Render each card type's icon once per indexing job

[habdelra]

During indexing, the prerenderer's index visit renders a card's icon alongside its search-doc data. Icon markup is a pure function of the card's type — it comes from the type's static icon component, never from instance data — yet the visit rendered it once per file. A from-scratch index of an 885-card realm with 11 distinct card types performs 885 icon renders where 11 would do; each one is a route transition plus render-settle inside the hot path's per-visit floor.

The prerender server now keeps a job-scoped memo of icon markup keyed on the card's type. The first index visit for a type renders the icon; every later visit of that type in the same job reuses the captured markup and skips the icon route entirely. File icons get the same treatment keyed on the file's type — a file visit whose icon is memoized performs no page work at all in its file-render pass (the file-data stash is skipped too).

Rows are unchanged: icon_html on an index row is byte-identical to what a fresh render produces.

How it works

• The index visit's card pass enters the render app via render.meta and renders the icon afterward as an in-page child transition. Meta goes first because the memo key is types[0], which the meta pass resolves. The fused visit is unchanged: it still runs meta last, after the format renders mark the linked fields they read as used.
• A meta error short-circuits the pass: the page is showing the error route at that point, so a subsequent icon capture would wait on fresh render output the page can no longer produce and ride out the full render timeout, evicting the tab. The row lands as an error row with no icon — the same skip the html-route entry applies when its isolated render errors. A card whose meta errors deterministically (e.g. a computed that reads a broken link) takes this path on every visit.
• The memo lives on the RenderRunner, one slot per affinity, keyed by the visit's jobId plus the loader epoch. Indexing serializes per realm, so at most one job is active per slot; a visit carrying a different job key replaces the slot outright. A module edit arrives as a new job, so stale icon markup cannot survive a module change.
• Visits without a jobId (on-demand renders) never read or write the memo.
• clearCache visits bypass the memo read — they are asked for a pristine render — and overwrite the entry with their fresh capture.
• releaseBatch also drops the affinity's memo. Correctness doesn't depend on this (the job-key check already isolates jobs); it just frees the memory promptly.

Testing

New tests in prerendering-test.ts cover same-job reuse (byte-identical markup, hit/miss counters), per-type keying (a second card type renders its own distinct icon), job-change reset, jobless visits leaving the memo untouched, clearCache bypass, releaseBatch and affinity-disposal cleanup, and the meta-error short-circuit (an instance whose computed reads a broken link produces an error row promptly, with no icon render, no render timeout, and no tab eviction). The existing lossless-partition test continues to pin the index visit's output against the fused visit's, byte for byte.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: cfeb5a24ef

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Copilot]

Pull request overview

This PR optimizes realm indexing prerenders by memoizing rendered icon HTML per card/file type within an indexing job, so repeated visits of the same type can reuse captured markup and skip the icon route work.

Changes:

• Add an affinity-scoped, job-keyed icon HTML memo to RenderRunner and use it during visitType: 'index' card + file passes to avoid redundant icon rendering.
• Clear the icon memo when an indexing batch releases an affinity, and expose a test-only accessor for memo stats/keys.
• Add coverage in prerendering-test.ts for memo hits/misses, per-type behavior, job boundaries, jobless visits, clearCache bypass, and releaseBatch cleanup.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
packages/realm-server/prerender/render-runner.ts	Introduces job-scoped icon memoization and integrates it into index-visit card/file render flows.
packages/realm-server/prerender/prerenderer.ts	Clears the icon memo on releaseBatch and exposes a test-only accessor for memo observability.
packages/realm-server/tests/prerendering-test.ts	Adds new tests validating icon memo correctness and lifecycle behaviors.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[github-actions]

Host Test Results

    1 files  ±0      1 suites  ±0   2h 33m 23s ⏱️ +35s
3 470 tests ±0  3 455 ✅ ±0  15 💤 ±0  0 ❌ ±0 
3 489 runs  ±0  3 474 ✅ ±0  15 💤 ±0  0 ❌ ±0 

Results for commit 3335610. ± Comparison against earlier commit 1edeb49.

Realm Server Test Results

    1 files  ±0      1 suites  ±0   11m 27s ⏱️ - 1m 53s
1 822 tests +1  1 822 ✅ +2  0 💤 ±0  0 ❌  - 1 
1 901 runs  +1  1 901 ✅ +2  0 💤 ±0  0 ❌  - 1 

Results for commit 3335610. ± Comparison against earlier commit 1edeb49.

[lukemelia]

It feels funny to me that you define the icon on card def the same way you define fitted, embedded, etc, but it has different semantics. No need to solve this in this PR though, lol!