Contributors/FY2016 Retrospectives/VE

Next Steps

• Action items for team members
  • JF to organise ES/TC/JF/NQ meeting to discuss VE metrics and metric goals
  • JF/LA: Need to bring more people on-board with new Wikitext Editor work as we progress.
  • JF to think about how to work with Research & Data and come back to the team.
  • Ed: Use the new browser test framework to address shortage of test coverage out of DM.
  • Ed: Explore adding (browser tests) to CI, blocking merge?
• Action items for department
  • JF: Coordinate VE and Language team to coordinate dependencies between VE in CX  and getting Content Translation out of Beta.  Decide who lead implementation work and by when.
• Decisions
  • JF to continue not deploying in the face of opposition.
  • Keep high priority for completing Wikitext editor (i.e., reject other requests, especially requests for more features in Wikitext editor, including from other departments)
  • Slow release process for new Wikitext editor will be needed.
• Follow-up at next retro (6 mo?)

Ground Rules

• Разговор этот час является конфиденциальной в рамках группы
  • A version of written notes, after a review pass and voting, to be published to Editing and to general public
• Purpose of Retrospective
• Scope
  • What did we accomplish last year?
  • What was great about the way we worked, or the org, or anything else job-related?
  • What was not great about the way we worked, or the org, or anything else job-related?
• Review key events and results over the time period
• Brainstorm (but do not discuss in depth) Things To Keep/Preserve and Things to Change
• Add a +1 to any (Action is important | I want to discuss | I want to escalate) items by including your initials
• Discuss top-voted items

Heavily upvoted items will be percolated up for larger department discussion

What happened (5 mins)

• VE roadmap was aggressively, repeatedly, and ad-hoc-ly revised
• Hasty prep work for demo at Wikimania 2015
  • Mobile editor etc.; not very convincing as a plan, for developers or editors alike
• Integration work
  • Implemented switching from wikitext to VE
  • Single edit tab
  • ProofreadPage work (for Wikisources)
  • Maps work (albeit unused due to templates)
  • Link suggestion feature
• Deployments
  • Launched lots of VE to lots of populations on lots of wikis
    • But not all
    • Resistance on a few key deploys
      • Second enwiki consultation on anon rollout was problematic
      • NL rejected/couldn’t make a decision
      • A few people blocked (the rollouts necessary to do testing leading to) VE for anonymous on en.wp.
  • Assisted with Wikisource deployment
  • Assisted with Wikivoyage deployment
• Feature work
  • Started on companion wikitext editor
  • David C’s huuuge completed work on IMEs, typing, etc.
  • Migrated to JS for automated screenshots (∴ new framework for browsers tests)
  • Table editing improvements
  • Gallery editing
  • Formula editing
  • Wikitext sequences (type {{ to trigger a template)
  • Link inspector
  • Syntax highlighting in <syntax> tool
• Team Changes
  • Added one new Senior Software Dev (David L.)
  • Added one new Manager (LA)
• Lots of SLU meetings with other teams
• Imposed our CSS standards on others via stylelint as part of our wider CI standards work

Teams/groups with whom we worked:

• Services, Parsing, Language, Collaboration, UI S13n, Front End Standards Group, Release Engineering, Community Liaisons, Product Analytics

Things:

• Gerrit, Phabricator, Phlogiston, Jenkins/CI, scap, …

Travel, meetings, face time:

• Two Wikimanias (Mexico City and Esino Lario); hackathon (Jerusalem), offsite (London June 2016), All-hands and Dev Summit in SF in January

Identify things that went well and should be preserved (10 mins)

• Deployments
  • (DL) (JA) during FY 2015–16, lower levels of complaint compared to the past
• Good progress on features
  • (ES) (TC) New wikitext editor
  • (ES) Gallery editing
  • Formula editing
  • Table editing
  • Link inspector
• Working with other teams
  • (ES) Collaboration with Zelijko on JS browser tests/screenshots
• Working methods
  • VE bug triage rarely got more than a month of backlog
  • Ed: “no complaints” about Phabricator / gerrit
  • Distributed team working well together; time zones continuing to be a tolerable and surmountable difference (special thanks to David C on his insane hours)
  • Exported and expounded stylelint, VE leading the way in s13n (standardisation) again :-)

Identify things that could be improved and/or changed (10 mins)

• (ES*) (DL) (TC) (AM) (JA) (JA) (LA) Aaron's quantitative study was inconclusive as to whether VE (compared to Wikitext editor) led to a decreased edit abandonment rate.
  • Not confident we correctly tested this properly given the range of data saying different things to the conclusion – e.g. discarding all cases where edit was abandoned immediately - ES
  • Could also learn lessons about what are the right questions to ask
• (ES) (DC) (DL) Testing lacking outside of the DM, which has good coverage (cf. new browser test framework)
• (AM) Maps work was extensive, but not usable because of templates?
• There was some occasional resistance by a few community members to VE rollouts
• Communication on IME issues was pretty patchy with some communities, which has made determining deployment readiness a bit of a stab in the dark at times
  • e.g. limited contact/feedback between team and Japanese-speaking editors
• (DC) (ES) Significant drift of responsibilities into certain Language team-related products (e.g. we rewrote jQuery.ime’s support for ContentEditable)
  • I think it would have been difficult to do some of the work without highly specific expertise, e.g. about browser ContentEditable support. We had a meeting in January that was technically enlightening to David C and me - ES
  • I agree handing stuff off is not always the best solution - DC
• (AM) (RY) VE usually carried a few weeks' worth of untriaged bugs, and only got that down to zero a few times in the year
  • possibly increase triage frequency?
• (AM) The deep backlog/icebox of 1000+ items not very useful
  • (re: the general backlog, not the categorised and prioritised Tranche backlogs)
• (DL) Jenkins CI was somewhat flaky throughout
• (AM) Git submodules can be a pain (especially in Gerrit)
• Phabricator and Gerrit somewhat redundant, with Differential
  • more of a Wikimedia tech-wide issue though
• (TC) (DL) Working relationship with Research & Data not clear
  • Link suggestion feature collaboration with Research did not go so well
    • Didn't see any concrete product emerge despite providing many hours of assistance.
    • Initial request for help was unclear how unable they were to implement the feature themselves, meaning we were expected to move from "give advice" to "do work".
  • Is R&D's model following up on VE ideas, producing their own ideas, or handing finished technology over for implementation?
• Vagrant can be flaky (James F, David L. & Thalia all reported issues)

Discuss top items, including Actions (15 mins)

Yays

• (DL) (JA) during FY 2015–16, lower levels of complaint compared to the past
  • ACTION/KEEP: JF stopping deployments in the face of motivated, consensual opposition.
• (ES) (TC) New wikitext editor
  • Ed: A fair amount still to do.  At some point, I have to get more people involved.
  • ACTION:
    • Don't screw it up? (Insulate the team from priority interruptions.  Ed: And from external product meddling … e.g. from other departments with specific feature requests)
    • Need to bring more people on-board with the work as we progress.
    • Slow release process will be needed.

Nays

• (ES*) (DL) (TC) (AM) (JA) (JA) (LA) Aaron's quantitative study was inconclusive as to whether VE (compared to Wikitext editor) led to a decreased edit abandonment rate.
  • Not confident we correctly tested this properly – e.g. editor bounce rate - ES
  • Could also learn lessons about what are the right questions to ask
  • DC: Good to have more solid evidence, but scrabbling for it and not finding it is bad
  • DL: People will ask, we should have the stats available
  • It's qualitatively obvious it's way better, loads of anecdotes to that effect; more, it's concerning about our lack of ability to be data-driven.
  • Thalia: (was that) the question to ask?
  • ACTION: JF to organise ES/TC/JF/NQ meeting to discuss and move on.
• (ES) (DC) Testing lacking outside of the DM, which has good coverage (cf new browser test framework)
  • ACTION: Use the new browser test framework to address this. Explore adding to CI blocking merge? Maybe too slow.
• (DC) (ES) Significant drift of responsibilities into certain Language team-related products (e.g. we rewrote jQuery.ime’s support for ContentEditable)
  • Ed: for Content Translation, it’s their remit since it’s their product, but affects everyone.  Definitely need VE integration before leaving beta.  If we and they both put it off, there’s potential for future live breakage.
  • Joel: make the issue clearer to more people via better definition of quality/done/whatever and earlier definition of timetable
  • DC: Ensure boundary of responsibility between teams is very clear, so any potential difficulties have a clear owner
  • ACTION: VE integration a blocker for Content Translation leaving beta to avoid a crisis. How to help Language team implement this efficiently?
• (TC) (DL) Working relationship with Research & Data
  • Link suggestion feature collaboration with Research did not deliver as initially envisaged
    • Didn't see any concrete product emerge despite providing many hours of assistance.
    • Initial request for help was unclear how much prior assistance they needed to familiarise with the VE code
  • Is R&D's model following up on VE ideas, producing their own ideas, or handing finished technology over for implementation?
  • Some manager-level discussion is in progress.
  • ACTION: In future, … JF to think about this and come back to the team.