Engineering Steering Committee Meeting, September 3, 2019
David Pursehouse, Patrick Hiesel, Luca Milanesio, Ben Rohlfs
Online, September 3, 12:30 - 13:30 CEST
The next meeting will be held on September 17, 12:30 CEST.
Gerrit News Page
There were no suggestions for news items. The draft for the next issue already exists, and is scheduled for publishing on September 27th. Anyone can send suggested news items for review before then.
Hackathon and User Summit in Gothenburg
The recent hackathon and summit were a great success. During the hackathon many bugs were fixed, and 3 new releases have since been made.
The talks at the summit were all recorded and will be published once they have been vetted by Volvo’s legal team.
Patrick’s remote presentation about Gerrit performance was well received, and he will present it in-person at the next summit in November.
Luca will prepare a more detailed article to be published on the project news page soon.
REST API for retrieving Git trees
The authors of the rejected design are planning to attend the upcoming user summit in Sunnyvale so there may be some opportunity to revisit the proposal and come to a compromise.
Gerrit versioning and criteria for accepting fixes
Luca’s updates to the versioning rules were reviewed, accepted, and submitted.
Design document for reverting multiple changes
The design for reverting multiple changes was reviewed and approved.
Plans for 3.1 release and EOL of 2.15
David suggested that we should start thinking about the release schedule for Gerrit 3.1 and therefore also bringing 2.15 to EOL.
Everyone agreed that the upcoming hackathon in November would be a good time to release 3.1. This date also aligns with the planned schedule for completing the migration to Polymer2.
We will cut the stable-3.1 branch some weeks before (mid October; exact date TBC) so that we have time to make release candidates and stabilise it.
We need to make sure the release notes get updated in a more timely manner than 3.0 so that we don’t have to do them at the last minute again.
The planned EOL for 2.15 should be announced early so that nobody is caught by surprise. We will make a separate news announcement, and update the homepage with more explicit details of support levels for recent releases.
Proposal for making more frequent patch releases
Luca proposed to make more frequent patch releases (for example every two weeks) to avoid that we have to make large release note updates and that the releases include too many fixes (i.e. from 3.0.1 to 3.0.2).
David and Patrick disagreed. Making releases is time consuming and it doesn’t always make sense to release on a fixed schedule, for example if there are only a couple of changes. It’s better to wait and make a release only when there are a reasonable number of fixes. Release note updates are not really a problem; 3.0.2 was a rather large release but most of the release notes could be copied from 2.15.x and 2.16.x.
End-to-end testing can be automated to reduce the time spent in making a release (see the next section). Also, Edwin is working on a way to automate creation of the release notes.
Improved end-to-end testing
Most of the release process is now automated, but there is no end-to-end testing. Work on this is in progress at GerritForge - see the example change posted for review.
We will follow up on this in the next meeting.
Polymer 2 and customization/themes
Ben talked about the current status of Polymer 2 migration in terms of what level of support should be expected for customization.
In Polymer 2 the shadow DOM makes it more difficult than before for plugins to override styles and behaviors, so we need to decide which parts we want to allow to be customizable, for example the header.
The frontent team would prefer to limit the amount of customization, but are aware that there are users that might want more. For example Wikimedia has some very specific styling, and GerritForge has customizations on the search box.
In some cases the frontend team might be willing to make global changes based on customer customizations (for example the GerritForge search box) but would need to run them by the UX team.
Consensus among ESC members is that we should apply good judgement on what should be customizable but don’t necessarily need to support every use case.
Migration of CI to Zuul
Monty Taylor (Redhat) has offered to help with the adoption of Zuul for Gerrit’s CI. They are also keen to help with support for the checks plugin in Zuul.
Everyone agrees that this would be a good move.
Migration of our existing CI jobs should not be too difficult since we already define them in a yaml format similar to the one that Zuul expects.
There is no timeline yet. Luca will follow up.
We had planned to include JGit upgrades in 2.16.11 and 3.0.2 but at the last minute a regression was found and the updates were reverted.
The root cause of the regression was found and fixed by Han-Wen Nienhuys earlier this week. Matthias Sohn is making new JGit releases, and we will include the upgrades for the next 2.16 and 3.0 patch releases.
Patrick reminded that we were planning to change Gerrit to build JGit from source, and asked what the status is. This is still planned, but has stalled during the ongoing work to stablize JGit.