solidjs / solid / 35894344323

74%

main: 89%

push

github

perf(server): materialize scoped hole ids lazily (follow-up to #3599) (#3614)

`ssrScope` reserved its slot by formatting the scope's id string up front
(`nextChildIdFor` → `formatChildId`: a radix-36 toString plus concats)
for every scoped hole. Since #3599 every hole that is not provably
primitive is scoped, and most of those are text (`{row.label}`) — nothing
inside ever asks for a child id, so the string was pure waste: ~60% of
the +7–10% renderToString regression on spread-static-tail, ~22 ns/hole.

The reservation is now a bare counter increment. The wrapper swaps the
pair (`id = prefix`, `_scopeSlot = slot`) onto the parent owner and
`materializeId` folds it into the string on the first derivation — the
same `formatChildId(prefix, slot)` the eager form built, so every id
string produced anywhere is byte-identical; a scoped hole still consumes
exactly one parent slot at argument-evaluation time.

Readers of a possibly-pending owner's id are the two derivation sites
(`nextChildIdFor`, `allocateOwner`'s transparent copy); every other
`.id` read is an owner's own freshly created one and now goes through
`ownerId`, whose dev build asserts the premise (0 B in prod).

Microbench (dist, 5,000 `<li>{row.label}</li>` hydratable): 24.3 → 9.5
ns/hole over pre-#3599. spread-static-tail (vitest, 15 interleaved
rounds): −1.3…−3.9% vs #3599 as landed; HTML identical.

Co-authored-by: Claude via Cursor <noreply@cursor.com>

592 of 865 branches covered (68.44%)

Branch coverage included in aggregate %.

873 of 1121 relevant lines covered (77.88%)

19.81 hits per line