Difference between revisions of "Eclipse/Mars Retrospective"

Latest revision as of 11:56, 5 August 2015

This page captures notes for the Mars retrospective to be discussion at one of our Eclipse project weekly architecture call.

Suggested improvements for Neon

• Ensure all tests are executed by the Gerrit build trigger (Lars) -> Agreed, master bug 474328 created
• Review days for projects in which the focus is on reviewing community patches (Lars) -> To do for Lars: create a wiki for review days and send it around
• Use test day only for testing - no last minute commits (Dani) -> Agreed
• Analysis/investigation, design decision etc. should always be part of bug reports, so that others can follow (Dani) -> Agreed
• Have a dedicated team member (on rotation) look at test failures and make sure they get fixed for the next I-build at the latest (Dani)-> Agreed
• Have a dedicated team member (on rotation) look at the inbox (Dani) -> Agreed
• Component leads update the status wiki page in the interest of saving time for others during the Architecture call (Dani) -> Agreed
• Some components like resources or text have very little activity, maybe combine them with other projects to reduce the overheard? (Lars) -> No action necessary
• Problems reported by users via error reporter tool should be used as an input for release planning, e.g. as a dedicated plan item. (Szymon) -> Agreed that all components at least look at the top ten list
• Could we migrate our night test run to Tycho? AFAIK it is still based on Apache Ant and I'm not familiar with this setup (Lars) -> Agreed as a long term goal. There are two big problems: Tycho does not run the tests against the official build/bits and we might face issues with cross-repository tests

Retrospective

Platform UI / Lars

• What went well:
  • Lot of onboarding of new committers.
  • Code cleanup and updates of plug-ins to Java 7.
  • Gerrit build trigger includes more tests.
  • Critical functionality like customize perspective dialog were fixed.
  • E4 tooling were migrated to platform.

• What didn't:
  • Error reports from Marcel were not really used.
  • Platform tests not working well with GTK3 SWT made it harder to run them locally on Ubuntu 14.04.
  • Gerrit patch for JDT UI to show annotations in Outline view was not reviewed.

Releng

• What went well:

• What didn't:

Debug/ Ant

• What went well:
  • Inbox tracked, items delivered as per the plan
  • Build and Test results tracked and attended

• What didn't:
  • Last minute checkin does create confusion like Nashorm Debugging bug

JDT Core

• What went well:
  • New Owner for Formatter and a brand new and much improved formatter
  • Wider adoption of Gerrit among team members

• What didn't:
  • At times there was not enough communication with respect to review requests, thus resulting in bugs being left out of releases at last minute
  • Several new test failures in areas where no changes were made. Most due to short-comings of test frameworks and took significant amount of time early during the release.

JDT UI

• What went well:
  • Quality of N&N from old-timers has improved, less corrections to be made
    • New contributors need to improve adherence to the instructions
  • Build failures, test failures, attended to quickly, didn't have many long standing failures

• What didn't:
  • High DPI work didn't start soon enough, didn't have time to address all the issues detected when starting downstream adoption. Should address the remaining concerns earlier in this release.

PDE

• What went well:
  • Lars coming in with contributions and proactively which is helping PDE's cause.
  • Migration of the E4 project wizard to e4

What didn't:

• • Lot of things got moved out of plan. Needed to have planned moving out of release rather than moving iteration
  • Less expertise in PDE build causing issues in junit failures.

SWT

• What went well:
  • Delivered majority of the planned work items with high quality on schedule
  • Communicated well with UI team for accurately capturing requirements for the Transparent Color feature

• What didn't:
  • Need to provide Quicker review/response to patches from external community.
  • Need to plan for major changes including all feature work and important bugs to be done earlier in the release to avoid last minute regressions (at best by M6). Should try to finalize API contract with other stakeholders by M5 if possible to leave room for re-work if necessary.
  • Need to improve testing coverage for features + platforms to catch regressions and important bugs.
  • Should capture all API and design discussions only on bugzilla (we tried to use wiki for this but did not work out).
  • Should plan milestone work for 5 weeks and avoid pushing changes as much as possible during testing week.

Platform Workspace

• What went well:
  • The number of open bugs in Platform Workspace is under control (about the same compared to last year).
  • Improving traceability of several most important plugins across SDK as part of Mars plan item already pays off in much easier and faster troubleshooting.
  • Error reporter tool is very useful in gathering actual problems from users.

• What didn't:
  • There is lack of discussion on bugs and lack of reflection on fixes which cases many reverts.
  • Contributions, API proposals, bug fixes, etc. are sometimes merged without reflection and agreement about the best course of action.
  • Revert "Revert "Revert "Bug XXX - Bug summary""" is not very helpful commit comment.
  • A lot of fixes are pushed through Gerrit without prior verification in local workspace which causes poor quality of commits.
    • Green tests in Gerrit are helpful but not sufficient.

Dani

• What went well:
  • Delivered all milestones and RCs on time.
  • Builds went well and we had continuous and fast response to build issues.
  • Lots of achievements/deliverables despite a smaller team.
  • Team and cross-component interaction went well.
  • Architecture call was reasonable short and only used for mostly good discussions.

• What didn't:
  • Sometimes decisions made in bugs not explained in a way that non-team members can understand it.
  • Bugzilla inbox list is growing in some components, too many open items to be triaged/addressed.
  • Tests for some components sometimes fail for days without interaction.
  • Had an M5a again after a one year pause ==> use test day only for testing - no last minute commits.
  • Component leads should update status wiki page in the interest of saving time for others during the Architecture call.