https://build.chromium.org/p/chromium.memory.fyi/builders/Linux%20Tests%20%28valgrind%29%282%29/builds/54616/steps/memory_test%3A%20extensions_unittests/logs/stdio reports:

Leak_DefinitelyLost
64 bytes in 1 blocks are definitely lost in loss record 1,567 of 2,723
  operator new(unsigned long) (m_replacemalloc/vg_replace_malloc.c:1140)
7 new files were left in /tmp: Fix the tests to clean up themselves.
  mojo::edk::js::HandleWrapper::Create(v8::Isolate*, unsigned int) (/mnt/data/b/build/slave/chromium-rel-linux-valgrind-tests-2/build/src/out/Release/extensions_unittests)
  gin::Converter<mojo::Handle, void>::ToV8(v8::Isolate*, mojo::Handle const&) (mojo/edk/js/handle.cc:53)
  v8::Local<v8::Value> gin::ConvertToV8<mojo::Handle>(v8::Isolate*, mojo::Handle) (/mnt/data/b/build/slave/chromium-rel-linux-valgrind-tests-2/build/src/out/Release/extensions_unittests)
  gin::ToV8Traits<mojo::Handle, false>::TryConvertToV8(v8::Isolate*, mojo::Handle, v8::Local<v8::Value>*) (gin/converter.h:239)
  bool gin::TryConvertToV8<mojo::Handle>(v8::Isolate*, mojo::Handle, v8::Local<v8::Value>*) (/mnt/data/b/build/slave/chromium-rel-linux-valgrind-tests-2/build/src/out/Release/extensions_unittests)
  bool gin::Dictionary::Set<mojo::Handle>(std::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, mojo::Handle) (gin/dictionary.h:47)
  mojo::edk::js::(anonymous namespace)::CreateDataPipe(gin::Arguments const&) (mojo/edk/js/core.cc:230)


HandleWrapper::Create is:

  static gin::Handle<HandleWrapper> Create(v8::Isolate* isolate,
                                           MojoHandle handle) {
    return gin::CreateHandle(isolate, new HandleWrapper(handle));
  }


gin::CreateHandle is:

template<typename T>
gin::Handle<T> CreateHandle(v8::Isolate* isolate, T* object) {
  v8::Local<v8::Object> wrapper = object->GetWrapper(isolate);
  if (wrapper.IsEmpty())
    return gin::Handle<T>();
  return gin::Handle<T>(wrapper, object);
}

I see a path in CreateHandle that returns without freeing |object|. I think object->GetWrapper() intends to 'delete this' when it returns an empty object, but the logic to do that is complex and hard to follow, and I'm not sure it's working.

It's not obvious which change in https://build.chromium.org/p/chromium.memory.fyi/builders/Linux%20Tests%20%28valgrind%29%282%29/builds/54616extensions_unittests made this start happening, but it isn't flaky.