humanspeak / svelte-virtual-chat / 25116581459

97%

main: 98%

push

github

perf(virtual-chat): memoize rendered slice by visible range (#16)

* perf(virtual-chat): memoize rendered slice by visible range

Replaces `messages.slice(visibleRange.start, visibleRange.end + 1)`
with a memoized `$derived.by` that returns the cached array reference
when both bounds are unchanged AND the head/tail still point at the
same message instances. At 60fps that drops a throwaway array per
frame on smooth scroll (the common case where scroll moves within a
single message's height window), and lets Svelte's keyed each block
short-circuit when the array reference is identical to the previous
derivation.

The head/tail identity check covers the chat semantics this component
supports: append (tail changes), prepend (visibleRange recomputes,
start/end change), whole-array replace (head/tail change), and
streaming (same array reference, same elements). Mid-array
replacement with unchanged head/tail isn't covered — it isn't part
of the supported semantics, and a full slice walk would defeat the
optimization.

The perf-bench fixture's `mutations10s` doesn't move on this change
because that metric is dominated by the per-frame `transform: translateY()`
mutation on the messages container, not by message-row add/remove or
slice churn. The win here is in JS execution time and GC pressure
(no allocation per scroll frame) plus Svelte's keyed-each fast path,
which the current bench doesn't measure directly.

Validation: all 56 unit tests pass, all 360 Playwright e2e tests pass
(append, prepend, streaming, bulk, history, header-footer, scroll-away,
genai-like). Type check + ESLint clean.

Refs humanspeak/svelte-virtual-chat#10

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(virtual-chat): address CodeRabbit feedback on slice memoization

Two changes:

- Strengthen the cache-hit predicate to also check the messages
  array reference. Without this, a consumer-side immutable
  mid-replace (`messages = [...slice(0,i), newMsg, ...slic... (continued)

73 of 78 branches covered (93.59%)

Branch coverage included in aggregate %.

154 of 155 relevant lines covered (99.35%)

92.85 hits per line