llvm-cov produces inconsistent result when file appears multiple times in gcno

[tenta4]

Summary

My gcno/gcda files contain some of sources multiple times and sometimes they have different paths formats.
For example

llvm-cov gcov test_unit_MXR_Settings_SettingsEngine_CSettingsHandler.dir/__/__/__/src/CSettingsHandler.cpp.gcno -b -x -i  | grep CSettingsHandler.cpp 
File '../../../src/Settings/SettingsEngine/test/unit_test/CSettingsHandler/../../../src/CSettingsHandler.cpp'
Lines executed:75.00% of 8
Branches executed:100.00% of 12
Taken at least once:41.67% of 12
No calls

--
File '../../../src/Settings/SettingsEngine/src/CSettingsHandler.cpp'
Lines executed:98.37% of 184
Branches executed:97.34% of 601
Taken at least once:56.91% of 601
No calls

llvm-cov randomly takes one of records and adds it to cov-file.
That is why my coverage results usually wrong and furthermore they differ every now and than

My patch unifies filenames, so coverage results are merged.

llvm-cov gcov /home/tenta4/work/arcreator/mxr2/build/build_products_cov/build_host_release/src/Settings/SettingsEngine/test/unit_test/CSettingsHandler/CMakeFiles/test_unit_MXR_Settings_SettingsEngine_CSettingsHandler.dir/__/__/__/src/CSettingsHandler.cpp.gcno -b -x -i  | grep CSettingsHandler.cpp -A5
File '../../../src/Settings/SettingsEngine/src/CSettingsHandler.cpp'
Lines executed:97.37% of 190
Branches executed:97.39% of 613
Taken at least once:56.61% of 613
No calls

[bogner]

llvm/lib/ProfileData/GCOV.cpp

146

Why not SmallString<256> P(filename)?

Regardless, it might help readability to add a function named something like "addNormalizedPathToMap" that does this and calls emplace as below rather than doing this all inline here.

[tenta4]

Please review 🙏

[bogner]

Looks good. Do you need me to commit this for you?

[tenta4]

Yes, do it please, since I do not have access.
Many thinks.
Regards, Oleksii Odynochenko <[e-mail]>

llvm/include/llvm/ProfileData/GCOV.h

public:
   std::map<uint32_t, GCOVFunction *> identToFunction;
   uint32_t runCount = 0;
   uint32_t programCount = 0;
 
   using iterator = pointee_iterator<
       SmallVectorImpl<std::unique_ptr<GCOVFunction>>::const_iterator>;
   iterator begin() const { return iterator(functions.begin()); }
   iterator end() const { return iterator(functions.end()); }
+
+private:
+  unsigned addNormalizedPathToMap(StringRef filename);
 };
 
 struct GCOVArc {
   GCOVArc(GCOVBlock &src, GCOVBlock &dst, uint32_t flags)
       : src(src), dst(dst), flags(flags) {}
   bool onTree() const;
 
   GCOVBlock &src;

llvm/lib/ProfileData/GCOV.cpp

if (tag == GCOV_TAG_FUNCTION) {
         if (!buf.readString(filename))
           return false;
         fn->startLine = buf.getWord();
         fn->startColumn = buf.getWord();
         fn->endLine = buf.getWord();
         if (version >= GCOV::V900)
           fn->endColumn = buf.getWord();
       }
-      auto r = filenameToIdx.try_emplace(filename, filenameToIdx.size());
-      if (r.second)
-        filenames.emplace_back(filename);
-      fn->srcIdx = r.first->second;
+      fn->srcIdx = addNormalizedPathToMap(filename);
       identToFunction[fn->ident] = fn;
     } else if (tag == GCOV_TAG_BLOCKS && fn) {
       if (version < GCOV::V800) {
         for (uint32_t i = 0; i != length; ++i) {
           buf.getWord(); // Ignored block flags
           fn->blocks.push_back(std::make_unique<GCOVBlock>(i));
         }
       } else {
         uint32_t num = buf.getWord();
         for (uint32_t i = 0; i != num; ++i)
           fn->blocks.push_back(std::make_unique<GCOVBlock>(i));
for (const GCOVFunction &f : *this)
     f.print(OS);
 }
 
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
 /// dump - Dump GCOVFile content to dbgs() for debugging purposes.
 LLVM_DUMP_METHOD void GCOVFile::dump() const { print(dbgs()); }
 #endif
 
+unsigned GCOVFile::addNormalizedPathToMap(StringRef filename) {
+  // unify filename, as the same path can have different form
+  SmallString<256> P(filename);
+  sys::path::remove_dots(P, true);
+  filename = P.str();
+
+  auto r = filenameToIdx.try_emplace(filename, filenameToIdx.size());
+  if (r.second)
+    filenames.emplace_back(filename);
+
+  return r.first->second;
+}
+
 bool GCOVArc::onTree() const { return flags & GCOV_ARC_ON_TREE; }
 
 //===----------------------------------------------------------------------===//
 // GCOVFunction implementation.
 
 StringRef GCOVFunction::getName(bool demangle) const {
   if (!demangle)
     return Name;

llvm/test/tools/llvm-cov/gcov/Inputs/test_no_gcda.h.gcov

-        -:    0:Source:./test.h
+        -:    0:Source:test.h
         -:    0:Graph:test.gcno
         -:    0:Data:-
         -:    0:Runs:0
         -:    0:Programs:0
         -:    1:#include <stdint.h>
     #####:    2:struct A {
         -:    3:  virtual void B();
         -:    4:};

llvm/test/tools/llvm-cov/gcov/Inputs/test_no_options.h.gcov

-        -:    0:Source:./test.h
+        -:    0:Source:test.h
         -:    0:Graph:test.gcno
         -:    0:Data:test.gcda
         -:    0:Runs:1
         -:    0:Programs:1
         -:    1:#include <stdint.h>
         1:    2:struct A {
         -:    3:  virtual void B();
         -:    4:};

llvm/test/tools/llvm-cov/gcov/basic.test

 # Print to stdout.
 RUN: llvm-cov gcov -t test.c > stdout
 RUN: llvm-cov gcov --stdout test.c | cmp stdout -
 RUN: cat test_no_options.cpp.gcov test_no_options.h.gcov | diff -u - stdout
 RUN: llvm-cov gcov -n -t test.c | count 0
 
 RUN: llvm-cov gcov test_paths.cpp 2>/dev/null | FileCheck %s --check-prefix=MISSING
 RUN: ls a.c.gcov
-MISSING: File 'src/./n/../a.c'
+MISSING: File 'src/a.c'
 MISSING: Creating 'a.c.gcov'
 
 # Preserve paths. This mangles the output filenames.
 RUN: llvm-cov gcov -p test_paths.cpp 2>/dev/null | FileCheck %s --check-prefix=PRESERVE
-RUN: ls src#n#^#a.c.gcov
-PRESERVE: File 'src/./n/../a.c'
-PRESERVE: Creating 'src#n#^#a.c.gcov'
+RUN: ls src#a.c.gcov
+PRESERVE: File 'src/a.c'
+PRESERVE: Creating 'src#a.c.gcov'
 
 # Long file names.
 RUN: llvm-cov gcov -l test_paths.cpp
 RUN: ls test_paths.cpp##a.c.gcov
 
 # Long file names and preserve paths.
 RUN: mkdir -p src && llvm-cov gcov -lp -gcno test_paths.gcno -gcda test_paths.gcda src/../test_paths.cpp
-RUN: ls src#^#test_paths.cpp##src#n#^#a.c.gcov
+RUN: ls src#^#test_paths.cpp##src#a.c.gcov
 
 # Hash pathnames.
 RUN: llvm-cov gcov -x -gcno test_paths.gcno -gcda test_paths.gcda src/../test_paths.cpp
-RUN: ls a.c##4784150e272908907eaa7380ca3eced8.gcov
+RUN: ls a.c##0c546a4dd99c1774b7b06e4fad16158c.gcov
 
 # Function summaries. This changes stdout, but not the gcov files.
 RUN: llvm-cov gcov test.c -f | FileCheck %s --check-prefixes=OUT,OUTFILE,OUT-F
 RUN: FileCheck --input-file=test.cpp.gcov %s --check-prefix=C
 RUN: FileCheck --input-file=test.h.gcov %s --check-prefix=H
 
 # All blocks. This doesn't affect stdout, only the gcov files.
 RUN: llvm-cov gcov test.c -a | FileCheck %s --check-prefixes=OUT,OUTFILE

  OUT-F-EMPTY:
          OUT:File 'test.cpp'
     OUT-NEXT:Lines executed:81.40% of 43
   OUT-B-NEXT:Branches executed:100.00% of 15
   OUT-B-NEXT:Taken at least once:86.67% of 15
   OUT-B-NEXT:No calls
 OUTFILE-NEXT:Creating 'test.cpp.gcov'
    OUT-EMPTY:
-    OUT-NEXT:File './test.h'
+    OUT-NEXT:File 'test.h'
     OUT-NEXT:Lines executed:100.00% of 1
   OUT-B-NEXT:No branches
   OUT-B-NEXT:No calls
 OUTFILE-NEXT:Creating 'test.h.gcov'
 
 # Summarize unconditional branches too.
 RUN: llvm-cov gcov test.c -a -b -u | FileCheck %s --check-prefixes=OUT,OUTFILE,OUT-B
 RUN: FileCheck --input-file=test.cpp.gcov %s --check-prefixes=C,C-A,C-B,C-U
 RUN: FileCheck --input-file=test.h.gcov %s --check-prefixes=H,H-A,H-B,H-U
 
 # Absolute counts for branches.
 RUN: llvm-cov gcov test.c -a -b -c -u
 RUN: FileCheck --input-file=test.h.gcov %s --check-prefix=H-C
 H-C: unconditional  0 taken 1
 
 # Missing gcda file just gives 0 counts.
 RUN: llvm-cov gcov test.c -gcda=no_such_gcda_file | FileCheck %s --check-prefix=NO-GCDA
 RUN: diff -ub test_no_gcda.cpp.gcov test.cpp.gcov
 RUN: diff -ub test_no_gcda.h.gcov test.h.gcov
 NO-GCDA:       File 'test.cpp'
 NO-GCDA-NEXT:  Lines executed:0.00% of 43
 NO-GCDA-NEXT:  Creating 'test.cpp.gcov'
 NO-GCDA-EMPTY:
-NO-GCDA-NEXT:  File './test.h'
+NO-GCDA-NEXT:  File 'test.h'
 NO-GCDA-NEXT:  Lines executed:0.00% of 1
 NO-GCDA-NEXT:  Creating 'test.h.gcov'
 
 # Invalid gcno file.
 RUN: llvm-cov gcov test.c -gcno=test_read_fail.gcno
 
 # Not a .gcda file. Error but keep the .gcov output.
 RUN: echo invalid > not.gcda
C-NEXT:4294967296:   70:    thing.B();
    C-U:unconditional  0 taken 100%
 C-NEXT:        -:   71:
 C-NEXT:        1:   72:  return a + 8 + grid[2][3] + len;
    C-A:        1:   72-block  0
    C-U:unconditional  0 taken 100%
 C-NEXT:        -:   73:  return more_useless();
 C-NEXT:        -:   74:}
 
-     H:        -:    0:Source:./test.h
+     H:        -:    0:Source:test.h
 H-NEXT:        -:    0:Graph:test.gcno
 H-NEXT:        -:    0:Data:test.gcda
 H-NEXT:        -:    0:Runs:1
 H-NEXT:        -:    0:Programs:1
 H-NEXT:        -:    1:#include <stdint.h>
    H-B:function _ZN1AC2Ev called 1 returned 100% blocks executed 100%
 H-NEXT:        1:    2:struct A {
    H-A:        1:    2-block  0
    H-U:unconditional  0 taken 100%
 H-NEXT:        -:    3:  virtual void B();
 H-NEXT:        -:    4:};

llvm/test/tools/llvm-cov/gcov/gcov-fake-4.2.c

 /// Test that llvm-cov supports a fake gcov 4.2 format used before clang 11.
 
 // RUN: rm -rf %t && mkdir %t && cd %t
 // RUN: echo -e '\n\n\n\n\n\n\n\n\n' > test.cpp && echo > test.h
 // RUN: llvm-cov gcov test. --gcno=%S/Inputs/gcov-fake-4.2.gcno --gcda=%S/Inputs/gcov-fake-4.2.gcda | FileCheck %s
 // RUN: FileCheck %s --check-prefix=C < test.cpp.gcov
 // RUN: FileCheck %s --check-prefix=H < test.h.gcov
 
 // CHECK:       File 'test.cpp'
 // CHECK-NEXT:  Lines executed:84.21% of 38
 // CHECK-NEXT:  Creating 'test.cpp.gcov'
 // CHECK-EMPTY:
-// CHECK-NEXT:  File './test.h'
+// CHECK-NEXT:  File 'test.h'
 // CHECK-NEXT:  Lines executed:100.00% of 1
 // CHECK-NEXT:  Creating 'test.h.gcov'
 // CHECK-EMPTY:
 
 //      C:        -:    0:Source:test.cpp
 // C-NEXT:        -:    0:Graph:{{.*}}gcov-fake-4.2.gcno
 // C-NEXT:        -:    0:Data:{{.*}}gcov-fake-4.2.gcda
 /// `Runs` is stored in GCOV_TAG_OBJECT_SUMMARY with a length of 9.
 // C-NEXT:        -:    0:Runs:2
 // C-NEXT:        -:    0:Programs:1
 // C-NEXT:        -:    1:
 // C-NEXT:        -:    2:
 // C-NEXT:        -:    3:
 // C-NEXT:        -:    4:
 // C-NEXT:        -:    5:
 // C-NEXT:        -:    6:
 // C-NEXT:        -:    7:
 // C-NEXT:        -:    8:
 // C-NEXT:        -:    9:
 // C-NEXT:8589934592:   10:
 
-//      H:        -:    0:Source:./test.h
+//      H:        -:    0:Source:test.h
 // H-NEXT:        -:    0:Graph:{{.*}}gcov-fake-4.2.gcno
 // H-NEXT:        -:    0:Data:{{.*}}gcov-fake-4.2.gcda
 // H-NEXT:        -:    0:Runs:2
 // H-NEXT:        -:    0:Programs:1
 // H-NEXT:        4:    1:

llvm/test/tools/llvm-cov/gcov/intermediate-format.test

 RUN: cmp test.cpp.gcov saved.cpp.gcov
 
          OUT:File 'test.cpp'
     OUT-NEXT:Lines executed:81.40% of 43
   OUT-B-NEXT:Branches executed:100.00% of 15
   OUT-B-NEXT:Taken at least once:86.67% of 15
   OUT-B-NEXT:No calls
    OUT-EMPTY:
-    OUT-NEXT:File './test.h'
+    OUT-NEXT:File 'test.h'
     OUT-NEXT:Lines executed:100.00% of 1
   OUT-B-NEXT:No branches
   OUT-B-NEXT:No calls
    OUT-EMPTY:
 
        I:file:test.cpp
        I:function:10,4294967296,_ZN1A1BEv
   I-NEXT:function:12,0,_Z7uselessv
   I-NEXT:function:14,0,_Z12more_uselessv
   I-NEXT:function:18,1,_Z3foov
   I-NEXT:function:23,0,_Z3barv
   I-NEXT:function:28,4,_Z6assignii
   I-NEXT:function:32,1,_Z15initialize_gridv
   I-NEXT:function:38,1,main
        I:lcount:10,4294967296
        I:lcount:33,3
 I-B-NEXT:branch:33,taken
 I-B-NEXT:branch:33,taken
-       I:file:./test.h
+       I:file:test.h
   I-NEXT:function:2,1,_ZN1AC2Ev
   I-NEXT:lcount:2,1