humanspeak / svelte-virtual-chat

92%

main: 92%

push

github

fix(chat): let programmatic scrolls escape follow-bottom (#49)

* test(chat): add failing spec for programmatic slow-scroll livelock (#45)

A sustained programmatic upward scroll (30px/frame in the rAF phase — the
shape of a consumer scroll animation or a11y tool) can never leave the
bottom: with no scroll intent to suppress it, snap-to-bottom answers
every step, and each snap-back also resets the upward-travel accumulator
that the unfollow decision needs. The viewport oscillates at the bottom
forever while following=true.

A control test guards the intended magnet: one sub-threshold nudge keeps
following, before and after the fix.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(chat): keep upward-travel evidence through snap-backs so programmatic scrolls can escape

The travel accumulator reset whenever the viewport was within the follow
threshold — which is exactly where the component's own snap-backs land.
A sustained programmatic upward scroll (no scroll intent to suppress the
snap) had its evidence erased every cycle: step up, snap back, reset,
repeat — a livelock where the viewport could never leave the bottom.

Upward user-attributable movement now always accumulates, inside or
outside the threshold; only a non-upward arrival at the bottom (snap-
backs, downward scrolls, clamps) wipes the slate. Trackpad noise still
resets on every downward half-tick, wheel input still escapes via
intent, and the sub-threshold magnet is unchanged (guarded by the new
control test). Fixes #45

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* docs(readme): add tokenmaxing badge

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* refactor(chat): move the upward-travel transition rule into chatScrollPolicy

Cleanups from a reuse/simplification/efficiency/altitude review pass —
no behavior changes:

- accumulateUpwardTravel is a pure policy function now, co-located with
  its consumer (decideFollowBottomAfterScroll), its shared attribut... (continued)

190 of 218 branches covered (87.16%)

Branch coverage included in aggregate %.

4 of 4 new or added lines in 1 file covered. (100.0%)

346 of 365 relevant lines covered (94.79%)

48.42 hits per line