Repository navigation
perf(llc/frontend): Gantt axis ticks are generated for the whole range with no viewport culling #14769
Description
Problem
axisTicks emits one tick for every step across the entire data range with no
viewport culling.
autobot-frontend/src/views/llc/GanttTimelineView.vue:284-298
const axisTicks = computed(() => {
const ticks: { x: number; label: string }[] = []
const step = zoom.value === 'day' ? 1 : zoom.value === 'week' ? 7 : zoom.value === 'month' ? 30 : 91
for (let ms = rangeStart.value; ms <= rangeEnd.value; ms += step * DAY_MS) {
...
ticks.push({ x: xForMs(ms), label })
}
return ticks
})At zoom: 'day' the step is 1 day and ZOOM_PX_PER_DAY.day is 40 (line 161),
so a multi-year range produces thousands of SVG tick elements, nearly all of
them outside the visible scroll window. Each carries a toLocaleDateString
call as well.
Proposed fix
Clamp the loop to the visible horizontal window (scroll offset + client width,
converted back to ms via the inverse of xForMs) plus a small overscan, and
recompute on scroll. Label formatting then only runs for ticks that render.
Acceptance criteria
- Tick count is bounded by the viewport width divided by tick spacing, not
by the data range. - Ticks still appear correctly when scrolling to either end of the range at
every zoom level (day/week/month/quarter). - Test asserts tick count stays bounded for a multi-year range at
zoom: 'day'.
Source
Found while reading our own Gantt grid code during
docs/research/canvas-grid-rendering-review.md Phase 2. Recorded per the
report-and-fix-every-discovered-problem rule rather than folded into an
unrelated change. Sibling finding in the same file: #14767.
Activity
mrveiss commentedon Aug 23, 2026
Closed — verified in base
Delivered by #14796, merged into
Dev_new_guiasd88b6d8aa.Closed by hand — PRs here target
Dev_new_gui, not the default branch, so theCloses #...keyword never fires. Evidence read fromorigin/Dev_new_gui, notfrom the PR diff.
Axis ticks are culled to the visible window (
TICK_OVERSCAN_PX, 400px eitherside), driven by a
ResizeObserverattached from awatchon the element ref—
.gantt-scrollsits behindv-elseand does not exist until a timelineloads, so an
onMounted-only hookup would have observed nothing and left thefallback permanently engaged while appearing to work.
Two invariants matter more than the culling itself, and both are in base:
indistinguishable from a chart with no data — the same conflation bug(llc/frontend): board and timeline views render a load failure as a legitimately empty board — no error state, no distinction from 'no work' (from #13993) #14064
records for this view's empty state.
if (from > to)).Added for cluster layout still animates for a reduced-motion user #14806... correction, Gantt axis renders no ticks after a zoom change while scrolled #14799, which found that the
viewWidth > 0guard covered only the unmeasured case while a viewport measured under a
previous
pxPerDayreached an empty axis by a different route. The guard wasnarrower than the invariant its own docstring stated.
The interaction with #14767 is the part worth remembering. Culling is
suspended during a PNG export, or the export would have silently shipped a
chart missing every gridline and date label outside the viewport — a worse
defect than the 1x softness being fixed, and invisible in review, because a
complete PNG and a gutted one both look like "an export happened". There is a
test that counts gridlines at serialisation time via a stubbed
XMLSerializer, and another asserting culling is restored after a failedexport.