Engineering Steering Committee Meeting, October 1, 2019
David Pursehouse, Alice Kober-Sotzek, Luca Milanesio, Ben Rohlfs, Patrick Hiesel
Online, October 1, 12:30 - 13:30 CEST
The next meeting will be held on October 15, 12:30 CEST.
Gerrit News Page
The latest issue of the project news was published on September 27.
A draft has been created for the next issue which is to be published on November 29.
There were no news items proposed during this meeting. Members of the community may propose items by adding a change on the draft post.
Renaming of ‘reviewdb’ package to something else
Since we migrated from ReviewDb to NoteDb it doesn’t make sense to have
classes in the
reviewdb package any more. David Ostrovsky proposed a
series of change
to rename the packge to
We all agreed that
reviewdb is no longer an appropriate name for the
package and that it should be renamed. Ben also pointed out that it can
cause confusion for new developers.
Alice mentioned that we should also consider that some of the classes may
not really need to be kept in the
entities package and can be moved to
other, more appropriate packages.
We concluded that we should do this in two steps. First, rename the package, which can be included in the upcoming 3.1 release. And second, move classes to more appropriate packages. The latter should be deferred until after 3.1, so it goes into 3.2.
Make the reviewers plugin a core plugin
Issue 40010542 proposes to make the reviewers plugin a core plugin.
We discussed this without reaching any concrete conclusion. We will defer to the newly formed Plugin Working Group to define criteria for promoting a plugin to core, and then revisit this issue.
Make checks a core feature instead of a plugin
Issue 40011251 proposes to make the checks plugin a core plugin. In that issue David Ostrovsky suggested the alternative of making checks a core feature rather than a plugin, with the motivation being that a number of current design limitations in checks plugins are coming from its nature being a plugin and not a pure gerrit core feature.
Several points were raised during the discussion:
Checks is a useful feature that should be core (either actually in core or as a plugin).
There are some user bases who have their own custom solutions and hence might not be interested in checks.
If it were a core feature it could be switched off with a config setting, but on the other hand this is the reason we have plugins: if someone doesn’t want the feature they don’t install it.
Long term it might be better as a core feature, i.e. at the point where gerrit-review is using it rather than labels.
Does the community want this as a core feature?
We didn’t reach a concrete conclusion as to whether checks should be a core feature, nor whether checks should be promoted to a core plugin. We will revisit the discussion when the integration with gerrit-review is more complete.
If any community members have opinions on whether checks should be a core feature, we ask them to add comments on issue 40011251.
Follow up on blocking issues for 3.1
Per the release plan for 3.1 we expect to be cutting the stable-3.1 branch and making the first release candidate in a couple of weeks. We had a look at the list of issues that are labelled Blocking-3.1.
Ben confirmed that all the Polymer 2 related issues are expected to be fixed in time.
Patrick will review the changes related to issue 40009176 (“Replace deprecated numeric types with new dimensional numeric types”).
For issue 40011264 we decided that reducing the default SSH idle timeout will be a breaking change for many users, and we prefer to only submit that change on master after stable-3.1 has been cut.
Support for deletion of groups
David implemented the framework of the endpoint, but the actual deletion of the group is not done. The reason for adding it in the ESC agenda was to confirm whether or not this feature is actually still wanted.
Alice confirmed that the feature is OK, but we need to be careful that all references to groups are handled when they are deleted. She will be involved in the review, but only when content is added (i.e. when the actual deletion is implemented).
Support for role-based access controls
Support for role-based access controls was started by Sven Selberg and Gustaf Lundh during one of the previous hackathons, but the changes got stale and were abandoned. See change 104743 and the acl-refs-for-removal topic.
This was added into the ESC agenda for discussion on whether or not this feature is still wanted and can be revived.
We all agreed that role-based access controls is a great idea, and would be very useful to reduce confusion around Gerrit’s permissions system, however it is something that should be done carefully and would need to be designed properly.
Luca suggested that this would be a good topic for a hackathon.
We are almost ready to release 3.1 so there’s no point including it in the roadmap. We will just write the release notes.
We will define the roadmap for 3.2, and continue to add features there until there is anything that we know will not make it into 3.2, then start the roadmap for 3.3.
For any more long term (than 3.3) features we will keep a backlog which will be cleaned regularly.
We will set up a dedicated meeting to discuss this in more detail.
Review of issues on the ESC component
We briefly went over the issues that have been added to the ESC component on the issue tracker.
Urgent or important issues will typically already have been added to the agenda for the meeting, and we did cover a couple of them today. Issue 40011246 was not on the agenda and we discussed that (see next section).
We added this activity as a standing item on the ESC meeting agenda.
Browsing repository should be a first class citizen
Issue 40011246 requests that finding and browsing code repositories should be provided by the Gerrit UI.
Alice commented that ‘first class’ means there is a code browser in Gerrit, and it’s difficult to get that right because browsing is a product on its own. Maybe we can improve the integration while keeping it separate. Ben also agreed that since Gerrit is a code review tool, browsing should be separate.
Patrick also raised the point that some people might expect that a code browser provides cross reference and search capabilities.
Everyone agreed that code browsing should be kept separate, but there is the possibility to improve the integration with plugins such as gitiles.