher>>;
 
   struct Section {
     Section(std::unique_ptr<Matcher> M) : SectionMatcher(std::move(M)){};
 
     std::unique_ptr<Matcher> SectionMatcher;
     SectionEntries Entries;
   };
 
   std::vector<Section> Sections;
   bool IsCompiled;
 
   SpecialCaseList();
   /// Parses just-constructed SpecialCaseList entries from a memory buffer.
   bool parse(const MemoryBuffer *MB, StringMap<size_t> &SectionsMap,
              std::string &Error);
   /// compile() should be called once, after parsing all the memory buffers.
   void compile();
 
   // Helper method for derived classes to search by Prefix, Query, and Category
   // once they have already resolved a section entry.
-  bool inSection(const SectionEntries &Entries, StringRef Prefix,
+  unsigned inSectionBlame(const SectionEntries &Entries, StringRef Prefix,
                           StringRef Query, StringRef Category) const;
 };
 
 }  // namespace llvm
 
 #endif  // LLVM_SUPPORT_SPECIALCASELIST_H
 

lib/Support/SpecialCaseList.cpp

 #include <string>
 #include <system_error>
 #include <utility>
 
 #include <stdio.h>
 namespace llvm {
 
 bool SpecialCaseList::Matcher::insert(std::string Regexp,
+                                      unsigned LineNumber,
                                       std::string &REError) {
   if (Regexp.empty()) {
     REError = "Supplied regexp was blank";
     return false;
   }
 
   if (Regex::isLiteralERE(Regexp)) {
-    Strings.insert(Regexp);
+    Strings[Regexp] = LineNumber;
     return true;
   }
   Trigrams.insert(Regexp);
 
   // Replace * with .*
   for (size_t pos = 0; (pos = Regexp.find('*', pos)) != std::string::npos;
        pos += strlen(".*")) {
     Regexp.replace(pos, strlen("*"), ".*");
   }
 
+  Regexp = "^(" + Regexp + ")$";
+
   // Check that the regexp is valid.
   Regex CheckRE(Regexp);
   if (!CheckRE.isValid(REError))
     return false;
 
-  if (!UncompiledRegEx.empty())
-    UncompiledRegEx += "|";
-  UncompiledRegEx += "^(" + Regexp + ")$";
+  RegExes.emplace_back(
+      std::make_pair(make_unique<Regex>(std::move(CheckRE)), LineNumber));
   return true;
 }
 
-void SpecialCaseList::Matcher::compile() {
-  if (!UncompiledRegEx.empty()) {
-    RegEx.reset(new Regex(UncompiledRegEx));
-    UncompiledRegEx.clear();
-  }
-}
-
-bool SpecialCaseList::Matcher::match(StringRef Query) const {
-  if (Strings.count(Query))
-    return true;
+unsigned SpecialCaseList::Matcher::match(StringRef Query) const {
+  auto It = Strings.find(Query);
+  if (It != Strings.end())
+    return It->second;
   if (Trigrams.isDefinitelyOut(Query))
     return false;
-  return RegEx && RegEx->match(Query);
+  for (auto& RegExKV : RegExes)
+    if (RegExKV.first->match(Query))
+      return RegExKV.second;
+  return 0;
 }
 
 SpecialCaseList::SpecialCaseList() : Sections(), IsCompiled(false) {}
 
 std::unique_ptr<SpecialCaseList>
 SpecialCaseList::create(const std::vector<std::string> &Paths,
                         std::string &Error) {
   std::unique_ptr<SpecialCaseList> SCL(new SpecialCaseList());
if (std::error_code EC = FileOrErr.getError()) {
       return false;
     }
     std::string ParseError;
     if (!parse(FileOrErr.get().get(), Sections, ParseError)) {
       Error = (Twine("error parsing file '") + Path + "': " + ParseError).str();
       return false;
     }
   }
-  compile();
   return true;
 }
 
 bool SpecialCaseList::createInternal(const MemoryBuffer *MB,
                                      std::string &Error) {
   StringMap<size_t> Sections;
   if (!parse(MB, Sections, Error))
     return false;
-  compile();
   return true;
 }
 
 bool SpecialCaseList::parse(const MemoryBuffer *MB,
                             StringMap<size_t> &SectionsMap,
                             std::string &Error) {
   // Iterate through each line in the blacklist file.
   SmallVector<StringRef, 16> Lines;
-  SplitString(MB->getBuffer(), Lines, "\n\r");
+  MB->getBuffer().split(Lines, '\n');
 
-  int LineNo = 1;
+  unsigned LineNo = 1;
   StringRef Section = "*";
+
   for (auto I = Lines.begin(), E = Lines.end(); I != E; ++I, ++LineNo) {
+    *I = I->trim();
     // Ignore empty lines and lines starting with "#"
     if (I->empty() || I->startswith("#"))
       continue;
 
     // Save section names
     if (I->startswith("[")) {
       if (!I->endswith("]")) {
         Error = (Twine("malformed section header on line ") + Twine(LineNo) +
                  ": " + *I).str();
         return false;
       }
for (auto I = Lines.begin(), E = Lines.end(); I != E; ++I, ++LineNo) {
     std::pair<StringRef, StringRef> SplitRegexp = SplitLine.second.split("=");
     std::string Regexp = SplitRegexp.first;
     StringRef Category = SplitRegexp.second;
 
     // Create this section if it has not been seen before.
     if (SectionsMap.find(Section) == SectionsMap.end()) {
       std::unique_ptr<Matcher> M = make_unique<Matcher>();
       std::string REError;
-      if (!M->insert(Section, REError)) {
+      if (!M->insert(Section, LineNo, REError)) {
         Error = (Twine("malformed section ") + Section + ": '" + REError).str();
         return false;
       }
-      M->compile();
 
       SectionsMap[Section] = Sections.size();
       Sections.emplace_back(std::move(M));
     }
 
     auto &Entry = Sections[SectionsMap[Section]].Entries[Prefix][Category];
     std::string REError;
-    if (!Entry.insert(std::move(Regexp), REError)) {
+    if (!Entry.insert(std::move(Regexp), LineNo, REError)) {
       Error = (Twine("malformed regex in line ") + Twine(LineNo) + ": '" +
                SplitLine.second + "': " + REError).str();
       return false;
     }
   }
   return true;
 }
 
-void SpecialCaseList::compile() {
-  assert(!IsCompiled && "compile() should only be called once");
-  // Iterate through every section compiling regular expressions for every query
-  // and creating Section entries.
-  for (auto &Section : Sections)
-    for (auto &Prefix : Section.Entries)
-      for (auto &Category : Prefix.getValue())
-        Category.getValue().compile();
-
-  IsCompiled = true;
-}
-
 SpecialCaseList::~SpecialCaseList() {}
 
 bool SpecialCaseList::inSection(StringRef Section, StringRef Prefix,
                                 StringRef Query, StringRef Category) const {
-  assert(IsCompiled && "SpecialCaseList::compile() was not called!");
+  return inSectionBlame(Section, Prefix, Query, Category);
+}
 
+unsigned SpecialCaseList::inSectionBlame(StringRef Section, StringRef Prefix,
+                                         StringRef Query,
+                                         StringRef Category) const {
   for (auto &SectionIter : Sections)
-    if (SectionIter.SectionMatcher->match(Section) &&
-        inSection(SectionIter.Entries, Prefix, Query, Category))
-      return true;
-
-  return false;
+    if (SectionIter.SectionMatcher->match(Section)) {
+      unsigned Blame =
+          inSectionBlame(SectionIter.Entries, Prefix, Query, Category);
+      if (Blame)
+        return Blame;
+    }
+  return 0;
 }
 
-bool SpecialCaseList::inSection(const SectionEntries &Entries, StringRef Prefix,
-                                StringRef Query, StringRef Category) const {
+unsigned SpecialCaseList::inSectionBlame(const SectionEntries &Entries,
+                                         StringRef Prefix, StringRef Query,
+                                         StringRef Category) const {
   SectionEntries::const_iterator I = Entries.find(Prefix);
-  if (I == Entries.end()) return false;
+  if (I == Entries.end()) return 0;
   StringMap<Matcher>::const_iterator II = I->second.find(Category);
-  if (II == I->second.end()) return false;
+  if (II == I->second.end()) return 0;
 
   return II->getValue().match(Query);
 }
 
 }  // namespace llvm

unittests/Support/SpecialCaseListTest.cpp

std::unique_ptr<SpecialCaseList> SCL =
                           "src:z*=category\n");
   EXPECT_TRUE(SCL->inSection("", "src", "hello"));
   EXPECT_TRUE(SCL->inSection("", "src", "bye"));
   EXPECT_TRUE(SCL->inSection("", "src", "hi", "category"));
   EXPECT_TRUE(SCL->inSection("", "src", "zzzz", "category"));
   EXPECT_FALSE(SCL->inSection("", "src", "hi"));
   EXPECT_FALSE(SCL->inSection("", "fun", "hello"));
   EXPECT_FALSE(SCL->inSection("", "src", "hello", "category"));
+
+  EXPECT_EQ(3u, SCL->inSectionBlame("", "src", "hello"));
+  EXPECT_EQ(4u, SCL->inSectionBlame("", "src", "bye"));
+  EXPECT_EQ(5u, SCL->inSectionBlame("", "src", "hi", "category"));
+  EXPECT_EQ(6u, SCL->inSectionBlame("", "src", "zzzz", "category"));
+  EXPECT_EQ(0u, SCL->inSectionBlame("", "src", "hi"));
+  EXPECT_EQ(0u, SCL->inSectionBlame("", "fun", "hello"));
+  EXPECT_EQ(0u, SCL->inSectionBlame("", "src", "hello", "category"));
+}
+
+TEST_F(SpecialCaseListTest, CorrectErrorLineNumberWithBlankLine) {
+  std::string Error;
+  EXPECT_EQ(nullptr, makeSpecialCaseList("# This is a comment.\n"
+                                         "\n"
+                                         "[not valid\n",
+                                         Error));
+  EXPECT_TRUE(
+      ((StringRef)Error).startswith("malformed section header on line 3:"));
+
+  EXPECT_EQ(nullptr, makeSpecialCaseList("\n\n\n"
+                                         "[not valid\n",
+                                         Error));
+  EXPECT_TRUE(
+      ((StringRef)Error).startswith("malformed section header on line 4:"));
 }
 
 TEST_F(SpecialCaseListTest, SectionRegexErrorHandling) {
   std::string Error;
   EXPECT_EQ(makeSpecialCaseList("[address", Error), nullptr);
   EXPECT_TRUE(((StringRef)Error).startswith("malformed section header "));
 
   EXPECT_EQ(makeSpecialCaseList("[[]", Error), nullptr);