[PATCH] reduce churn-induced conflicts in selftest/knownfail and flapping

Ralph Böhme slow at samba.org
Fri Jun 2 06:50:41 UTC 2017

On Fri, Jun 02, 2017 at 04:31:02PM +1200, Douglas Bagnall via samba-technical wrote:
> Broadly speaking, there are two kinds of entries in selftest/knownfail:
> the "WONTFIX-at-least-not-today" lines (some apparently dating from
> 2008), and the "proof of fix" lines that are added and removed at either
> end of a changeset to demonstrate a decrease in brokenness. Both kinds
> are often added to the end of the knownfail file, which is a natural
> thing to do given it has no intrinsic order. This leads to a lot of
> flickering activity that causes git conflicts when in semantic terms
> there is no conflict (it's an unordered set of lines).
> This patch makes the subunit framework look at all files in
> selftest/knownfail.d/ and treat them as if they were one. The intention
> is that long-lived entries can reside in selftest/knownfail.d/knownfail,
> and the fleeting ones can come and go in arbitrarily named files in that
> directory. For example (taken from the probable near future), supposing
> I was fixing a problem with the way we handle single value constraints
> in linked attributes, I could add tests and a knownfail file called
> selftest/knownfail.d/ldap-linked-attributes (I wish I could say the name
> was intentionally imperfect to show it doesn't really matter). Then a
> few patches latter when all is good, I would remove the file but leave
> the tests. We can all add fleeting knownfails with little risk of
> clashes.
> This retains the benefit of the tests-first, proof-of-previous-failure
> approach without the version control nuisance.

I generally like the idea, but can we keep the existing selftest/knownfail file
in place and not move it into the knownfail.d directory? That would make it work
like all the other kids like cron handle this.


More information about the samba-technical mailing list