> On April 17, 2015, 1:31 p.m., Michael Park wrote: > > src/local/local.cpp, lines 141-145 > > <https://reviews.apache.org/r/31265/diff/8/?file=929832#file929832line141> > > > > What's the point of setting `_allocator` here? It looks like it only > > gets passed to the `new Master(...)` call. > > > > Can we just leave `_allocator` alone and do > > > > ``` > > if (allocator_.isError()) { > > LOG(ERROR) << "Failed to create an instance of > > HierarchicalDRFAllocator: " > > << allocator_.error(); > > allocator = NULL; > > } > > > > // Save the instance for deleting later. > > allocator = allocator_.get(); > > ``` > > > > then pass `allocator` to `new Master` below?
`_allocator` is passed by non-const pointer, so theoretically we can modify caller's data. I can't find any code where this is the case, but I still don't feel like doing this drive-by change. Let's create a separate JIRA if you feel this should be fixed. > On April 17, 2015, 1:31 p.m., Michael Park wrote: > > src/tests/hierarchical_allocator_tests.cpp, line 77 > > <https://reviews.apache.org/r/31265/diff/8/?file=929836#file929836line77> > > > > I see this is the place where you wanted to use `ASSERT_SOME` but > > couldn't due to the [gtest > > limitation](https://code.google.com/p/googletest/wiki/AdvancedGuide#Assertion_Placement). > > > > One way to solve this I think is to use `SetUp/TearDown` rather than > > `Ctor/Dtor`. > > Another way may be to use `CHECK_SOME` in place of `ASSERT_SOME`. I > > think this makes sense as well since success of > > `HierarchicalDRFAllocator::create` isn't actually a testing condition. > > > > Thoughts? > > Michael Park wrote: > Actually, is it possible just use a `TestAllocator` constructed from > `TestAllocator::create<>()` as suggested below? Good idea! - Alexander ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/31265/#review80445 ----------------------------------------------------------- On April 15, 2015, 2:19 p.m., Alexander Rukletsov wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/31265/ > ----------------------------------------------------------- > > (Updated April 15, 2015, 2:19 p.m.) > > > Review request for mesos, Kapil Arya, Michael Park, and Niklas Nielsen. > > > Bugs: MESOS-2160 > https://issues.apache.org/jira/browse/MESOS-2160 > > > Repository: mesos > > > Description > ------- > > The factory creates allocator instances in a way identical to how instances > from modules are created. It allows us to use same typed tests for built-in > and modularized allocators. > > > Diffs > ----- > > src/examples/test_allocator_module.cpp PRE-CREATION > src/local/local.cpp 289b9bced7bab0d745fe14823efa4e90ec36905e > src/master/allocator/mesos/allocator.hpp > fb898f1175b61b442204e6e38c69ccc2838a646f > src/master/main.cpp 7cce3a0bb808a1cb7bac9acab31eb1c67a15ea9f > src/tests/cluster.hpp a56b6541adcdebc5866571bbdbb6828df97b34ec > src/tests/hierarchical_allocator_tests.cpp > 8861bf398e4bb17b0f74eab4f4af26202447ccef > src/tests/master_allocator_tests.cpp > 03a1bb8c92b44bc1ad1b5f5cff8d1fb971df2302 > src/tests/mesos.hpp 42e42ac425a448fcc5e93db1cef1112cbf5e67c4 > > Diff: https://reviews.apache.org/r/31265/diff/ > > > Testing > ------- > > make check (Mac OS 10.9.5, Ubuntu 14.04) > > > Thanks, > > Alexander Rukletsov > >
