perf: bound collection update and viewport work for large data sets

[subinium]

Priority: P2
Evidence: Source-supported workload investigation
Baseline: v0.23.0 source 37933f9, audited 2026-09-05. Native execution/compile probes used Rust 1.98.0 on aarch64 macOS unless explicitly stated otherwise.

Problem and Reproduction

push_item rebuilds cache-coupled views on every append; sequential loading can repeat whole-list work. Variable-height End navigation still advances top linearly despite prefix lookup support. Non-virtual lists/tables/palettes can prepare offscreen commands; table pagination and existing file-picker virtualization must be distinguished.

src/widgets/collections.rs:94 · src/widgets/collections.rs:354 · src/context/widgets_interactive/rich_markdown.rs:1374 · src/context/widgets_interactive/selection.rs:366

Scope

• Measure10^3/10^4/10^5-item append/filter/navigation/render workloads at a fixed viewport.
• Evaluate batch mutation/incremental search caches and direct prefix-based jumps.
• Optimize offscreen preparation without removing supported full-list or pagination semantics.

Expected Effect

Keep large collections responsive with work tied to actual changes and visible rows where appropriate.

Acceptance Criteria

• Compare push loop against set_items/batch alternatives and document the contract.
• End/Home/filter jumps preserve selected-item and variable-height mappings.
• Measure release latency, command counts and memory; do not claim all widgets are O(viewport) beforehand.
• Keep already bounded VecDeque logs and file-picker virtualization; add only remaining gaps.
• Follow the repository quality gates and the evidence rules in tracking: post-v0.23 correctness, resource efficiency and WASM completion #366; record actual results rather than treating a proposed test as passed.

Boundaries and Dependencies

Do not weaken cache coherence to obtain lower timings.

• perf: establish release-mode workload, allocation and idle-resource baselines #414
• widgets: keep virtual-list navigation consistent with ListState filtered selection #424
• widgets: keep per-item heights aligned across list mutations #425
• widgets: resolve percentage table columns against the actual table content width #426

Related closed work: #345, #319. This is a scoped follow-up at the audit baseline, not a claim that every prior fix is absent.

Tracking: #366. This issue records unresolved work, not an implemented fix.

[subinium]

v0.24.0 Disposition

Incremental append, filtered/variable-height metadata and prefix jumps are implemented, alongside the parent-geometry recording cache. Measured 100k list append completes in about 4.77 ms; the original baseline did not finish 100k within 3 s. Ordinary list and unpaged table still emit n+4 and n+6 commands; virtual/paged fixtures emit constant 45/27 commands. Full 100k rendering samples regressed (list 18.766 -> 23.116 ms; table 22.037 -> 28.563 ms) in the combined tree, without causal attribution. The later 10k tiny-table recording regression was independently fixed (884.31 -> 8.06 ms). Keep broader ordinary-rendering/retained-memory work open and avoid a blanket virtualization/speedup claim.

Implementation and evidence mapping: PR #433. Release v0.24.0 and its 20/20 release workflow are verified. This issue stays open only for the remaining acceptance/investigation described above; publication alone does not establish it.