[Discussion] Drop the frame model for mind map / flowchart — treat them as free-floating like shapes?

[vibhavkatre]

Reported by Vibhav (QA, 2 Aug 2026).

Question: do we need the frame model for mind map / flowchart on the unified canvas, or can they be free-floating like normal shapes? Several current bugs (#96 gap, #98 tidy-up hidden, #100 text edit, #104-adjacent, this batch's Delete/marquee) stem from the frame abstraction.

Action: I'll prepare a pros/cons (frame model vs free-floating) — covering interaction consistency, layout/auto-arrange, selection/delete/marquee, collab/CRDT mapping, and migration cost — for us to discuss before deciding. Applies to both mind map and flowchart.

[vibhavkatre]

Frame model vs free-floating — pros/cons (for a decision)

Where we are: on the unified canvas a mind map / flowchart is a frame — a positioned region with an origin (top-left on the shared canvas), content auto-laid-out inside, and a hit-rect behind it for selecting/dragging the whole thing. Block shapes, by contrast, are free-floating individual objects handled by the shared selection/interaction machinery.

Option A — keep frames

Pros

• Auto-layout is always-on; the frame is a natural container for "this cluster arranges itself."
• Move-as-a-unit: drag the frame, the whole diagram moves together.
• Map-wide ops (Tidy up / direction / number) map cleanly onto the frame.
• Clean separation between auto-layout sub-models and free shapes.

Cons — and note how many open bugs are frame artifacts

• Second coordinate space (origin translate) → conversion bugs: the off-screen insert and the Flowchart on-node '+' menu opens too far from the node and can hide behind the palette #96 picker-gap live here.
• Interaction is special-cased and inconsistent with block shapes: Cannot edit text inside flowchart nodes (unified canvas) #100 (text edit), Mind map: Delete does nothing when the frame is selected #120 (delete), Mind map: click-drag marquee selection doesn't work (only single-click frame select) #121 (marquee) are all "works for shapes, broken in the frame."
• Map-wide actions aren't exposed because the frame has no consistent toolbar → Expose flowchart layout actions (Tidy up / direction / number) on the unified canvas #98 (Tidy up hidden).
• Insert placement logic causes the canvas to pan on add (Adding a shape/node must never pan the canvas; click-to-add is visible-area only #119).
• Every new interaction (click-to-place Inserting a flowchart/mind map places it far below the viewport; want click-to-place #75, curved connectors Mind map: curved connectors #117, hover handles Mind map: hide connectors by default, reveal on hover; show default (+) handles on both sides #118) must be re-implemented for the frame context.

Option B — free-floating (nodes are ordinary canvas objects)

Pros

• One interaction model — selection, marquee, delete, text-edit, drag all reuse the shared block machinery. This dissolves Cannot edit text inside flowchart nodes (unified canvas) #100/Mind map: Delete does nothing when the frame is selected #120/Mind map: click-drag marquee selection doesn't work (only single-click frame select) #121/Flowchart on-node '+' menu opens too far from the node and can hide behind the palette #96 rather than fixing each.
• No second coordinate space → no origin-conversion bugs.
• Connectors, hover handles, click-to-place reuse the block implementations.
• Matches the product direction: Whimsical-style parent-shape/child-text (Mind map: only the parent is a shape node; child nodes are transparent text nodes (Whimsical), convertible via context menu #125), curved connectors (Mind map: curved connectors #117), hover + handles (Mind map: hide connectors by default, reveal on hover; show default (+) handles on both sides #118) all fit free objects.

Cons

• Lose automatic "move the whole map as one unit" → needs group/selection semantics (the model already has grouping for block — reusable).
• Auto-layout becomes an explicit Tidy up action rather than always-on (nodes can drift). Arguably fine — you already asked for manual placement + tighter Tidy up.
• Migration cost: existing unified docs store mind-map/flowchart sub-models with origin; free-floating needs a data migration (flatten nodes to canvas coords + promote edges/parent-links to first-class connectors) or a compat layer.
• Collab/CRDT mapping changes (nodes become top-level objects).

Recommendation

Lean B (free-floating) for the long run — most of the QA bugs are frame-abstraction tax, and every enhancement you've asked for fits free objects better. But it's a significant refactor + data migration, so I'd sequence it:

1. Now: keep frames; land targeted fixes for the worst bugs (Cannot edit text inside flowchart nodes (unified canvas) #100 text edit, Mind map: Delete does nothing when the frame is selected #120 delete, Mind map: click-drag marquee selection doesn't work (only single-click frame select) #121 marquee).
2. Then: plan the free-floating migration as a dedicated effort (peer of Drive/commenting) — it removes a whole bug class and unlocks Mind map: curved connectors #117/Mind map: hide connectors by default, reveal on hover; show default (+) handles on both sides #118/Mind map: node corner radius too large (looks oval); keep it subtle, rounded-rectangle on expansion #124/Mind map: only the parent is a shape node; child nodes are transparent text nodes (Whimsical), convertible via context menu #125 cleanly.

Your call: is frames' "move-as-one + always-on layout" worth the ongoing interaction-inconsistency/bug tax, or do we commit to the one-time free-floating refactor?

[vibhavkatre]

Free-floating refactor — implementation & migration plan (for review)

Decision recap: mind maps / flowcharts become free-floating canvas objects (like block shapes), not framed sub-models. Below is how, and how existing diagrams migrate.

1. Today's model (what changes)

• document.shapes[] / connectors[] — free block objects, each with an id + x/y. Per-object CRDT in collab (concurrent edits merge).
• document.mindmap / flowchart — sub-models with their own node/edge structures + a frame origin. In collab they're one blob in the meta ymap → concurrent edits clobber.
• The frame model is the source of the interaction bugs (Flowchart on-node '+' menu opens too far from the node and can hide behind the palette #96 picker gap, Cannot edit text inside flowchart nodes (unified canvas) #100 text edit, Mind map: Delete does nothing when the frame is selected #120 delete, Mind map: click-drag marquee selection doesn't work (only single-click frame select) #121 marquee) — nodes are special-cased instead of reusing block handling.

2. Target model

• A mind-map / flowchart node → a shape in shapes[], tagged with a role: { kind: 'mindmap' | 'flowchart', nodeType?, branches?, … }, positioned in canvas coords (no origin).
• Links → connectors in connectors[]: mind-map parent→child edges and flowchart edges become ordinary connectors (the connector model + the new anchor system from Connectors: show logical connection points on shapes while drawing (snap + hover), like Google Slides #138 already fit).
• Node-type geometry stays (reuse flowchartShapes.nodeShape / mind-map rendering) — only the ownership moves to the shared shape/connector arrays.
• Auto-layout ("Tidy up") becomes an explicit action that repositions these free shapes, instead of an always-on frame layout. Exposed on selection (fixes Expose flowchart layout actions (Tidy up / direction / number) on the unified canvas #98).

3. Migration (schema v2)

• Bump SCHEMA_VERSION → 2; add a step in migrateDocument() (runs on load, so it's lazy + backward-safe).
• For each doc with a mindmap/flowchart sub-model: flatten nodes to shapes at node.x + origin.x (positions preserved), convert branches/edges to connectors, then drop the sub-model. Idempotent; v1 docs upgrade on first open and re-save.
• Keep reading v1 docs forever (never hard-break old data). crdt_state seeds are re-derived from JSON, so no separate CRDT migration.

4. Collab/CRDT impact — a win

• Nodes move from the coarse meta blob into the per-object shapes/connectors ymaps → concurrent node edits merge instead of clobbering. META_KEYS drops mindmap/flowchart. Net simplification of useCollaboration.

5. Interaction/layout — deletions, not additions

• Selection / marquee / delete / drag / text-edit all come free from the block/shape machinery → Flowchart on-node '+' menu opens too far from the node and can hide behind the palette #96 / Cannot edit text inside flowchart nodes (unified canvas) #100 / Mind map: Delete does nothing when the frame is selected #120 / Mind map: click-drag marquee selection doesn't work (only single-click frame select) #121 dissolve rather than needing individual fixes.
• Mind-map/flowchart-specific ops (add-child, add-node, decision branches, tidy/direction/number) become operations over the tagged shapes + connectors.

6. Phasing (de-risked, incremental)

1. Model + migration behind a read path: teach the store/render to draw mind-map/flowchart shapes + connectors; add the v2 migration; keep the old sub-model render for un-migrated docs. (No user-visible change yet.)
2. Interaction: point mind-map/flowchart creation + editing at the shared shape/connector handlers; remove the frame hit-rect/drag.
3. Auto-layout as an action (Tidy up on selection) + the mind-map/flowchart palette inserts create tagged shapes.
4. Collab: move the keys out of meta; verify two-client merge.
5. Cleanup: delete the frame code paths + origin; then build Mind map: only the parent is a shape node; child nodes are transparent text nodes (Whimsical), convertible via context menu #125 (parent-shape/child-text) + Reintroduce a Settings dialog for Draw; first setting: mind-map child-node style (text vs shape) #126 (Settings) on this model.

7. Risks / open questions

• Scale of the diff — touches the store, canvas render, mind-map/flowchart composables, collab, and export/thumbnail (documentToSvg). I'd land it as the phased PRs above, each green, not one mega-PR.
• Auto-layout drift — free nodes can be dragged out of layout; Tidy up re-flows. Acceptable per your earlier feedback (tighter tidy-up).
• Decision-branch semantics — flowchart decision ports/labels need to survive as connector metadata; needs care.
• Export parity — documentToSvg must render the tagged shapes; covered by moving them into shapes.

Question for you: OK to proceed on this phased plan (starting with phase 1 — model + lazy migration, no visible change), or do you want to adjust the model tagging / auto-layout approach first?