fix(FlexPanel): honor VerticalAlignment.Stretch as definite block axis

[codemonkeychris]

Summary

• FlexPanel's block axis was keyed off the panel's declared Height only — the parent's offer was ignored even when VerticalAlignment.Stretch + a finite parent slot meant "fill". The canonical web flex pattern of header(auto) / body(flex:1) / footer(auto) filling a viewport required a hardcoded .Height(N), and a flex:1 body resolved to an empty pool.
• Three coordinated changes in MeasureOverride, ArrangeOverride, and the Yoga MeasureFunction wrapper fix this without a parent-type sniff and without regressing the wobble fix from fix(FlexPanel): eliminate +/-1 px height wobble on resize #172.
• New self-test Flex_HeaderBodyFooterFillsParentSlot covers the canonical pattern under a Reactor host with no explicit .Height(N) on the column.

Why three changes

1. MeasureOverride — when no explicit Height, VerticalAlignment.Stretch, and parent offer is finite, treat block axis as definite. Symmetric counterpart to the existing HorizontalAlignment.Stretch inline-axis fill — same WinUI Stretch contract on both axes.
2. ArrangeOverride — same opt-in reruns Yoga at finalSize.Height to cover the case where the parent arranges at a different size than the measure offer. _arranging continues to suppress child re-measure, preserving the resize-wobble fix from fix(FlexPanel): eliminate +/-1 px height wobble on resize #172.
3. MeasureFunction wrapper (load-bearing) — differentiate Yoga's Exactly mode (real stretch-fit, pass finite) from AtMost / Undefined (basis / content phase, pass infinity). Without this, a nested FlexPanel with default VerticalAlignment.Stretch would treat AtMost's soft cap as a fill target and report the cap as its DesiredSize, defeating the outer's flex-grow distribution across siblings. With it, nested FlexPanels behave correctly under any intermediate (Border, Grid, ContentControl, etc.).

Test plan

• dotnet test tests/Reactor.Tests — 6819 passed, 0 failed
• dotnet test tests/Reactor.SelfTests — 668 passed, 0 failed (includes the new Flex_HeaderBodyFooterFillsParentSlot fixture)
• dotnet run --project tests/Reactor.AppTests.Host -- --self-test --filter \"Flex\" — all 31 flex fixtures pass, including new HBF
• Manual: a demo app with FlexColumn(titleBar, header, ..., listView.Flex(grow: 1, basis: 0)).Padding(24) (no explicit .Height) now fills the window and the ScrollView produces a scrollbar instead of overflowing
• Recommended for reviewers: spot-check the wobble-prone scenarios (pure-width window resize over a header-rich column) — _arranging should still suppress the rerun's child re-measure, and Measure-time DesiredSize is still content-sized when parent offer is infinite

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR updates FlexPanel’s layout policy so a FlexColumn can treat the block axis as definite when VerticalAlignment.Stretch and the parent provides a finite slot, enabling the common header(auto) / body(flex:1) / footer(auto) pattern to fill a viewport without requiring an explicit .Height(N).

Changes:

• Update FlexPanel.MeasureOverride to treat block-axis size as definite when VerticalAlignment.Stretch and the parent’s height offer is finite.
• Update FlexPanel.ArrangeOverride to optionally re-run Yoga at finalSize.Height (while preserving the _arranging guard from #172).
• Add a new self-test fixture and register it in the self-test registry.

Show a summary per file

File	Description
src/Reactor/Yoga/FlexPanel.cs	Adjusts block-axis definiteness rules in measure/arrange and changes Yoga→WinUI measure-mode mapping.
tests/Reactor.AppTests.Host/SelfTest/Fixtures/FlexLayoutFixtures.cs	Adds an end-to-end self-test for header/body/footer filling behavior without explicit height.
tests/Reactor.AppTests.Host/SelfTest/SelfTestFixtureRegistry.cs	Registers the new self-test fixture name and factory mapping.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 3/3 changed files
• Comments generated: 2

[codemonkeychris]

Update pushed (3aec44e) addressing both Copilot review comments and the broader agent review.

What I changed

1. AtMost → infinity mapping reverted (Copilot point 1, agent point 1). Both reviewers correctly flagged that mapping YogaMeasureMode.AtMost to +∞ would under-report wrapped text height: TextBlock.Measure(width=∞) yields a 1-line result, TextBlock.Measure(width=200) yields the correct wrapped multi-line height. The wrapper now passes finite w + margin for both AtMost and Exactly, restoring the standard WinUI Measure contract.

The nested-FlexPanel discrimination (the original reason I changed the wrapper) is now done via a [ThreadStatic] _outerYogaHeightMode the wrapper publishes around child.Measure. FlexPanel.MeasureOverride reads it to tell:

• null → not nested under a flex measurement → use the WinUI Stretch contract.
• Exactly → stretch-fit allocation, fill.
• Undefined / AtMost → outer flex wants content size, do not fill (otherwise the inner reports the cap as DesiredSize and defeats the outer's flex-grow distribution).

This handles nesting through any intermediate (Border, Grid, ContentControl) without a brittle parent-type sniff, which was the failure mode of an earlier draft.

2. Class name (Copilot point 2). Renamed FlexColumnFillsParentSlot → FlexHeaderBodyFooterFillsParentSlot to match the registry key.

3. New regression test for AtMost (agent point 5). Flex_AtMostPreservesTextWrapping measures long text in a 200px column and asserts wrapped height ≥ 40px. Under a broken AtMost-as-infinity wrapper the height collapses to ~18px (1 line) — guards against the regression both reviewers flagged.

What I'm pushing back on

Agent point 2 — "keep auto block-size content-sized during measure unless Height is explicit; the arrange-time rerun handles fill."

I tried this. It does not fix the user's primary scenario. Diagnostic from running the HBF fixture with measure-time fillBlockAxis removed:

# DIAG col=784x750 sv.Render=640 sv.Viewport=640 sv.Extent=640 sv.Scrollable=0 sv.Desired=640

The column resolved to its content height (750) instead of the host's slot. The host's ContentPresenter arranges at min(slot, child.DesiredSize) = DesiredSize when the child is shorter than the slot, so ArrangeOverride.sizeChanged is false and the rerun never fires. Without measure-time fill, flex-grow children have no pool to expand into in the underflow case, which is the normal pre-overflow state of any web flex layout.

The arrange-time rerun is necessary but not sufficient — it covers the slot-overflow case (parent passes slot, child wants more) but misses the slot-underflow case (parent passes slot, child wants less). The user's dryrun.cs is the underflow case (8 items × 80px + chrome = 750 < 768 window), which is exactly when measure-time fill matters.

The trade-off the agent is right about is real, though: a Border with auto-height wrapping a default-Stretch FlexColumn will inflate to the parent's offer rather than shrink-wrap to content, because we are mapping "Stretch + finite parent" to CSS web-flex height: 100% at measure time. New self-test Flex_BorderAutoHeightInflatesWithStretch makes this testable and documents the opt-out — .VAlign(VerticalAlignment.Top) makes the inner shrink-wrap as expected. Asserts both halves so a future refactor that lands web-flex semantics without auto-height inflation has a known signpost to flip.

Agent points 3 (AlignContent default) and 4 (arrange-time stale measurements) — both fair observations but out of scope for this PR. AlignContent default is a separate semantic change (would affect any wrapped multi-line flex layout); arrange-time stale measurements is the existing wobble-fix tradeoff from #172, which this PR doesn't change.

Test results

• dotnet test tests/Reactor.Tests — 6819 passed, 0 failed
• dotnet test tests/Reactor.SelfTests — 670 passed, 0 failed (3 new fixtures: HBF, Border auto-height, AtMost text wrapping)