Page MenuHomePhabricator

[spike] Lead paragraph hoisting and lazy images are disabled for users in the experiment on legacy parser in treatment but NOT in control
Closed, ResolvedPublic3 Estimated Story PointsSpike

Description

Explore options for resolving T416405: Lead paragraph hoisting and lazy images are disabled for users in the experiment on legacy parser which include

  • Making lead paragraph work on non-Parsoid views
  • Use Parsoid for all page views in experiment
  • Approach the experiment differently. Do not disable SectionTransform via $wgMFNamespacesWithoutCollapsibleSections and instead disable the associated JavaScript.

Outcome:

  • We have decided a strategy for ensuring the control and treatment group are consistent with the only difference being the floating table of contents in the treatment group.
  • (jon and brooke agreed that while it sorta works to hack up the transforms, it's potentially risky to non-experiment code. conservative approach is to go ahead and switch the control and transform groups to both use parsoid, which is a known quantity)
  • We can estimate T416405: Lead paragraph hoisting and lazy images are disabled for users in the experiment on legacy parser
  • (i'm putting it at a 2)

Event Timeline

Restricted Application changed the subtype of this task from "Task" to "Spike". · View Herald TranscriptWed, Feb 4, 5:54 PM
Restricted Application added a subscriber: Aklapper. · View Herald Transcript
Jdlrobson-WMF updated the task description. (Show Details)
egardner set the point value for this task to 3.Wed, Feb 4, 5:57 PM

Change #1237347 had a related patch set uploaded (by Bvibber; author: Bvibber):

[mediawiki/extensions/MobileFrontend@master] Proposed fix for lead para transform when section collapsing is disabled

https://gerrit.wikimedia.org/r/1237347

Change #1237347 abandoned by Bvibber:

[mediawiki/extensions/MobileFrontend@master] Fix for lead para transform without legacy section collapsing

Reason:

abandoning proposed patch as still too risky; we'll go with switching the control group to parsoid, which is a known quantity

https://gerrit.wikimedia.org/r/1237347

Jon and I agreed that while hacking the transforms seems to mostly work, it's a bit risky to non-experiment code and we're going to take the more conservative approach of switching both control and treatment groups to parsoid. This avoids any potential to break non-experiment renderings.

I'm assigning an initial 2-point estimate to the work ticket T416405