Record #18's three agreed changes, the measurement, and the wrapping-text divergence the random-tree rig found

docs/IRIS_EXTRACTION_HANDOFF.md

@@ -9,40 +9,76 @@ Canonical `main` is **`ca2b4b2`** (#17, the headless rig). Sixteen slices are
 in; #16's size work and #17's rig both merged on 2026-09-14.
 
 **#18 `split/18-position-chain`** is open, worktree `/home/bob/repos/iris-pr18`,
-head `db1751f`, five commits. LAYOUT.md §2's O(1) subtree movement, plus the
-`Remap` retirement the owner asked for on top of it.
+head `86a7e8d`, eight commits. LAYOUT.md §2's O(1) subtree movement, the
+`Remap` retirement, and then the three changes the owner chose on 2026-09-14:
+the slot carries a box, slots are opt-in, and a widget's region is held in the
+coordinates of the slot it draws in.
 
-- Every active widget owns a slot in `UiRenderState::moves` -- a translation in
-  physical pixels and the slot it is relative to. A primitive instance and a
-  mask each name one, and `prelude.wgsl` walks the chain and adds the
-  accumulated delta. A mask resolves its own chain rather than the drawn
-  primitive's, so a stationary viewport can clip content moving inside it.
-- `try_reuse`'s translation case writes one slot instead of remapping a
-  subtree: 100 primitive region writes to 0 on a span of 20 rows five
-  primitives deep.
-- `window_region` walks the same chain on the CPU, so hit testing and anyone
-  asking in window pixels agree with the shader. `Moves::resolve` stops at
-  `CHAIN_LIMIT` as the shader does, and debug-asserts that it reached the end.
-- `Vec2` is `repr(align(8))`, WGSL's alignment for a `vec2<f32>`, so a GPU
-  struct holding one is laid out the way its shader reads it without saying so
-  itself. `GlyphPrimitive` no longer states its own alignment; both it and
-  `MoveOffset` keep a manual `unsafe impl Pod`, because the trailing padding
-  that alignment introduces is what `derive(Pod)` refuses. **No manual padding
-  fields** -- the owner rejected one on 2026-09-14.
-- `Remap` is gone, with `UiScalar::outside`, `UiSpan::outside` and
-  `LerpUtil::lerp_inv`. A translation never needed an inversion: shifting a box
-  shifts everything composed into it by the same amount, since
-  `lerp(s + d, e + d, t) == lerp(s, e, t) + d` on both channels, whatever the
-  box's relative extent. Only a change of length needs each part's fraction
-  recovered, which is `UiRegion::stretch` behind `UiRegion::stretchable`. The
-  decision is made once before the walk and neither relocation method
-  branches, which is how the owner asked for it.
+- **A slot holds a box, not a translation** (`1f9dc48`), given in the
+  coordinates of the slot it names, and `prelude.wgsl` composes the chain with
+  `within` instead of adding a delta. A translation is the special case where
+  the box has its parent's relative extent. Measured against the translate
+  slot on the same binary: +0.5-0.8% at depth 1-4, +9.6% at 8, +32.2% at 64.
+  Free where opt-in slots put the chain. The identity is `UiRegion::FULL`, not
+  zero -- a zeroed entry collapses its subtree to a point.
+- **`Painter::place` is how a container asks for a slot** (`d989691`): it
+  draws a child it decides the box of and may decide again. `Span`, `Aligned`
+  and `Scroll` use it; everything else shares its nearest ancestor's slot,
+  which `tests/layout.rs` pins at one deep with four widgets in between.
+  Placing a child again is one entry to write, moved or resized alike.
+- **Nothing is inverted any more.** `UiRegion::stretch`, `stretchable` and
+  `UiScalar::stretch` are gone, with the case they could not express: a
+  40-tall row now stretches on its other axis, which `stretchable` refused.
+- **`redraws_under` is the CPU walk that remains**, and it is a question
+  rather than a marking: before keeping a drawing, ask whether anything under
+  it would have to be drawn again for the new length, and give up the reuse if
+  so. Marking instead does not terminate -- the mark escalates to the
+  descendant's size reader, which re-places the child, which marks it again.
+  A part of a box with no relative extent holds its contents as offsets, and
+  composing into it leaves none either, so the walk stops where a length did
+  not change: an 80-wide child of a widened row is never asked.
+- **`Span`, `Pad`, `Stack`, `Offset`, `Aligned`, `SetSize` and `LayerOffset`
+  say `Scale`.** Each places in fractions and offsets of its own box and none
+  reads its pixel length. `Scroll` and `MaxSize` do read pixels and stay
+  `Redraw`. Measured as not the cause of the divergence below: taking every
+  one of these claims back out leaves the failing set identical.
+- **`OnResize::Scale` keeps its name.** The owner rejected `Stretch` on
+  2026-09-14: stretch has an opposite and scale does not, and the answer is
+  per axis so the axis is already established where it is read.
 
-**Two invariants everything here rests on.** `resolved = region +
-resolve(slot)`, so anything that rewrites a region owes that slot a zero --
-`stretch` does it for the subtree it rewrites, `draw_inner` for the widget it
-draws. And a stretch is only expressible out of a box with a relative extent;
-a fixed length holds its parts as offsets from its start and keeps no fraction.
+**The open defect, and it is older than this branch.** `tests/generated.rs`
+grows a seeded random tree and checks that laying it out again lands where
+growing it cold does. It does not always. Every case seen is a **wrapping text
+on a span's own axis**: the span measures the child in the whole box, the
+child shapes to that width and reports the width it used, the span places it
+in exactly that width, which is a length change, so it shapes again and its
+longest line is shorter than the box it was just given. Where it settles
+depends on how many passes it has had, and a warm tree has had a different
+number from a cold one -- so layout is not a function of the state, which it
+has to be. A span whose axis is not the wrap axis is stable, which is every
+real column of text, and why nothing had run into it.
+
+    seeds 1-30, three checks each      db1751f (before)   86a7e8d (now)
+    diverging                          7 of 90            30 of 90
+
+The same defect either way: a placed child reaches the second shaping more
+often now. It wants fixing where the two draws meet, LAYOUT.md §4, rather than
+anywhere in the chain -- and **it should be fixed before #18 merges**, since
+tripling how often a tree lands somewhere unrepeatable is not something to
+land even on a defect that was already there.
+`a_wrapping_child_of_a_row_settles_somewhere_else_each_time` is the ignored
+reproduction; the six seeds the live tests use are ones that agree.
+
+**Rigs added.** `iris::random` grows the tree (spans in every direction,
+stacks, rects with varying opacity, text wrapping and overflowing, a declared
+size over half of it, stopping at a depth), `examples/random.rs` draws one --
+`IRIS_SEED` and `IRIS_DEPTH` pick it -- and `tests/generated.rs` compares warm
+against cold under a size change, a resize, and both. It found the
+non-terminating marking above immediately.
+
+**Not verified yet:** the headless render shots and the live resize under
+sway. The owner asked for those last, after the CPU side is right, and the CPU
+side is not right until the divergence is.
 
 ### The plan after #18, agreed with the owner on 2026-09-14
 
@@ -91,10 +127,29 @@ the row grows. The CPU work on a resize becomes asking `on_resize` down the
 widgets and redrawing the ones that say so -- one ask for a transcript row
 against a few hundred glyph rewrites. It replaces #18's chain rather than
 extending it, and it contradicts LAYOUT.md §2's line that slots carry
-translation only. Before choosing it, run `chain_cost` with a region per slot
-in place of a `Vec2`: the per-level cost was measured as the dependent load
-rather than the arithmetic, so the extra lerps should be free, but that is an
-assumption until the rig says so.
+translation only -- a line an agent wrote into the design on 2026-09-04, not
+something the owner asked for. Her own statement of the requirement was "if it
+needs to be moved then that can be done after the fact efficiently, or resized
+just done after as well".
+
+**Measured, 2026-09-14.** `MoveOffset` widened from a `Vec2` to a `UiRegion`
+(36 bytes) and `resolve_move` composing with `within` instead of adding
+deltas, against the same binary's translate slots: depth 1 78.5 us against
+78.0, depth 2 78.7 against 78.3, depth 4 79.6 against 79.0, then depth 8 89.8
+against 81.9, depth 16 128.6 against 111.3, depth 64 331.2 against 250.6. So a
+box slot is free at the 2-4 depth opt-in slots produce (+0.5-0.8%) and costs
+10-30% past depth 8, where the chain is already expensive. All 42 tests pass
+against the composing shader and `tabs` at 1920x1200 is byte-identical,
+because composing through a translation box is the same map as adding its
+delta. The experiment is `stash@{0}` in `/home/bob/repos/iris-pr18`.
+
+The slot has to carry the whole box rather than a scale and an offset: a
+pixel-space affine map scales everything under it, including a child that must
+keep its pixel length, and the `rel`/`abs` pair is exactly what distinguishes
+the two. What a box slot does not buy is a shorter CPU walk. Asking each
+widget whether its drawing depends on the length that changed is the same walk
+either way; what the box removes is the region rewrite and the upload at the
+widgets that answer `Scale`, and the inversion at the ones that do not.
 
 What is settled either way: recalculating rather than repositioning is in, as
 the `Remap` retirement above. It costs the per-axis carry: a box that changed