CI is the killer, followed by patch tracking (for Samba)

Posted Sep 4, 2021 20:44 UTC (Sat) by abartlet (subscriber, #3928)
In reply to: Emacs discusses web-based development workflows by daenzer
Parent article: Emacs discusses web-based development workflows

For Samba, which I led on the transition from e-mail based workflow to GitLab it was CI that was sufficient carrot that we moved.

We already had pre-commit CI, a system we called autobuild, however if a dud patch was posted, the price paid for a CI failure was paid by the reviewer/commiter (particularly if the patch was not by a Samba Team Member.

We first encouraged pushing to a Shared development repository hooked up to enough runners to do a full Samba test, and then natrually folks started proposing merge requests. Pretty soon this became the standard workflow (even as the official Samba git repo remained elsewhere.

So, oddly, Samba doesn't actually merge any Samba merge requests, but applies them to a developer local machine and pushes them into the old process, but new developers don't see this, they just see a mostly new process that makes sense to them.

The big saving is that a developer, new or otherwise, who submits a not-yet-passing patch is given feedback by the computer, and not by a now-grumpy maintainer who spent time reviewing and pushing a patch that doesn't have a hope of passing CI.

I gave a Samba XP Talk on the transition.