Fix large GEDCOM files failing to load or freezing the browser

[robekl]

Summary

Large GEDCOM files (10K+ individuals) either failed silently in Firefox/Safari or froze Chrome for 10–20+ seconds. This PR fixes five bugs in the viewer. The underlying chart rendering bug (exponential pedigree traversal) is addressed separately in PeWu/topola#110.

Bugs fixed

1. Firefox and Safari reject the uploaded file immediately

navigate(..., {state: {data: gedcom}}) passes the GEDCOM through history.pushState. Firefox caps this at 640KB; Safari at ~512KB. A 10MB GEDCOM caused an immediate SecurityError — the chart never loaded in either browser.

Fix: Store the GEDCOM in a bounded in-memory store (gedcom_store.ts, capped to 1 entry — new upload evicts the previous). navigate() carries only a content fingerprint hash.

2. JSON.stringify blocks the browser for 10–20 seconds

prepareData serialized the full TopolaData (33MB total) to sessionStorage on every load — always failing the 5MB limit.

Fix: Skip serialization when the raw GEDCOM exceeds 512KB.

3. Chart re-renders 3–5 times per load, each freezing the browser

The Chart useEffect has no dependency array, so it ran after every React render. Each call re-ran the full D3 layout.

Fix: Return early in renderChart() when neither initialRender nor resetPosition is set.

4. Progress indicator disappears before the chart renders

The "Loading…" pill was shown only during AppState.LOADING, disappearing before the D3 layout ran.

Fix: Persistent fixed overlay driven by loadingStatus state, cleared via onFirstRender callback after the chart DOM is ready (intentionally pre-animation — the chart is interactive at this point, and keeping the pill during the 400ms fade-in would create a confusing overlap).

5. Donatso chart: progress pill never cleared

DonatsoChart did not call onFirstRender, so loadingStatus was never cleared.

Fix: Add onFirstRender? to DonatsoChartProps and call it in the first-render branch of DonatsoChart's useEffect.

Additional improvements

• Async parsing with progress — convertGedcom is async with await setTimeout(0) yields between the four major steps, keeping the spinner responsive and allowing incremental GC. An onProgress callback threads through the full load pipeline (all five datasources: uploaded, URL, Google Drive, WikiTree, embedded) showing step-by-step status.
• Search index idle pre-build — Lunr index (~600ms for 23K individuals) is scheduled via requestIdleCallback after data loads. Event listener for idle callbacks uses const-scoped handles to eliminate type casts.
• Fast file fingerprint — replaced full-file md5() (~240ms for 10MB) with metadata + 4KB content sample (~1ms), extracted to src/util/file_fingerprint.ts.
• Embedded listener cleanup — the window.addEventListener in EmbeddedDataSource.loadData was never removed, accumulating listeners across calls. Now removed on first settlement.
• Test isolation — gedcom_store.spec.ts adds beforeEach(clearStoredGedcom) to prevent Jest module cache from causing order-dependent test failures.

Tests

All passing: 89 unit tests, 13 visual regression tests (macOS snapshots updated to include new progress overlay).

Dependency note

The root-cause chart rendering fix (exponential pedigree traversal in topola) is in PeWu/topola#110. This viewer PR is independently mergeable.

Co-authored-by: Claude Sonnet 4.6 noreply@anthropic.com

[PeWu]

Thank your for all the fixes! Do you think you could ask Claude to split the PR into multiple smaller ones? Like one per fix? It will make reviewing the changes easier.

[PeWu]

In the Donatso chart, the "Rendering chart" pill doesn't disappear.
To reproduce, start with npm start and open http://localhost:3000/#/view?url=https%3A%2F%2Fraw.githubusercontent.com%2FPeWu%2Ftopola%2Frefs%2Fheads%2Fmaster%2Fdemo%2Fdata%2Ffamily.ged&view=donatso

Also makes npm run test:visual fail.

[PeWu]

I reviewed the code. You don't need to split it (unless you already did 😄 )
I left some minor comments. The most important thing is to make tests pass

[robekl]

Fixed the Donatso pill issue — added onFirstRender to DonatsoChartProps (called from its first-render useEffect branch), consistent with how the D3 Chart component already handles it. Updated all visual regression snapshots for macOS; the 12 non-Donatso snapshot mismatches were pre-existing staleness, chart-donatso was the only new failure. All tests pass: unit (89), visual (13).

[robekl]

Fixed — added onFirstRender? to DonatsoChartProps and call it in the first-render branch of DonatsoChart's useEffect, consistent with how the D3 Chart component handles it.

While fixing this I also addressed a few issues surfaced during code review:

• onProgress now threads through all five datasources — WikiTree and Embedded were previously omitting it at the call site. Embedded additionally needed it wired through the event-listener closure into loadGedcom.
• Embedded event listener leak fixed — window.addEventListener in EmbeddedDataSource.loadData was never removed. It now wraps resolve/reject to self-remove on first settlement.
• gedcom_store test isolation — added beforeEach(clearStoredGedcom) so tests are independent of execution order (Jest caches module-level state).
• Search useEffect handle scoping — replaced shared let handle union type with const-scoped handles in each branch, eliminating the type casts.

All 89 unit tests and 13 visual tests pass.

[PeWu]

Thank you!