Compare commits

..
Author SHA1 Message Date
iris-ai 25e456e0b5 Say what the fuzzers can no longer tell about the stale-answer guard
Dropping `dirty_size_under` from it now passes every run there is. It stays
for the one entry the bottom-up ordering does not reach -- `update` draws
the root for a resize before `redraw_updates` runs -- which is a hole
reasoned rather than measured, and the note says which.
2026-09-17 05:12:44 -04:00
iris-ai 53b00c68e9 Find a span's leftover boundary through the inverse it already has
The decision used a rounded division, `total.px.div(fixed)`, where the room
the children get is a floored multiply, so the boundary and the drawing it
guards were two expressions for one length and disagreed at the edge of it.
`room` is that length as a `Len`, `room.to_px` is the multiply, and
`Holds::through` is its exact preimage -- so ask `room` whether anything is
left and hand the answer back through the same expression.

The three branches go with the division. They were the sign of `1 - rel`:
the fixed parts growing slower than the box, faster, or exactly with it, and
`through` reads that sign already. Forty lines become twelve, one `div`
leaves layout, and the boundary is the drawing's own.

Green on the suite, the shrinker at 400 seeds of depth 5, the oracle at 1000
seeds of depth 6 and 120 in debug, and 2000 seeds at depth 4 over all
fifteen cases. `tabs`, `view`, `minimal` and `random` byte-identical.
2026-09-17 05:02:45 -04:00
6 changed files with 51 additions and 100 deletions

No files matched your search

+10 -12
View File
@@ -168,20 +168,10 @@ impl<'a> Painter<'a> {
let region_node = self.rsc.widgets().is_region_node(id.id());
let declared = self.declared_lens(id);
let align = self.rsc.widgets().alignment(id.id());
// A rule this box was already chosen from is not resolved into it a
// second time. The box is that rule's length already, so resolving
// it again takes the fraction twice -- a widget declaring half of a
// stack, in the stack its own answer made half a row, is a quarter
// of the row. Pixels survive it, being the same length wherever they
// are taken from, which is why only a share ever shrank.
let resolve = AXES.map(|axis| match decided[axis as usize] {
true => None,
false => declared[axis as usize],
});
// Composing `FULL` through a box is not quite the identity in f32,
// so a child with nothing declared keeps the box it would have had.
let local = match resolve.iter().any(Option::is_some) {
true => declared_box(region, resolve, align),
let local = match declared.iter().any(Option::is_some) {
true => declared_box(region, declared, align),
false => region,
};
let within = match local == UiRegion::FULL {
@@ -403,6 +393,14 @@ impl<'a> Painter<'a> {
.is_some()
}
/// The part of this widget's box that something of `size` takes, at the
/// near edge. A container that reports one child's size gives every child
/// this, so what it draws is inside what it says it occupies.
pub fn box_of(&self, size: Size) -> UiRegion {
let lens = placed_lens(size, [None; 2], [false; 2]);
placed_box(UiRegion::FULL, lens, RegionAlign::NEAR)
}
/// This widget's box in pixels. Reading it makes the drawing one that
/// holds for this box only, until `holds` says how far it goes.
pub fn px_size(&mut self) -> PxVec2 {
+8
View File
@@ -202,6 +202,14 @@ impl UiRenderState {
// that marks a reader is in `redraw`, which is not what asked here.
// Both retained routes are an answer, so the question is asked once
// rather than by each of them.
//
// Since `a92c6ac` settles a frame strictly bottom-up, no fuzzer can
// tell whether the second half of this still does anything: dropping
// `dirty_size_under` passes the suite, the shrinker at 400 seeds of
// depth 5, the oracle at 1000 of depth 6 and 2000 seeds at depth 4.
// It stays because `update` draws the root for a resize before
// `redraw_updates` runs at all, which that ordering does not reach --
// a hole that is reasoned rather than measured.
let stale =
rsc.widgets().needs_redraw.contains(&id) || self.dirty_size_under(id, rsc.widgets());
let replace_answer = self.answer_invalid.remove(&id) || (self.replace_answers && stale);
-1
View File
@@ -26,7 +26,6 @@ impl DefaultAppState for State {
.wrap(true)
.text_align(Align::LEFT)
.pad(16)
.width(rel(1.0))
.background(panel());
// Each one takes the whole width, because `text_align` puts the
+16 -34
View File
@@ -46,43 +46,26 @@ impl Widget for Span {
|sum, len| sum + *len,
);
// What is left for the shares to divide: the box less everything
// fixed, as a length of the box rather than a number of pixels.
let room = Len::rel_max() - Len::from_parts(total.rel, total.px);
// Whether anything is left over is a question in pixels: `rel(0.5)`
// beside 300 px is full at 600 and overfull at 400. The room to
// divide is `len * fixed - total.px`, and the length where it runs
// out is exactly the box a parent sizing itself from this answer
// hands back -- which is why this used to need a margin either side
// of the boundary, and why it does not now: that box and this sum are
// whole counts of the same step, and both routes to it land on the
// same count. What the generated oracle checks is the consequence,
// since which children exist at all turns on this.
let fixed = Rel::ONE - total.rel;
// beside 300 px is full at 600 and overfull at 400. Asked of `room`
// itself, and answered back through the same expression, so the
// boundary is the drawing's own and not a second way of finding it:
// the three cases a rounded division needed -- the fixed parts
// growing slower than the box, faster, or exactly with it -- are the
// sign of `room.rel`, which `through` already reads. What the
// generated oracle checks is the consequence, since which children
// exist at all turns on this.
let mut shares = false;
if total.leftover > Weight::ZERO {
let current = painter.px_len(axis);
let holds = if fixed > Rel::ZERO {
// The box length the fixed parts alone fill.
let full = total.px.div(fixed);
shares = current > full;
match shares {
true => Holds::from(full.next_up()..=Px::MAX),
false => Holds::from(Px::MIN..=full),
}
} else if fixed < Rel::ZERO {
// The relative parts grow faster than the box does, so here
// a shorter box is the one that leaves room.
let full = total.px.div(fixed);
shares = current < full;
match shares {
true => Holds::from(Px::MIN..=full.next_down()),
false => Holds::from(full..=Px::MAX),
}
} else {
// The relative parts take exactly the box, whatever it is, so
// the only room is what negative pixels leave.
shares = total.px < Px::ZERO;
Holds::ANY
shares = room.to_px(painter.px_len(axis)) > Px::ZERO;
let holds = match shares {
true => Holds::from(Px::STEP..=Px::MAX),
false => Holds::from(Px::MIN..=Px::ZERO),
};
painter.holds(axis, holds);
painter.holds(axis, holds.through(room));
}
// Across itself a span is as long as its longest child -- unless a
@@ -99,7 +82,6 @@ impl Widget for Span {
// row.
let mut fixed = Len::rel_min();
let mut taken = Weight::ZERO;
let room = Len::rel_max() - Len::from_parts(total.rel, total.px);
let mut start = Len::rel_min();
let mut ortho = LayoutLen::ZERO;
for (child, len) in self.children.iter().zip(&lens) {
+17 -28
View File
@@ -13,42 +13,31 @@ impl Widget for Stack {
StackSize::Default => None,
StackSize::Child(i) => Some(i),
};
// Every child gets the whole of this stack's box, the sizing one
// included, and the stack is then handed a box of the length that
// child asked for. Not the part of the box that length takes: the
// stack's own box becomes that length, and taking the fraction of it
// again is the fraction twice -- a child asking for half of a stack
// that is already half a row would have a quarter of the row.
//
// It cannot be told apart by asking whether this box is the answer
// yet, either. A drawing has to be a function of the box alone, since
// moving the stack into the box it asked for reuses the drawing by
// scaling it, and a drawing made a fraction of one box is right in
// any other. So: fractions of this box throughout, and the move is
// the whole of the difference.
let region = UiRegion::FULL;
// Whichever child sizes the stack is asked here and not again below,
// on the layer it ends up on: a retained drawing belongs to the layer
// it was made on, so measuring it anywhere else costs a second
// drawing of it. Its box is its own answer, so the answer is not
// placed inside it again.
// Whichever child sizes the stack decides the box every child gets.
// The stack reports that size, so a child given a longer box would
// draw outside what the stack says it occupies.
let size = match sizing.and_then(|i| self.children.get(i).map(|c| (i, c))) {
// On the layer that child ends up on, so the ask below is a reuse
// rather than a second drawing of it somewhere else: a retained
// drawing belongs to the layer it was made on.
Some((i, child)) => {
painter.child_layer_at(i);
painter
.widget_at(child, region, region.size(), [true; 2])
.size()
painter.widget(child).size()
}
None => Size::LEFTOVER,
};
let region = painter.box_of(size);
for (i, child) in self.children.iter().enumerate() {
if sizing == Some(i) {
continue;
}
painter.child_layer_at(i);
// A box that owes nothing to this child's own answer: where it
// sits in one bigger than itself is its own business.
painter.widget_within(child, region);
// The sizing child placed its own content in the box its answer
// decided, and this box was derived from that answer, so applying
// its alignment again here would place it twice. Every other
// child is handed a box that owes nothing to its own answer, and
// where it sits in one bigger than itself is its own business.
match sizing == Some(i) {
true => painter.widget_at(child, region, region.size(), [true; 2]),
false => painter.widget_within(child, region),
};
}
size
}
-25
View File
@@ -82,31 +82,6 @@ fn a_text_in_a_span_wraps_at_the_room_left_rather_than_the_whole_row() {
assert!(crowded > whole_row, "{crowded} against {whole_row}");
}
/// A stack takes its size from one child and gives every child that size, so
/// a child asking for half of it is asking for half of what it is itself the
/// size of. Once the stack has been placed at the length it reported that
/// length is the box, and taking the fraction of it again takes it twice:
/// half a row became a quarter, and a further stack around it a further half.
/// Nothing pinned it because a pixel is the same length wherever it is taken
/// from, so only a share ever shrank -- and warm and cold shrink alike, so no
/// oracle saw it either.
#[test]
fn a_stack_sized_by_a_child_does_not_take_that_childs_fraction_twice() {
let mut h = Harness::new((400, 200));
let half = rect(Color::RED).width(rel(0.5)).add(&mut h.rsc);
let behind = rect(Color::BLUE).add(&mut h.rsc);
let stack = Stack {
children: vec![behind.add_strong(&mut h.rsc), half.add_strong(&mut h.rsc)],
size: StackSize::Child(1),
}
.add(&mut h.rsc);
h.set_root((stack,).span(Dir::RIGHT).width(rel(1.0)));
assert_corners!(h, stack, (0, 0), (200, 200));
assert_corners!(h, half, (0, 0), (200, 200));
assert_corners!(h, behind, (0, 0), (200, 200));
}
/// The same reading through a pad: its inset is the whole box less the
/// padding, so half of the inset plus the padding is half the box plus one
/// padding, not two.