On 10/4/19 5:52 PM, Brendan Higgins wrote:
On Fri, Oct 4, 2019 at 4:30 PM Theodore Y. Ts'o tytso@mit.edu wrote:
On Fri, Oct 04, 2019 at 04:47:09PM -0600, shuah wrote:
However, if I encourage arbitrary tests/improvements into my KUnit branch, it further diverges away from torvalds/master, and is more likely that there will be a merge conflict or issue that is not related to the core KUnit changes that will cause the whole thing to be rejected again in v5.5.
The idea is that the new development will happen based on kunit in linux-kselftest next. It will work just fine. As we accepts patches, they will go on top of kunit that is in linux-next now.
I don't see how this would work. If I add unit tests to ext4, they would be in fs/ext4. And to the extent that I need to add test mocks to allow the unit tests to work, they will involve changes to existing files in fs/ext4. I can't put them in the ext4.git tree without pulling in the kunit changes into the ext4 git tree. And if they ext4 unit tests land in the kunit tree, it would be a receipe for large numbers of merge conflicts.
That's where I was originally coming from.
So here's a dumb idea: what if we merged KUnit through the ext4 tree? I imagine that could potentially get very confusing when we go back to sending changes in through the kselftest tree, but at least it means that ext4 can use it in the meantime, which means that it at least gets to be useful to one group of people. Also, since Ted seems pretty keen on using this, I imagine it is much more likely to produce real world, immediately useful tests not written by me (I'm not being lazy, I just think it is better to get other people's experiences other than my own).
That doesn't make sense does it? The tests might not be limited to fs/ext4. We might have other sub-systems that add tests.
So, we will have to work to make linux-next as the base for new development and limit the number of tests to where it will be easier work in this mode for 5.5. We can stage the pull requests so that kunit lands first followed by tests.
We have a similar situation with kselftest as well. Sub-systems send tests that depend on their tress separately.
thanks, -- Shuah