gallery/vendor/i-slint-core/textlayout/sharedparley/selection.rs
Greg Pomerantz 2f4768f354 Fix 120Hz scroll jank end-to-end: render-ahead present + full-cache eviction off the UI thread
Three root causes, all fixed and verified on the Pixel 9 Pro:

1. Present phase (backend + renderer): VSYNC_LOCKED_RENDER defers the
   render to the vsync wake; RENDER_AHEAD presents the previous frame at
   the start of render() (phase = wake + ~1.1 ms) and holds the new
   frame one slot. Crash fix: a held wgpu frame must be PRESENTED, never
   merely dropped, before the next acquire (AlreadyAcquired panic at
   launch otherwise); discard_pending_frame presents too.

2. Full-res prefetch churn (src/lib.rs): the 8-slot full cache cycled
   through a ~20-cell viewport forever (137 decodes/20 s) and each
   eviction free()'d a 50 MB Vec on the UI thread (20-52 ms CPU spikes,
   ~7/s = the residual SLOWEV NonMain jank). Fix: 60 s tombstone
   cooldown on evicted paths (idle prefetch skips, urgent tap bypasses)
   + off-thread reaper for the freed blobs (insert: 21 ms -> 4 us).

3. (earlier) Skia final flush + Mailbox present mode (vendor, M3).

Instrumentation (all behind --features perf): OTHR/INVK/INVK_TAG/FLCK
lines for cross-thread closure timing and push-site tagging.

Warm measured swipe, 8580-file grid: cadence median 8.33 ms (exact
120 Hz), p99 15.5 ms, 5/552 gaps >16.7 ms (0.9%, max 28.3 ms, zero
>33 ms); PRES phase median ~994 us, zero late frames. Docs updated
(docs/BUGS/scroll-jank-120hz.md, full scorecard + line reference).

Also: vendored i-slint-core + wgpu-hal (vkshim interposition patch),
vkshim profiling tooling, backend made a non-optional dep, debuggable
flag for simpleperf (REMOVE before shipping).
2026-10-30 02:49:23 -04:00

196 lines
9.0 KiB
Rust

// Copyright © SixtyFPS GmbH <info@slint.dev>
// SPDX-License-Identifier: GPL-3.0-only OR LicenseRef-Slint-Royalty-free-2.0 OR LicenseRef-Slint-Software-3.0
// cSpell: ignore bidi
//! Selection geometry, resolved once per draw into per-line horizontal spans.
//!
//! The same rounded span edges fill the highlight and clip the glyph runs, so the two can't
//! disagree -- see [`SelectionSpan`].
use super::*;
/// One contiguous run of selected text on a single line.
///
/// Selection is deliberately *not* expressed as a text style. A style can only ever recolor a
/// whole glyph, but a selection boundary may fall in the middle of one: with an `fi` ligature,
/// selecting just the `i` leaves parley with a single glyph whose style comes from the cluster's
/// first character (`Glyph::style_index` is `char_infos[cluster_id]`), so the whole ligature would
/// be painted unselected while the highlight covers only its right half. Instead the spans below
/// are used twice — to fill the highlight background, and to clip the glyph runs that straddle a
/// boundary so each half is drawn in its own color.
#[derive(Clone, Debug)]
pub(super) struct SelectionSpan {
/// Index into `Layout::paragraphs`.
pub(super) paragraph: usize,
/// Line within that paragraph.
pub(super) line: usize,
/// Highlight rectangle in item coordinates, ready to fill. Its horizontal edges are snapped to
/// whole device pixels where they are computed, and [`Self::x`] hands the very same edges to
/// the glyph clip -- so the highlight edge and the clip edge cannot disagree and leave a sliver
/// of wrongly-colored glyph on top of the highlight.
background: PhysicalRect,
}
impl SelectionSpan {
/// Horizontal extent of the highlight, in the same coordinate space as `GlyphRun::offset()`.
pub(super) fn x(&self) -> Range<f32> {
self.background.min_x()..self.background.max_x()
}
}
/// Sorted by `(paragraph, line, x.start)`: the spans belonging to one line form a contiguous slice,
/// and within that slice they run left to right. Both halves are load-bearing -- see
/// [`Self::for_line`] and the segment walk in `draw_glyph_run_with_selection`.
#[derive(Clone, Debug, Default)]
pub(super) struct SelectionSpans(pub(super) Vec<SelectionSpan>);
impl SelectionSpans {
pub(super) fn is_empty(&self) -> bool {
self.0.is_empty()
}
pub(super) fn backgrounds(&self) -> impl Iterator<Item = PhysicalRect> + '_ {
self.0.iter().map(|span| span.background)
}
/// The spans covering one line. Both the stored spans and the draw loop walk paragraphs and
/// lines in order, but a binary search keeps this independent of that ordering.
pub(super) fn for_line(&self, paragraph: usize, line: usize) -> &[SelectionSpan] {
let key = (paragraph, line);
let start = self.0.partition_point(|span| (span.paragraph, span.line) < key);
let len = self.0[start..].partition_point(|span| (span.paragraph, span.line) == key);
&self.0[start..start + len]
}
}
/// How selected glyphs are painted, resolved once per draw call.
pub(super) struct SelectionRendering<'a, R: GlyphRenderer> {
pub(super) spans: &'a SelectionSpans,
/// Forced fill for selected glyphs. It wins over `Brush::override_fill_color` and
/// `Brush::link_color`, so a colored span or a link inside the selection still reads as
/// selected.
pub(super) foreground: <R as GlyphRenderer>::PlatformBrush,
}
/// How much of one glyph run a selection covers.
pub(super) enum RunCoverage {
/// No selected pixels: draw once, in the run's own brush.
Unselected,
/// Fully selected: draw once, in the selection foreground. No clip needed.
Full,
/// A boundary falls inside the run — possibly inside a ligature. The line's spans have to be
/// drawn separately, each clipped to its own horizontal band.
Partial,
}
/// Classifies `run_x` against the selection spans of the line it sits on.
pub(super) fn run_coverage(run_x: &Range<f32>, spans: &[SelectionSpan]) -> RunCoverage {
// Empty runs (and the degenerate zero-advance runs parley emits for ligature tails) can't
// show a boundary.
if spans.is_empty() || run_x.end <= run_x.start {
return RunCoverage::Unselected;
}
let mut overlapping = false;
for span in spans {
let span_x = span.x();
if span_x.start <= run_x.start && span_x.end >= run_x.end {
return RunCoverage::Full;
}
overlapping |= span_x.start < run_x.end && span_x.end > run_x.start;
}
if overlapping { RunCoverage::Partial } else { RunCoverage::Unselected }
}
impl Layout {
/// Resolves `selection_range` into per-line horizontal spans, for the lines whose boxes may
/// intersect `visible_band` (a physical y range in item coordinates; spans feed drawing only,
/// so off-screen lines need none). Pass an unbounded band to resolve everything.
///
/// Parley already splits a ligature into one cluster per character and apportions the
/// advance between them, so the geometry it reports is accurate to sub-glyph precision --
/// selecting the `i` of an `fi` ligature yields exactly the ligature's right half. That
/// precision is what makes clip-based selection drawing possible; see [`SelectionSpan`].
pub(super) fn selection_geometry(
&self,
selection_range: Range<usize>,
visible_band: &Range<PhysicalLength>,
) -> SelectionSpans {
let mut spans = Vec::new();
for (paragraph_index, paragraph) in self.visible_paragraphs().iter().enumerate() {
// Like the draw cull, padded by an (average) line height for ink that overhangs its
// line box: such a line still draws, so it still needs its spans.
let paragraph_top = self.y_offset + paragraph.y;
let paragraph_height = PhysicalLength::new(paragraph.layout.height());
let line_pad = paragraph_height / paragraph.layout.lines().len().max(1) as f32;
if paragraph_top + paragraph_height + line_pad < visible_band.start
|| paragraph_top - line_pad > visible_band.end
{
continue;
}
let selection_start = selection_range.start.max(paragraph.range.start);
let selection_end = selection_range.end.min(paragraph.range.end);
if selection_start >= selection_end {
continue;
}
let local_start = selection_start - paragraph.range.start;
let local_end = selection_end - paragraph.range.start;
let selection = parley::editing::Selection::new(
parley::editing::Cursor::from_byte_index(
&paragraph.layout,
local_start,
Default::default(),
),
parley::editing::Cursor::from_byte_index(
&paragraph.layout,
local_end,
Default::default(),
),
);
selection.geometry_with(&paragraph.layout, |rect, line| {
// Snap the horizontal edges to device pixels once, here, so that the highlight
// rectangle and the glyph clip derived from the same span are pixel-identical.
let x = (rect.x0 as f32).round()..(rect.x1 as f32).round();
if x.end <= x.start {
return;
}
// A giant wrapped paragraph passes the paragraph test above with all its lines;
// keep only the ones that can reach the band.
let top = PhysicalLength::new(rect.y0 as _) + paragraph_top;
let bottom = PhysicalLength::new(rect.y1 as _) + paragraph_top;
if bottom + line_pad < visible_band.start || top - line_pad > visible_band.end {
return;
}
let background = PhysicalRect::new(
PhysicalPoint::from_lengths(
PhysicalLength::new(x.start),
PhysicalLength::new(rect.y0 as _) + paragraph_top,
),
PhysicalSize::new(x.end - x.start, rect.height() as _),
);
spans.push(SelectionSpan { paragraph: paragraph_index, line, background });
});
}
// Already in this order: paragraphs are visited in order, `geometry_with` walks a
// paragraph's lines in order, and within a line it accumulates x left to right over
// visually reordered items -- so even a bidi line yields ascending spans. Sort defensively
// anyway, since both consumers depend on it and neither would fail loudly: `for_line`
// needs the `(paragraph, line)` grouping, and the segment walk in
// `draw_glyph_run_with_selection` needs ascending x within a line.
spans.sort_by(|a, b| {
(a.paragraph, a.line)
.cmp(&(b.paragraph, b.line))
.then_with(|| a.x().start.total_cmp(&b.x().start))
});
SelectionSpans(spans)
}
}