Compare commits

..
2 Commits
Author SHA1 Message Date
iris-aiandClaude Opus 5 2dba90bd0f Grow images in the generated trees
`Image` is the only widget in the repository whose size hint is a length in
pixels -- everything else hints a share, or nothing -- so it is the only one
that exercises a rule beside a hint, a box a widget knows before it is drawn,
and the answer the commit before this one changed. The generated trees had
none, which is why nothing there could reach that case.

`Kind::Image` is a fifth leaf, drawn one time in five, and it steps to a plain
rect when the shrinker reduces it: a picture measures nothing either, but its
length is its own, so the leaf that takes whatever it is given is the simpler
one. The picture is a 64x64 checkerboard of purple and black in 8 px cells,
committed at `src/assets/checkerboard.png` beside the generator that draws it
-- the way `examples/tabs` keeps its own -- and included rather than opened, so
that growing a tree does not depend on a working directory and one seed is one
tree whatever anything else does.

One upload per tree, however many images it grows: a `TextureHandle` is a
counted reference, so the first image in a tree uploads the checkerboard and
every one after it clones the handle. Measured: seed 1 at depth 4 grows 13
images and holds 1 texture, seed 6 grows none and holds none, and
`a_tree_of_images_uploads_one_texture` asserts it. `Image::new` is what a
caller holding a handle needs, since `image` uploads what it is given.

A seed names a tree only while the generator draws the same things in the same
order, so every seed now grows a different tree. The seed list in
`generated.rs` says so: 20 and 86 no longer grow the trees whose defects they
once caught, and both of those live on as shrunk fixtures in `unsettled.rs`,
which are trees rather than numbers. The seeds those fixtures name are
similarly historical, and their file says that too.

Format, clippy with and without layout-diagnostics, and the suite (135 + 19 +
13 + 4) are clean. The cold dump is a new baseline of 34,571 boxes over the 400
depth-5 trees, since the trees themselves changed; all three seed scans pass
over the new ones -- 400 at depth 5 in 62.79s, 1,000 at depth 6 in 160.20s,
2,000 at depth 4 in 299.58s -- which is what actually checks that images lay
out warm the way they do cold.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 04:50:31 -04:00
iris-aiandClaude Opus 5 b295c8b97a Read a leftover as a minimum where nothing divides it
A share under a parent that divides nothing is still a share: the pixels and
fraction beside it are taken first, the share fills whatever the box has left,
and where those are already longer than the box they overflow it exactly as
they would without the share. So the length is `max(box, px + rel*box)`, a
minimum the share imposes rather than an addition to what was asked for
(Bryan, 2026-09-20, generalising the same `max` he gave for `Scroll`'s content
length two days earlier).

A span does that. Measured at `77ed7a2`, a probe recording the box it is asked
in, in a 400 px window, under `.wrapper()` against a one-child span:

    rule                      nothing divides   a span divides
    leftover(1)                           400              400
    px(50) + leftover(1)                  400              400
    px(500) + leftover(1)                 400              500
    rel(0.5) + leftover(1)                400              400
    px(500), no leftover                  500              500

One row disagreed, and the same length without the share overflows fine
(drawn -50..450, its alignment centring it), so what swallowed the overflow was
the share. `LayoutLen::declared` refuses to answer for anything carrying
leftover weight, so the non-dividing path never learned the fixed part and fell
back to the offer.

Said as the place the parent gives rather than as a declaration, because that
is what the retained record already keeps: where the fixed part is the longer,
`widget_at` hands the child `fixed.as_desc().fills()` -- a box of that length,
placed by the child's alignment, its own rel base -- which is what a declared
length already comes to, and `active.placed` stores it, so a recomposed subtree
reads the same box without resolving anything again. A place that is already
the child's placement is skipped: a parent that divides has given the share
whatever it was owed, and re-placing a span's slot moved its child.

Which of two lengths is longer is a question in pixels, so it is one operation
with the crossing kept as a window range, and both callers now share it.
`Painter::longer_than` is that operation -- the span's room for the shares it
divides, and a share past the box it was given -- and it narrows this widget's
range where the span replaced it, since a comparison the framework makes on an
arbitrary parent's behalf is one more reason its drawing holds, not the only
one. A `SizeRule::Min` of `rel(1.0)` is the same operation again, which is what
this is (Bryan, 2026-09-20); when that lands it belongs on this path.

`a_share_is_a_minimum_wherever_nothing_divides_it` walks the table above and
holds the two parents to the same length; the crossing case is checked from
both sides, by a window that crosses it and by the rule itself crossing while
the window holds still. Both fail at `77ed7a2` with 400 where 500 is wanted. A
change of rule needs nothing to escalate it: the reported size is the rule
resolved, so the answer changes and the parent refuses its own drawing --
verified by writing the escalation, finding the tests pass without it, and
dropping it.

Format, clippy with and without layout-diagnostics, and the suite (134 + 19 +
13 + 4) are clean. The cold dump over 400 depth-5 trees is byte-identical to
`77ed7a2` across all 34,488 boxes, since no generated tree carries a share with
pixels beside it -- which the next commit changes.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-20 04:49:50 -04:00
10 changed files with 236 additions and 34 deletions

No files matched your search

+86 -6
View File
@@ -187,10 +187,20 @@ impl<'a> Painter<'a> {
id: &'s StrongWidget<W>,
place: impl Into<PlaceDesc>,
) -> DrawResult<'s, 'a, W> {
let place = self.resolve_rel_base(place.into());
let mut place = self.resolve_rel_base(place.into());
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 share fills what the pixels and fraction beside it leave of the
// box and overflows where they are longer, which is the rule a span
// follows with one child. Only the overflow is a box of the child's
// own: a share that fits is the box it was given, which is what this
// place already says.
for axis in Axis::BOTH {
if let Some(len) = self.share_past_the_offer(id.id(), place, align, axis) {
place[axis] = len.as_desc().fills();
}
}
let declared = self.declared_lens(id);
let (rel_base, region) =
place.rel_base_and_region(self.region, self.rel_base, declared, align);
#[cfg(feature = "layout-diagnostics")]
@@ -299,6 +309,45 @@ impl<'a> Painter<'a> {
self.rsc.widgets().declared_lens(id.id())
}
/// The box a child's own share asks for where that is longer than the box
/// `place` gives it, and nothing where the share fits.
///
/// A share is a length only to whoever divides one, and nothing divides a
/// box handed to one child: what is left of it after the pixels and the
/// fraction beside the share is what the share takes, so the length comes
/// to the whole box until those are longer than it and to them once they
/// are. Only that second case is a box this widget did not give, and the
/// crossing between them is a question in pixels, so this widget's drawing
/// holds for the windows on one side of it. Narrowed rather than stated,
/// because this widget may have read its own box as well, and a range it
/// pinned for that still holds.
fn share_past_the_offer(
&mut self,
id: WidgetId,
place: PlaceDesc,
align: RegionAlign,
axis: Axis,
) -> Option<Len> {
// A place that is the child's placement outright is a box its parent
// decided, and a parent that divides one has already given the share
// whatever it was owed. Only an offer -- a box with the answer still
// to be placed inside it -- is a box a share reads.
if place[axis].fills {
return None;
}
// A share with nothing beside it is the box whatever the box is, so
// there is no comparison to make and no range to keep for one.
let stated = self.rsc.widgets().exact_len(id, axis)?;
if stated.leftover == Weight::ZERO || stated.is_only_leftover() {
return None;
}
let fixed = stated
.without_leftover()
.within_len(place.base(axis, self.rel_base));
let offer = place.of(self.region, align)[axis].len();
self.longer_than(fixed, offer, axis).then_some(fixed)
}
/// What a child says its length is without being drawn, if it can say,
/// as the length its draw would report: a fraction in it is resolved
/// against this widget's rel base, which is the rel base a child asked with
@@ -475,6 +524,40 @@ impl<'a> Painter<'a> {
len.to_px(window)
}
/// Whether `len` is longer than `than`, kept as the windows that comparison
/// comes out the same way on: a drawing that took one of two lengths holds
/// where the same one is the longer, and nowhere else.
///
/// Which is longer is a question in pixels -- `rel(0.5)` is longer than 300
/// px at a box of 600 and shorter at 400 -- and it is asked of the
/// difference and answered back through that same difference, so the
/// boundary is the drawing's own rather than a second way of finding it.
/// Narrowed rather than stated, because whatever else this widget read
/// about the window is a reason its drawing holds where it does too.
///
/// This is the one operation a length that is the longer of two needs: the
/// room a container has left for the shares it divides, and a share that
/// overflows the box it was given because the pixels beside it are longer
/// than the box.
pub fn longer_than(&mut self, len: Len, than: Len, axis: Axis) -> bool {
let over = len - than;
let window = self.window[axis];
let longer = over.to_px(window) > Px::ZERO;
let side = match longer {
true => Px::STEP..=Px::MAX,
false => Px::MIN..=Px::ZERO,
};
let holds = Holds::from(side).through(over);
debug_assert!(
holds.contains(window),
"'{}' ({:?}) compared two lengths and kept a range without this window",
self.label(),
self.id
);
self.own[axis].window = self.own[axis].window.and(holds);
longer
}
/// The windows this drawing holds for, stated rather than taken: a
/// container that branched on a length in pixels says which side of the
/// boundary it was on, which is wider than the one window reading that
@@ -725,10 +808,7 @@ impl PlaceDesc {
let mut rel_base = parent_rel_base;
let mut region = given;
for axis in Axis::BOTH {
let base = match self[axis].rel_base {
RelBase::Len(len) => len,
RelBase::Inherit | RelBase::WithRegion => parent_rel_base[axis],
};
let base = self.base(axis, parent_rel_base);
let len = declared[axis]
.map(|len| len.within_len(base))
.unwrap_or(base);
+11 -1
View File
@@ -1,5 +1,5 @@
use crate::util::impl_axis_index;
use crate::{Axis, AxisAlign, Len, PrimitiveHandle, RegionAlign, UiRegion, UiSpan};
use crate::{Axis, AxisAlign, Len, PrimitiveHandle, RegionAlign, UiRegion, UiSpan, UiVec2};
/// How a child's region along one axis comes from the region of the widget
/// asking, and what its fractions are of.
@@ -123,6 +123,16 @@ impl PlaceDesc {
self
}
/// What a child's fractions on one axis are of, as a length of the
/// window: a length this place names, or the rel base of the widget
/// giving it, which is `parent_rel_base`.
pub(super) fn base(&self, axis: Axis, parent_rel_base: UiVec2) -> Len {
match self[axis].rel_base {
RelBase::Len(len) => len,
RelBase::Inherit | RelBase::WithRegion => parent_rel_base[axis],
}
}
/// The box each axis names, in the coordinates `own` is in.
pub fn of(self, own: UiRegion, align: RegionAlign) -> UiRegion {
UiRegion::new(self.x.of(own.x, align.x), self.y.of(own.y, align.y))
Binary file not shown.

After

Width:  |  Height:  |  Size: 191 B

+29 -3
View File
@@ -190,6 +190,9 @@ pub enum Kind {
color: usize,
alpha: u8,
},
/// The one leaf whose own length is a number of pixels it knows before it
/// is drawn, which is the hint a rule beside it has to win over.
Image,
/// Scrolling reads the pixel length of its box, which nothing else here
/// does, and gives its child a box longer than its own.
Scroll {
@@ -443,9 +446,11 @@ impl Kind {
}
match self {
// The one leaf that reads the width it is given, then the one
// that does not, then the one that measures nothing at all.
// that does not, then the one that measures nothing at all. A
// picture measures nothing either, but its length is its own, so
// it steps to the leaf that takes whatever it is given.
Kind::Wrapped => out.push(Kind::OneLine),
Kind::OneLine => out.push(Kind::Rect {
Kind::OneLine | Kind::Image => out.push(Kind::Rect {
color: 0,
alpha: 255,
}),
@@ -627,9 +632,10 @@ struct Sow<'a> {
impl Sow<'_> {
fn leaf(&mut self) -> Plan {
Plan::bare(match self.rng.below(4) {
Plan::bare(match self.rng.below(5) {
0 => Kind::Wrapped,
1 => Kind::OneLine,
2 => Kind::Image,
_ => {
let color = self.rng.below(COLORS.len());
let alpha = (self.rng.below(5) * 63) as u8;
@@ -793,6 +799,7 @@ pub fn build<Rsc: UiRsc + 'static>(rsc: &mut Rsc, plan: &Plan) -> (StrongWidget,
let mut build = Build {
rsc,
tree: Tree::default(),
checkerboard: None,
};
let root = build.node(plan);
(root, build.tree)
@@ -801,6 +808,10 @@ pub fn build<Rsc: UiRsc + 'static>(rsc: &mut Rsc, plan: &Plan) -> (StrongWidget,
struct Build<'a, Rsc> {
rsc: &'a mut Rsc,
tree: Tree,
/// The checkerboard, uploaded when the first image in this tree is built.
/// A handle is a reference to the texture, so every image after that one
/// clones this rather than uploading the same picture again.
checkerboard: Option<TextureHandle>,
}
impl<Rsc: UiRsc + 'static> Build<'_, Rsc> {
@@ -826,6 +837,20 @@ impl<Rsc: UiRsc + 'static> Build<'_, Rsc> {
built
}
/// The one picture the generated trees draw: a 64x64 checkerboard of purple
/// and black in 8 px cells. Committed rather than drawn here, so that one
/// seed is one tree whatever anything else does, and included rather than
/// opened, so that growing a tree does not depend on a working directory.
fn checkerboard(&mut self) -> TextureHandle {
if self.checkerboard.is_none() {
let image = include_bytes!("assets/checkerboard.png")
.get_image()
.expect("the checkerboard is committed beside this file");
self.checkerboard = Some(self.rsc.ui_mut().textures.add(image));
}
self.checkerboard.clone().unwrap()
}
fn kind(&mut self, kind: &Kind) -> StrongWidget {
let id: StrongWidget = match kind {
Kind::Wrapped => wtext(WORDS).size(16).wrap(true).add_strong(self.rsc),
@@ -834,6 +859,7 @@ impl<Rsc: UiRsc + 'static> Build<'_, Rsc> {
.wrap(false)
.add_strong(self.rsc),
Kind::Rect { color, alpha } => rect(COLORS[*color].alpha(*alpha)).add_strong(self.rsc),
Kind::Image => Image::new(self.checkerboard()).add_strong(self.rsc),
Kind::Scroll { axis, inner } => {
let inner = self.node(inner);
let id = Scroll::new(inner, *axis).add(self.rsc);
+9
View File
@@ -16,6 +16,15 @@ impl Widget for Image {
}
}
impl Image {
/// One texture already uploaded, for a caller holding its handle: [`image`]
/// uploads what it is given, and several widgets showing one picture want
/// one upload and one slot between them.
pub fn new(handle: TextureHandle) -> Self {
Self { handle }
}
}
pub fn image<State: UiRsc>(image: impl LoadableImage) -> impl WidgetFn<State, Image> {
let image = image.get_image().expect("Failed to load image");
move |state| Image {
+7 -17
View File
@@ -52,25 +52,15 @@ impl Widget for Span {
// What is left for the shares to divide: the row less everything
// fixed, as a length of the rel base rather than a number of pixels.
let room = row - total.without_leftover();
// Whether anything is left over is a question in pixels: `rel(0.5)`
// 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
let all_fixed = total.without_leftover();
let room = row - all_fixed;
// 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.
// sign of `room.rel`, which the range `longer_than` keeps already
// reads. What the generated oracle checks is the consequence, since
// which children exist at all turns on this.
let any_leftover = total.leftover > Weight::ZERO;
let has_room = any_leftover && painter.to_px(room, axis) > Px::ZERO;
if any_leftover {
let holds = match has_room {
true => Holds::from(Px::STEP..=Px::MAX),
false => Holds::from(Px::MIN..=Px::ZERO),
};
painter.window_holds(axis, holds.through(room));
}
let has_room = any_leftover && painter.longer_than(row, all_fixed, axis);
// Across itself a span is as long as its longest child -- unless a
// rule beside it gives that length outright, and then reading them
+60
View File
@@ -260,6 +260,66 @@ fn a_share_rule_beats_the_widgets_own_pixel_size() {
assert_eq!(asked.get(), 400.0, "the share is all of the box");
}
/// A share with pixels or a fraction beside it is the longer of the two: it
/// fills what they leave of the box and overflows the box where they are
/// longer than it. A parent that divides nothing gives the same length as a
/// span with one child, because in both there is nobody else to divide with.
#[test]
fn a_share_is_a_minimum_wherever_nothing_divides_it() {
let asked = |rule: LayoutLen, in_a_span: bool| {
let mut h = Harness::new((400, 200));
let probe = rect(Color::RED).add(&mut h.rsc);
h.set_len(probe, Axis::X, rule);
match in_a_span {
true => h.set_root((probe,).span(Dir::RIGHT)),
false => h.set_root(probe.wrapper()),
}
h.region(&probe).unwrap().size().x
};
for (rule, want) in [
(LayoutLen::LEFTOVER, 400),
(LayoutLen::px(50) + LayoutLen::LEFTOVER, 400),
(LayoutLen::px(500) + LayoutLen::LEFTOVER, 500),
(LayoutLen::rel(0.5) + LayoutLen::LEFTOVER, 400),
(LayoutLen::px(500), 500),
] {
let want = Px::from_int(want);
assert_eq!(asked(rule, false), want, "{rule:?} where nothing divides");
assert_eq!(asked(rule, true), want, "{rule:?} in a span");
}
}
/// Which of the two is longer is a question in pixels, so the box is decided
/// again wherever the answer can change: a window that crosses the length the
/// pixels ask for, and the rule itself crossing it while the window holds
/// still. The first is a range the drawing holds for; the second cannot be
/// seen in what the widget declares, since a share declares nothing either
/// way, so it reaches the parent as a length only the parent can resolve.
#[test]
fn a_share_past_the_box_is_decided_again_on_either_side_of_the_crossing() {
let mut h = Harness::new((400, 200));
let probe = rect(Color::RED).add(&mut h.rsc);
h.set_len(probe, Axis::X, LayoutLen::px(500) + LayoutLen::LEFTOVER);
h.set_root(probe.wrapper());
assert_eq!(h.region(&probe).unwrap().size().x, Px::from_int(500));
h.resize((900, 200));
h.frame();
assert_eq!(h.region(&probe).unwrap().size().x, Px::from_int(900));
h.resize((400, 200));
h.frame();
assert_eq!(h.region(&probe).unwrap().size().x, Px::from_int(500));
h.set_len(probe, Axis::X, LayoutLen::px(50) + LayoutLen::LEFTOVER);
h.frame();
assert_eq!(h.region(&probe).unwrap().size().x, Px::from_int(400));
h.set_len(probe, Axis::X, LayoutLen::px(500) + LayoutLen::LEFTOVER);
h.frame();
assert_eq!(h.region(&probe).unwrap().size().x, Px::from_int(500));
}
#[test]
fn a_child_drawn_twice_moves_once() {
let mut h = Harness::new((400, 200));
+20 -3
View File
@@ -5,7 +5,9 @@
//! and the oracle another. And reducing a plan has to end, or a shrinker
//! searching for the smallest counterexample never returns.
use iris::random::{Edits, Kind, Plan, Rng, SpanEdit, plan};
use iris::harness::Harness;
use iris::prelude::*;
use iris::random::{Edits, Kind, Plan, Rng, SpanEdit, grow, plan};
use std::collections::HashMap;
fn some_edits(seed: u64, of: &Plan) -> Edits {
@@ -67,8 +69,6 @@ fn some_edits(seed: u64, of: &Plan) -> Edits {
}
}
use iris::prelude::*;
/// The two routes to an edited tree are one tree. `plan` resolves edits out
/// of the random stream as it draws; `edited` puts them on a tree that
/// already exists, which is the only route a shrunk plan has, since no seed
@@ -137,3 +137,20 @@ fn reducing_a_plan_all_the_way_ends() {
);
}
}
/// Every image in a tree is the same picture, and a handle is a reference to
/// the texture rather than a copy of it, so one upload and one slot serve all
/// of them however many a tree grows -- and the trees are grown in hundreds.
#[test]
fn a_tree_of_images_uploads_one_texture() {
let mut images = 0;
let mut tree = plan(1, 4, &Edits::default());
tree.walk_mut(&mut |p| images += (p.kind == Kind::Image) as usize);
assert!(images > 1, "a tree of {images} images tests nothing");
let mut h = Harness::new((900, 1200));
let (root, _) = grow(&mut h.rsc, 1, 4, &Edits::default());
h.state.root = Some(root);
h.frame();
assert_eq!(h.rsc.ui().textures.count(), 1);
}
+6
View File
@@ -9,6 +9,12 @@
//! reached through a region node's own entry rather than through the offer
//! that node was given. The last is a wrapping text handed back the width
//! it measured, rounded to a step below the line it measured there.
//!
//! Each says which seed it was shrunk from, of the generator as it stood when
//! it was found. Those numbers no longer grow those trees -- a seed names one
//! only while the generator draws the same things in the same order, and the
//! leaves have grown an image since -- so what is written out below is the
//! record of the case, and the seed is where it came from.
use std::collections::HashSet;
+8 -4
View File
@@ -25,10 +25,14 @@ fn depth() -> usize {
env("IRIS_GENERATED_DEPTH", 4)
}
/// The seeds the ordinary tests take. Eight that have never failed; 86,
/// which a `Scroll` fixed point once settled differently on; and 20, which
/// caught a locally redrawn widget being placed twice in the box its parent
/// had already placed it in.
/// The seeds the ordinary tests take: a corpus rather than a set of
/// regression cases, since a seed names a tree only for as long as the
/// generator draws the same things in the same order. Adding images to the
/// leaves moved every one of them, so 20 and 86 -- which once caught a widget
/// placed twice in a box its parent had already placed it in, and a `Scroll`
/// fixed point settling differently -- no longer grow those trees. Both
/// defects are pinned by the shrunk fixtures in `cases/unsettled.rs`, which
/// are trees rather than numbers.
const SEEDS: [u64; 10] = [1, 2, 3, 5, 8, 10, 13, 20, 86, 98];
fn check(seed: u64, depth: usize, case: Case) {