Cache realm setup for host acceptance/integration tests

[IanCal]

Summary

This PR centralizes SQLite snapshot caching for host test realm setup so we get speedups across the whole host suite without requiring per-test wrapper plumbing.

What changed

• Added snapshot lifecycle support to SQLiteAdapter (packages/host/app/lib/sqlite-adapter.ts):
  • exportSnapshot(snapshotName?)
  • importSnapshot(snapshotName)
  • hasSnapshot(snapshotName)
  • deleteSnapshot(snapshotName)
  • deleteSnapshotsByPrefix(prefix)
• Added robust snapshot copy/restore internals in SQLiteAdapter:
  • schema-aware table cloning/copy helpers
  • batched import transaction with rollback on failure
  • tracked snapshot metadata and cleanup on close()
• Added centralized cache helpers in packages/host/tests/helpers/index.gts:
  • withCachedRealmSetup(...)
  • module-derived cache key resolution via QUnit module name
  • deterministic snapshot naming with simpleHash
  • additional key support (for setup variants)
• Added automatic cache teardown by module prefix:
  • explicit hook helper: setupRealmCacheTeardown(...)
  • global registration via QUnit.moduleDone(...) so snapshots are cleaned per module and don’t grow unbounded during long runs
• Wired caching directly into realm setup entry points:
  • setupAcceptanceTestRealm(...) now goes through withCachedRealmSetup(...)
  • setupIntegrationTestRealm(...) now goes through withCachedRealmSetup(...)
  • additional cache key is derived from setup inputs (contents, realmURL, permissions, startMatrix) using stable serialization + hash

Why this approach

• The previous approach required opt-in per module/test file, which is easy to miss and hard to keep consistent.
• Centralizing at setupAcceptanceTestRealm / setupIntegrationTestRealm guarantees coverage for all current callers.
• Using module name + hashed setup inputs gives safe cache partitioning between different setups.
• Automatic teardown prevents snapshot accumulation from becoming an unbounded memory/perf issue during long test sessions.

Behavior impact

• No test behavior changes intended.
• Primary effect is faster repeated setup during test runs.
• Snapshot state is now bounded by module lifecycle cleanup.

Notes

• The branch also contains a minor whitespace-only cleanup in packages/host/tests/acceptance/host-mode-test.gts.
• Full suite was intentionally not run locally due runtime constraints; CI should validate full coverage.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 480ef88244

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".

[chatgpt-codex-connector]

Detach snapshots after export to avoid attach-limit failures

withCachedRealmSetup() now writes a new snapshot for every cache miss, but each snapshot stays attached until module teardown, so distinct setup variants accumulate throughout the module. SQLite permits only 10 attached databases; modules with many setupIntegrationTestRealm variants (for example packages/host/tests/integration/realm-test.gts, which calls it many times) can hit too many attached databases before moduleDone cleanup runs. This turns the cache into a hard test failure path once enough unique keys are seen in one module.

Useful? React with 👍 / 👎.

[github-actions]

Preview deployments

• Host staging preview
• Host production preview

[github-actions]

Host Test Results

    1 files  ±0      1 suites  ±0   1h 16m 35s ⏱️ - 29m 47s
1 854 tests ±0  1 697 ✅  - 142  14 💤 ±0    0 ❌ ±  0  143 🔥 +142 
1 869 runs  ±0  1 569 ✅  - 284  14 💤 ±0  143 ❌ +142  143 🔥 +142 

For more details on these errors, see this check.

Results for commit 4677c09. ± Comparison against base commit e7401f2.