Matt Kane
18593475bb
* chore: bump TS target from ES2022 to ES2024
Unlocks ES2023 Array methods (toSorted, toReversed, toSpliced, findLast)
and ES2024 types (Promise.withResolvers, Object.groupBy) in type
checks. Node 22 already supports these at runtime; the old target just
made the types invisible.
* perf: hydrate taxonomy terms on collection and entry queries
Adds eager hydration of taxonomy terms on getEmDashCollection and
getEmDashEntry results via a single batched JOIN query. Each entry
now exposes data.terms keyed by taxonomy name alongside the existing
bylines hydration. Templates that previously looped and called
getEntryTerms per entry can drop the N+1.
Measured on blog-demo home page (7 posts, 2 taxonomies), warm render:
use: 443ms -> 317ms (-28%)
euw: 274ms -> 178ms (-35%)
ape: 682ms -> 413ms (-40%)
aps: 1008ms -> 156ms (-85%)
The aps figure reflects the dominant cost of serial D1 round-trips
across a replica link: eliminating 6 of them dwarfs every other
homepage query put together.
New exports: getAllTermsForEntries, invalidateTermCache. A cached
hasAnyTermAssignments probe short-circuits the JOIN entirely on sites
that use no taxonomies; the cache is invalidated when term
assignments are written.
Reserved field slugs now also block terms, bylines, and byline so
new user fields cannot shadow the auto-hydrated values.
* chore(blog-demo): add account_id to wrangler.jsonc
Unblocks wrangler deploy in multi-account environments without
requiring CLOUDFLARE_ACCOUNT_ID in the shell. The EmDash CMS account
ID is non-secret.
* fix: harden hydration probes (per-DB cache, rethrow, chunk IN, cross-dialect)
Addresses review feedback on the taxonomy hydration PR. Applies the
same fixes to the pre-existing byline hydration, which had identical
shortcomings.
- Per-DB cache via WeakMap<Kysely<Database>, boolean>. The probe result
("does any row exist in _emdash_bylines / content_taxonomies?") used
to live at module scope, which meant preview sessions, D1 Sessions
API replicas, and Durable Object playground databases could see
cached values from an unrelated DB instance.
- Rethrow non-table-missing errors from the probe. The old code
swallowed every error and returned false, silently hiding real DB
failures (permissions, connectivity, syntax).
- Chunk IN (...) clauses with SQL_BATCH_SIZE so large collection
queries can't exceed D1's bound-parameter limit. Matches the
existing getContentBylinesMany pattern.
- Cross-dialect missing-table detection via a new
utils/db-errors.ts#isMissingTableError helper that matches SQLite,
PostgreSQL, and MySQL phrasings. Wired into loader.ts (replacing
three inline duplicates), query.ts hydrators, and both probes.
All 2393 core tests pass, including 7 new ones covering the helper
and the batched multi-taxonomy fetcher.
20 lines
426 B
JSON
20 lines
426 B
JSON
{
|
|
"compilerOptions": {
|
|
"target": "ES2024",
|
|
"module": "preserve",
|
|
"lib": ["ES2024"],
|
|
"moduleResolution": "bundler",
|
|
"resolveJsonModule": true,
|
|
"allowJs": false,
|
|
"strict": true,
|
|
"esModuleInterop": true,
|
|
"skipLibCheck": true,
|
|
"forceConsistentCasingInFileNames": true,
|
|
"declaration": true,
|
|
"declarationMap": true,
|
|
"sourceMap": true,
|
|
"noUncheckedIndexedAccess": true,
|
|
"noImplicitOverride": true
|
|
}
|
|
}
|