[RFC] Classifier state and feedback, interventions or similar

[mcbouslog]

Package
lib-classifier

Feature or Issue Description
Per #549 to show the feedback modal we use MST middleware to catch the action to advance to the next subject, which we then abort. The feedback modal is shown, and on close we reconnect to the subjects store and advance to the next subject. We're not wild about this implementation.

Why we're not wild about the current implementation:

• per this PR comment: One of my dislikes about this implementation is that it's recursive. Calling subjects.advance() at the end of feedback will trigger this handler to run again. We've hacked around that by checking to see if there's a modal open. It's possible to accidentally put the code into a never-ending loop where closing the feedback modal opens it again and you never get to the next subject.
• paraphrased from other PR comments: It'd be nice if we could use a pattern where feedback interrupts the action in progress, then resumes it. At the moment, it just blocks the subject queue from advancing and we have to remember to advance it again somewhere else in the code. At the moment, the feedback modal will always advance the subject queue on close, no matter where it's used in the classifier.
• per Slack: Ideally, this should pause the action, show feedback (or whichever intervention is happening) then continue the paused action. That will become more important when we implement interventions in the new classifier.

Let's use this RFC to discuss potential refactors

• an overarching classifier state store might be useful, to pause/resume the classifier flow, which feedback, interventions or other features could hook into that the rest of the classification process would react to.

[eatyourgreens]

The feedback modal isn’t hardwired to advance the subject queue, as of today’s changes.

[eatyourgreens]

A couple of issues with the current implementation:

• Recursion: we listen for subjects.advance, then show a modal which fires subject.advance on close, which checks to show feedback etc. etc. Change the order of the following two lines of code and the feedback modal will always re-open on hideFeedback.

  front-end-monorepo/packages/lib-classifier/src/store/FeedbackStore.js

  Lines 107 to 110 in 405f797

  	function hideFeedback () {
  	self.onHide()
  	self.showModal = false
  	}

• Going to Talk bypasses feedback. Done & Talk (Classifier - add "Done and Talk" button #516) redirects the window to a new URL before feedback can open. In PFE, we fixed this by hijacking Done & Talk and redirecting after feedback is shown. Store workflow task history in Classifier state. Panoptes-Front-End#4450 (comment)

[eatyourgreens]

Looking back at PFE, the Done button triggers showing feedback, which then triggers completeClassification when closed. The original click event is passed through to completeClassification which then branches based on whether the clicked element was a link (Done & Talk) or a button (Done.)

[srallen]

That's pretty ugly and another reason to remove the means to go to Talk from the task navigation.

[eatyourgreens]

For TESS, I'm thinking Done/Done & Talk triggers showFeedback, which either shows feedback or passes through to finishing the classification. Done & Talk would need to store something that tells us whether the next action after completing the classification is opening Talk or advancing to the next subject.

[eatyourgreens]

And we're going to have to deal with this once #516 merges.

[eatyourgreens]

I've opened the Done & Talk issue as #598.