fix: don't re-observe a detached element after destroy()

[AndreasOhlsson]

Fixes #246.

What happens

A position update queued with lowPriority() (requestIdleCallback) before destroy() can still run after it, for example when the component unmounts during a busy route change. updatePos(el) then calls observePosition(el), which creates a new IntersectionObserver rooted at document.documentElement on the now-detached element. Nothing disconnects that observer, and it keeps the detached subtree alive for the rest of the session.

The fix

observePosition() returns early for an element that is no longer in the document, after disconnecting any old observer. Every path that (re)creates an observer goes through this function, so one guard covers all of them:

• idle-callback polls
• updateAllPos
• observer callbacks
• animation finished handlers

Observing a disconnected element measures nothing, so no animation behaviour changes for connected elements.

Evidence

• Repro: in Chromium, a parent with 10 children; unmount it during a ~600 ms busy main-thread window; force GC.
  • 0.10.0: the detached subtree survived in 11 of 12 trials. Every leaked observer was created 200–600 ms after an idle callback that was queued before destroy() and ran after it.
  • With this guard: 0 of 12 leaked, even though idle callbacks still ran after destroy().
• In a production React app (useAutoAnimate on a timeline that remounts on navigation), over 60 navigations:
  • 0.10.0: about +120 DOM nodes per navigation.
  • With the guard (applied as a patch): DOM nodes and event listeners stay flat.

Relation to #242

#242 (which fixes #180) tracks the poll start timers and stops polling in cleanup. That closes the timer-side leaks, but an idle callback that is already queued still reaches observePosition() after destroy(). The two changes complement each other and don't conflict.

I didn't add a test. The race needs a busy main thread around unmount, which the existing coarse memory.spec.ts doesn't cover. Happy to add a targeted e2e test modelled on it if you'd like one.

Diagnosed with heap snapshots and a Playwright repro.

[vercel]

@AndreasOhlsson is attempting to deploy a commit to the Formkit Team on Vercel.

A member of the Team first needs to authorize it.