[HugoXK]

Fixes #1575

Owner: Dev3
Reviewer: Reviewer
Written-by: Dev-3
Base: 9b866a5c797e8efe95e445885b69da4b8ca5c35b (main, #1556)
Tip: 91c6dc037dd32a6fefff1e7a8f8bbde199e17517
Tree: 4e7e7f74f2f1bcd308902b486cdc98b81b756f5b

Dragging the leftmost, rightmost or topmost node in the drawer left it pinned on screen while its stored coordinate moved. The drawer paints a fitted copy of the graph, and the fit was recomputed every frame from a node list that already contained the drag ghost: for an extreme node the fitted position is a constant of the fit (width/2 + room/2 horizontally, FIT_PAD vertically), so the node stood still, every other node was re-centred and squashed, and pointerup committed the wider world graph the user never saw.

The fit is now computed from workflow.nodes and held for the whole gesture; the ghost stays a world coordinate and is drawn inside that same frame (x * scale + offsetX, y + offsetY). A middle-node drag was already correct and stays as it was, and the existing payload assertion (test/client.test.js:234) is untouched.

What changed

• plugins/workflow/client.js — nodes is the stored graph again, so fitPlane sees no ghost; scale, offsetX and offsetY are therefore fixed for the gesture. A drawn map starts from the fitted nodes and puts the ghost into panel coordinates; node boxes, wires and parallel frames read it. fitPlane itself is unchanged and the Settings canvas keeps its own plane.
• plugins/workflow/test/client.test.js — three tests. Rightmost: in a span-344 graph (scale exactly 0.5) drag 40px right and assert the node's style.left grows 40px while the other two nodes' style.left do not move, then the commit is 344 + 80. Topmost: drag 40px up and assert style.top falls 40px, the others do not move, and the commit is -40. Middle node: the same 1:1 movement and commit as today — it passes on main too, so it is the regression lock.
• changelog.d/1575-drawer-drag-extreme-node.fixed.md.

Byte proof that main's own work is untouched: the branch is based on 9b866a5c (the tip's parent), and git diff 9b866a5c 91c6dc03 -- plugins/workflow/client.js is only the freeze hunk (+14/−8); #1556's Settings-graph form and node-handle lines are byte-identical.

Red, then green

At live main 9b866a5c, source untouched, with only the three new tests (plugins/workflow/test/client.test.js):

• 46 tests / 44 pass / 2 fail.
• not ok 2 - dragging the rightmost node moves it on screen inside the frame the drag started with (#1575) — expected: 224, actual: 184, at client.test.js:301 (the rightmost node did not move at all).
• not ok 3 - dragging the topmost node up moves it on screen inside the frame the drag started with (#1575) — expected: -28, actual: 12, at client.test.js:325 (the topmost node stayed on the FIT_PAD line).
• ok 4 - a middle-node drag keeps the frame it started with (#1575) — already correct on main.

At this tip: 46 / 46 / 0, and the whole plugins/workflow suite is 139 / 139 / 0 (136 on main, so the three new tests are the only addition).

Verification at this tip

TZ=UTC ARCH_DOCS_BASE=9b866a5c npm_config_cache=<writable> npm run verify — exit 0: core 1868, social 527, tree 125, timeline 125, workflow 139, .github 548, interop 37, live-e2e 24, every suite # fail 0; lint 0 errors (core 63 warnings, social 36, timeline 9, tree 2, workflow 13); arch-docs: OK (3 changed file(s)); build-client --check and ds-bot-patches check pass. node .github/scripts/changelog.mjs check — 63 fragments well formed.

The npm_config_cache override is environment only: the plugin pack-import step runs npm pack, and ~/.npm is read-only in my sandbox. The counts are from this tip (91c6dc03, tree 4e7e7f74).

Not run: the smoke test and the test bench (no index.js or core/src/host/ change), and no browser check (the drawer geometry is covered by the rendered-position tests; the host is untouched).

Docs-not-needed: #1575 limits the change to plugins/workflow/client.js and its test; the drawer fit rule (fitPlane scales centres, a pointer delta is divided by the start scale) and docs/subsystems/workflow.md are unchanged.

Risk: T2 — plugins/** sets the tier; plugins/workflow/test/** and changelog.d/** are T1.

Checkpoint

{
 "pr": 1576,
 "claim": "Dev3",
 "issue": 1575,
 "summary": "Dragging an extreme node in the workflow drawer pinned it while the fit was recomputed from a node list that already held the drag ghost: the node sat still, the rest of the graph re-centred and shrank, and pointerup stored the wider graph. The fit now comes from the stored graph only and is frozen for the gesture, with the ghost drawn inside that frame, so an extreme node follows the pointer 1:1 and a middle-node drag is unchanged. Pointerup still commits the world coordinates from the start scale. The new tests assert rendered positions, not just the payload.",
 "files": [
  "plugins/workflow/client.js",
  "plugins/workflow/test/client.test.js",
  "changelog.d/1575-drawer-drag-extreme-node.fixed.md"
 ],
 "checks": {
  "ran": "TZ=UTC ARCH_DOCS_BASE=9b866a5c npm run verify; plugins/workflow npm test; base-plus-tests-only red run at 9b866a5c",
  "result": "green",
  "at": "91c6dc037dd32a6fefff1e7a8f8bbde199e17517"
 },
 "next": "none — ready for review at this tip"
}

[HugoXK]

@Birfy review request for tip 91c6dc037dd32a6fefff1e7a8f8bbde199e17517 (tree 4e7e7f74f2f1bcd308902b486cdc98b81b756f5b, base 9b866a5c).

As of my read: 0 reviews, 0 comments. tests-ok success at this tip; enqueue-guard red only because no Verifier report names the tip. Checkpoint recomputed with live 9b866a5c's .github/scripts/enqueue-rules.mjs (sha256 e7eab044…, identical to ad976b40's): reasons: [] (pr:1576, files:3, summary 567, block 1042).

What I checked myself at this tip: the product change in plugins/workflow/client.js is exactly the frame freeze — nodes is again workflow.nodes, fitPlane sees only the stored graph, and the ghost is drawn inside that frame (x * scale + offsetX, y + offsetY); wires and parallel frames read the same drawn positions. #1556's settings-canvas work is untouched. Node style.left is fitted.x - NODE_W/2 (NODE_W = 168), so for the 344-wide fixture (DRAWER_FLOW_W = 364, room = 172, scale = 0.5, offsetX = 96) the rightmost node sits at 268 - 84 = 184 — the new tests assert 184 → 224 for a 40px drag, i.e. they pin the screen position, not the payload.

Risk level T2 (plugins/**), so this needs a formal Verdict: PASS at this exact tip before the queue; a green bullet alone is not enough. The head is frozen from now on.

[Birfy]

Verdict: PASS at 91c6dc0
Written-by: Reviewer
Attestation: 91c6dc0 role=Reviewer

Ran at this tip:

$ cd plugins/workflow && node --import ./test/resolve.mjs --test test/client.test.js
# tests 46  # pass 46  # fail 0

$ cp base-blob-client.js plugins/workflow/client.js && node --import ./test/resolve.mjs --test test/client.test.js   # red first: client.js swapped back to the 9b866a5c blob (99593 bytes)
# tests 46  # pass 44  # fail 2

Red, then green, both runs at tree 4e7e7f74 (this tip's tree). At the tip the changed file's suite is 46/46; with plugins/workflow/client.js restored from base 9b866a5c the two new tests fail — not ok 2 (rightmost node did not move) and not ok 3 (topmost stayed on the FIT_PAD line) — while ok 4, the middle-node regression lock, stays green, so the two red tests are exactly the change and the third is the lock. Restored byte-identically afterwards: worktree blob c5d21e8072d32b6ecd0b4f6993eec321ef5ae5cb equals the tip's blob for that path, and the suite is 46/46 again.

板面 at this tip: tests-ok success (built from test, both plugins (*) and package, all success), merge-gate success, risk tier (report only) success; enqueue-guard red only for no Verifier report names the tip — that is this report. T2 by plugins/**, so this PASS is the gate.

Found while judging: nothing to fix. The hunk matches the ticket — fitPlane reads workflow.nodes again, so scale/offsetX/offsetY are frozen for the gesture, and the drag ghost is drawn through a drawn panel-coordinate map; the box, wire and parallel-frame paths all read that map, and the Settings canvas keeps its own plane. The three new tests assert rendered style.left/style.top, not the payload, which is what a stuck-on-screen node needs; the commit still uses the start scale. changelog.d/1575-drawer-drag-extreme-node.fixed.md is present and well formed.

Not run: the full npm run verify and the whole plugins/workflow suite — this review ran the one changed test file, by design. The Author's tip-wide verify is quoted in the body, and the queue entry re-runs everything on the merged tree.

Checkpoint in the body is well formed (1108 characters, pr 1576, claim Dev3, issue 1575, three files, checks.result green at this tip, next 35 characters), so the queue entry's read will pass.