Speed up the card-source-endpoints realm-server suite (per-module boots + template dedupe)

[lukemelia]

What

tests/card-source-endpoints-test.ts (55 tests) paid a full-stack realm boot per test — clone the template Postgres DB → tmp dir + fixture copy → construct a Worker → boot the Realm → logInToMatrix() → build a RealmServer → start a supertest server → tear down — because the harness default is mode: 'beforeEach'. It also built two distinct templates for size-limit modules that differ only by request-time limits.

This lands the two lowest-risk, highest-impact items from CS-12935.

1. mode: 'before' for the read-only modules

Boot once per module instead of once per test for the three modules that don't mutate shared realm state:

• card source GET request › public readable realm (12 tests)
• card source GET request › permissioned realm (4 tests)
• card source HEAD request › public readable realm (3 tests)

Each test either reads fixture files that no test mutates or writes to a path unique to that test, so a shared boot stays order-independent. The harness (setupDB) already pairs before/after and clones the template DB once under mode: 'before', so this is a mechanical change that removes ~16 per-test boots.

2. Consolidate the two size-limit templates

public writable realm with size limit now carries the same {file: 512, audio: 2048, video: 8192} limits as public writable realm with size limit for binary, so both hash to one cached template. The template cache key includes the limits, but limits are request-time config and don't affect indexed content, so the extra two are inert for the .gts test and collapse two identical builds into one.

Deferred (follow-ups from the ticket)

• Fixture downgrade (GET public-readable realistic → simple): the simple fixture lacks person-with-error.gts and hello.test.gts that the module reads, and adding them would perturb the shared simple fixture for other modules. Once mode: 'before' collapses that module to a single boot, the realistic-vs-simple boot differential is paid only once, so the remaining value is small.
• Worker-less boots for read-only modules, Matrix-login memoization, and a pooled-realm harness — larger harness lifts.

Verification

Changes are lint-clean. End-to-end timing was intended via scripts/measure-test-file.sh per the ticket's method note, but local runs are currently blocked by an unrelated stale local dev stack; CI runs this suite against a built host and is the authoritative source for the green run and before/after shard timings. The mode: 'before' path is already exercised by existing suites, and the size-limit consolidation preserves each module's own runtime limits (only the shared template is deduped).

🤖 Generated with Claude Code

[github-actions]

Host Test Results

    1 files      1 suites   2h 5m 13s ⏱️
4 734 tests 4 720 ✅ 14 💤 0 ❌
4 749 runs  4 735 ✅ 14 💤 0 ❌

Results for commit 9e2992f.

Realm Server Test Results

    1 files    210 suites   1h 14m 0s ⏱️
2 764 tests 2 764 ✅ 0 💤 0 ❌
2 803 runs  2 803 ✅ 0 💤 0 ❌

Results for commit 9e2992f.

[Copilot]

🟢 Approval recommended

No unresolved blocking issues were identified.

Pull request overview

This pull request speeds up card-source endpoint tests by sharing read-only realm boots and deduplicating equivalent templates.

Changes:

• Uses mode: 'before' for three read-only modules.
• Reuses cached templates with aligned size limits.

File summaries

File	Summary
packages/realm-server/tests/card-source-endpoints-test.ts	Updates test setup and size-limit template reuse.

Review details

• Files reviewed: 1/1 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.