ContextQMD
Back to Motion

Plan issue-2791: Reproduce-or-close NaN when spring-animating polygon points (defer to plan 032)

plans/issues/issue-2791.md

12.41.05.9 KB
Original Source

Plan issue-2791: Reproduce-or-close NaN when spring-animating polygon points (defer to plan 032)

Executor instructions: This issue already has a full investigation plan: plans/032-spring-complex-value-nan-investigation.md. Execute THAT plan; this file only records the triage classification, planner findings that 032 predates, and the gated close path. Update the status row for this plan in plans/issues/README.md (NOT plans/README.md) when done.

Drift check (run first):

  1. gh api repos/motiondivision/motion/issues/2791 --jq .state → expect "open". If closed, mark DONE and stop.
  2. git diff --stat 42bfbe3ed..HEAD -- packages/motion-dom/src/utils/mix packages/motion-dom/src/animation/JSAnimation.ts packages/motion-dom/src/value/index.ts — on any change, re-verify the excerpts below before acting; mismatch = STOP.

Status

  • Classification: NEEDS-REPRO
  • Priority: P2
  • Effort: M
  • Risk: LOW (investigation-gated; repo policy: no repro → no fix)
  • Depends on: plans/032-spring-complex-value-nan-investigation.md (execution vehicle)
  • Category: bug (unconfirmed mechanism)
  • Planned at: commit 42bfbe3ed, 2026-06-11
  • Issue: https://github.com/motiondivision/motion/issues/2791

Why this matters / verdict

Spring-animating a <polygon>'s points reportedly writes NaN point lists (Sep 2024, framer-motion 11.x). The mechanism could NOT be pinned statically at 42bfbe3ed, and the reproduction is unreachable, so per repo policy this is NEEDS-REPRO: run plan 032's reproduce-first ladder; if no layer fails, close gated.

Planner findings (new since plan 032 was written — feed into its Steps 1–3)

  • Repro fetch already attempted and failed: https://codesandbox.io/api/v1/sandboxes/5mwd29 → HTTP 403; https://codesandbox.io/p/sandbox/pensive-khayyam-5mwd29 → blank page (JS app). The issue body contains no code, only the link. Plan 032 Step 1's "STOP and ask" branch is the likely entry state — ask the operator/reporter for the polygon JSX + exact points strings.
  • The leading "mismatched token count + spring overshoot" hypothesis was tested and does NOT produce NaN. Against the built package at 42bfbe3ed:
    • mixComplex("0,0 10,10 20,20", "0,0 100,100")(1.2) → "0,0 118,118" (origin longer: extrapolates finitely; matchOrder in packages/motion-dom/src/utils/mix/complex.ts:79-98 pads with ?? 0).
    • mixComplex("0,0 10,10", "0,0 100,100 200,200")(0.5) → "0,0 100,100 200,200" (origin shorter: canInterpolate fails at complex.ts:107-110, falls to mixImmediate, which is (p) => (p > 0 ? b : a) — packages/motion-dom/src/utils/mix/immediate.ts:1-3 — an instant flip plus the "too different to mix" warning, never NaN).
  • New candidate to probe while reproducing — bogus string velocity: isFloat (packages/motion-dom/src/value/index.ts:43-45) is !isNaN(parseFloat(value)), and parseFloat("100,100 150,25 …") → 100, so canTrackVelocity is TRUE for points strings. getVelocity() (value/index.ts:406-428) then computes a finite-but-meaningless velocity from the first coordinate, which is passed into the [0,100] progress spring created in JSAnimation.initAnimation (packages/motion-dom/src/animation/JSAnimation.ts:131-141). That explains spring-specific violent overshoot, but is still finite — NOT proven to be the NaN source. Check what velocity reaches the spring in the live repro.
  • The reporter's separate "expected number, 'undefined'" (non-spring) suggests the string "undefined" reaching the DOM attribute — also unexplained; capture it in the repro too.

Steps

Step 1: Execute plan 032

Run plans/032-spring-complex-value-nan-investigation.md end-to-end with the findings above. Concretely: skip re-testing the two mixer cases listed above (already negative); start its Step 2 repro matrix at the JSAnimation/Cypress layers and add a velocity probe (onUpdate + log motionValue.getVelocity() before the animation starts).

Verify: plan 032 exits with either (a) a failing test + diagnosed root cause + fix, or (b) its documented "needs repro" exit after honest attempts at all layers.

Step 2 (only on exit (b)): gated close

ONLY after the row for this plan in plans/issues/README.md is marked APPROVED:

gh api repos/motiondivision/motion/issues/2791/comments -f body="We attempted to reproduce this against current main and couldn't trigger NaN output from the complex-value mixer, including with mismatched point counts and spring overshoot (progress > 1), and the original CodeSandbox is no longer accessible. If you can still reproduce on motion@12, please share the polygon's exact points keyframes and transition so we can reopen — happy to dig in with a working repro."
gh api -X PATCH repos/motiondivision/motion/issues/2791 -f state=closed -f state_reason=not_planned

Verify: gh api repos/motiondivision/motion/issues/2791 --jq .state → "closed". (If plan 032 instead lands a fix, close with -f state_reason=completed after merge — same APPROVED gate. gh pr edit is broken on this repo; use gh api -X PATCH for PR edits.)

Done criteria

  • Plan 032 executed to one of its two exits; its row AND this plan's row updated in the respective READMEs (032 → plans/README.md, this → plans/issues/README.md)
  • No fix or test landed without a test that failed on the bug first (repo policy)
  • Any close/comment action happened only under the APPROVED gate

STOP conditions

  • Plan 032's excerpts no longer match the live code (drift).
  • You are tempted to land "defensive" NaN guards or happy-path tests without a failing repro — policy forbids it; take exit (b) instead.
  • The sandbox becomes fetchable and reveals 3+ keyframes with spring: that hits the prod-silent two-keyframe truncation (JSAnimation.ts:120-129) — report, don't improvise.