archive mirror
 help / color / mirror / Atom feed
* kunit: what do we do with the 'kunit/alpha/master' branch?
@ 2019-09-17 18:41 David Gow
  2019-09-23 21:41 ` Brendan Higgins
  0 siblings, 1 reply; 8+ messages in thread
From: David Gow @ 2019-09-17 18:41 UTC (permalink / raw)
  To: Brendan Higgins, David Chiang, David Siebert, Kees Cook,
	Mike Salvatore, Pei Huang, Sagi Shahar, Sangsu Ha, Shuah Khan
  Cc: kunit-dev, linux-kselftest

[-- Attachment #1: Type: text/plain, Size: 8886 bytes --]

TL;DR: We expect KUnit to be accepted upstream into Linus' branch in
the next week or two, and we now need to figure out what we are going
to do with our non-upstream 'kunit/alpha/master' branch.

Hello everyone,

We've put together a rough proposal of what we should do with our
not-upstream branch, known to most people using it as
'kunit/alpha/master'[1], now that KUnit's acceptance into mainline
appears to be imminent (the KUnit MVP patchset is now in linux-next,
and the merge window just opened).


KUnit development is currently split between two versions: the
'kunit/alpha/master'[1] git branch, and the version being submitted to
the upstream Linux kernel. While there are some good reasons to
continue to have two separate versions of KUnit, at present there is
some uncertainty around the difference between these versions, and in
which circumstances each version is useful.

At present, the 'kunit/alpha/master' branch serves a few different
purposes. It is a place for code not-yet-ready for upstream -- such as
the mocking framework -- while being developed, while also acting as a
stable version for customers who do not wish to follow along with the
changes made during the upstreaming process. Adding to the confusion,
the name 'kunit/alpha/master' refers to an early (alpha) version of
KUnit, and the version of KUnit being upstreamed has now diverged
significantly from this version, requiring significant differences in
documentation, and requiring a number of changes to tests when porting
from one version to the other. Finally, it is not clear how the
'kunit/alpha/master' version should evolve as features it contains are

On the other hand, the version being upstreamed has its own
complications. It contains significantly fewer features (as features
such as the mocking frameworks will be upstreamed individually), and
so is less useful for the average customer. Until each feature is
upstreamed, it is iterated on rapidly to address comments from the
kernel community, so in-progress features are not stable enough to
reasonably build on. Finally, it exists only as a set of patches on
mailing lists, rather than as a maintained git repository (due to the
fact that the patches themselves are changing rapidly), making it
difficult for early adopters to incorporate into their own trees.

Whilst we believe there to be enough (at times conflicting) goals
above to justify having multiple versions of KUnit, we want to ensure
that they are meeting their goals, and that we have a process to
ensure that code finds its way into the correct version, that we can
deprecate and remove failed experiments or superseded versions, and
that we can keep pace with upstream kernel releases.

The Proposal

We propose having two tracks of development: the upstream kernel
(comprising both code that has been upstreamed, and code which is in
the process of being upstreamed -- i.e. is being reviewed on the
mailing lists), and an 'experimental' branch, which contains features
which are yet to be submitted upstream.

Periodically, the 'experimental' branch will be replaced with a new
'experimental' branch based on the latest upstream version. This will
allow us to focus on code based on upstream versions by deprecating
earlier, experimental versions of features. The older experimental
branch will stay around, but receive no new features or support:
customers will need to eventually migrate to either the upstream
kernel (if all of the features they require have been upstreamed), or
to the latest 'experimental' branch (if they still require
non-upstreamed features).

Where practical, 'experimental' branches will try to provide a stable
interface (and/or compatibility layers/shims) such that users can
write tests against an experimental version and expect our changes to
break them infrequently. We don't intend to make any solid guarantee,
however: so users will need to be able to deal with some breakage.

We expect most changes to KUnit to be submitted directly upstream, and
reserve the 'experimental' branch for features where:
a) the design is yet to be proven by early adopters, and is likely to change;
b) the feature is large enough that having a solid staging environment
in which several developers can work is necessary; and/or
c) it depends on existing features only available in the 'experimental' branch.

An example of features suited to the 'experimental' branch are the
mocking systems[2].

Lifetime of a feature

A new feature (that for one of the above reasons may require some
experimentation before being sent upstream) would typically be
developed as follows:

1. An initial version of the feature is developed against the latest
'experimental' branch.
2. The feature is tested by users of the 'experimental' branch.
3. It is revised as needed within the 'experimental' branch. If
breaking changes are required, and supporting some form of
compatibility layer is practical, such a layer is implemented to avoid
breaking existing 'experimental' customers.
4. Once the feature is deemed ready, it is submitted to the mailing
lists for the upstream kernel. Any compatibility layers for earlier
experimental version are removed -- this is a 'clean' version of the
5. Development of the change continues on the kernel mailing lists:
any particularly important changes or bugfixes are ported to the
'experimental' branch if feasible, but complete convergence is not
expected at this stage.
6. When the feature is accepted upstream, a new 'experimental' branch
is created based on the new upstream kernel. Development on the old
'experimental' branch ceases.
7. The upstream version of the feature has now superseded the
experimental version. All future development is done based on the
upstream version, either in the new 'experimental' branch, or directly
against the upstream kernel.
8. Any customers still using the experimental version of the feature
will need to update their code to work with the upstream version in
order to update to the new (supported) 'experimental' branch, or to
the upstream kernel.

What does this mean for me?
=========================== a user?

If you're using the upstream version of KUnit, nothing! Features will
continue to be pushed upstream when they're ready.

If you're using the 'kunit/alpha/master' branch, then you'll need to
switch to the new 'experimental-<version>' branches. Every so often,
we'll create a new 'experimental-<version>' branch, and you'll need to
rebase your code on top of that to keep receiving updated features and
bugfixes. While we intend to make this as painless as possible, there
may be breaking changes which require you to rework some of your

Once a feature exists in the upstream kernel, the version of that
feature in the 'experimental' branches should stabilise somewhat. a KUnit developer?

For most changes/bugfixes which apply cleanly to the upstream version,
you should directly submit these upstream as before. If it is feasible
and useful to backport them to the latest 'experimental' branch (e.g.,
they are bugfixes to code that the experimental branch already has),
please do so.

For more major features (or anything which depends on
yet-to-be-upstreamed code), you should develop them in the latest
'experimental' branch. For the most part, you can treat this the same
way as the old 'kunit/alpha/master' branch, but note that changes will
need to be rebased on top of a newer experimental branch periodically.
It may make sense to use this as an opportunity to tidy the code up
somewhat: by removing compatibility code for deprecated features, or
to merge 'fixup' commits. Once the code has been upstreamed, the next
rebase of the 'experimental' branch should only use the upstream
version: there should not be a separate 'experimental' version of
upstream code save for the development of a new 'experimental' feature
or bugfix.

For more details, see the 'Lifetime of a feature' section above.

In Conclusion

We'll continue to be working with upstream as before, but we intend to
rename and more explicitly specify the purpose of the
'kunit/alpha/master' branch as above. This will require a bit more
work from people using and developing against this 'experimental'
branch, but allows us to continue to use it for features such as the
mocking framework while limiting its divergence from upstream.

Do let us know if this sounds reasonable to you, if it will affect
your use case, or if you have any questions.

-- David & Brendan


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4847 bytes --]

^ permalink raw reply	[flat|nested] 8+ messages in thread

end of thread, other threads:[~2019-09-24  0:05 UTC | newest]

Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-09-17 18:41 kunit: what do we do with the 'kunit/alpha/master' branch? David Gow
2019-09-23 21:41 ` Brendan Higgins
2019-09-23 21:45   ` Siebert, David
2019-09-23 22:26     ` Brendan Higgins
2019-09-23 21:52   ` shuah
2019-09-23 23:00     ` David Gow
2019-09-23 23:19       ` shuah
2019-09-24  0:05         ` Brendan Higgins

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).