[00135] Implement Rust Widget APIs for Blade Navigation and Advanced Layout Components

[rorychatt]

Fixes #125

00135 — Implement Rust Widget APIs for Blade Navigation and Advanced Layout Components

Issue #125. Worktree:
Worktrees/Rusty-Framework, branched from c528820 on main.

What shipped

Five Rust widgets that previously existed only as React components, plus the adapter, harness,
test and doc work that makes them usable and gated.

File	Change
rusty/src/widgets/blade.rs	new — Blade, BladeContainer
rusty/src/widgets/breadcrumbs.rs	new — Breadcrumbs, BreadcrumbItem
rusty/src/widgets/pagination.rs	new — Pagination
rusty/src/widgets/toolbar.rs	new — Toolbar, ToolbarItem, ToolbarItemVariant
rusty/src/widgets/mod.rs	four pub mod + four pub use lines
rusty/src/shared/widget_names.rs	five ivy_widget entries; three counts 38 → 43, 25 → 30
rusty/src/shared/ivy_node.rs	IVY_EVENT_NAMES 8 → 12; five to_ivy_node tests
rusty/src/core/event_registry.rs	four EventName variants + case-insensitive On strip (deviation, below)
rusty-server/src/bin/widget_harness.rs	four WidgetKind variants, arms and impl View apps
e2e/app/index.html	five renderWidget case arms + CSS
e2e/tests/widgets/navigation.spec.ts	new — 10 specs
rusty-docs/docs/03_widgets/34_blade.md … 37_toolbar.md	new — four pages

Commits: 59eeb8e (widgets), 659f5dc (adapter), 45bfa47 (event names), 4d079bc (e2e),
d81cc62 (docs), fb73f7c (cargo fmt).

Design decisions worth knowing

Items are props, not widgets. BreadcrumbItem and ToolbarItem are plain serde structs, so
they carry no widget id and cannot hold a closure. The frontend fires one event on the widget —
OnItemClick with an index, OnSelect with a tag — and reads per-item booleans
(hasOnClick) to decide how to render. ToolbarItem.children is Vec<ToolbarItem>, not
Vec<Element>, which is also what keeps it clear of widget_checks.rs's rule that a Vec<Element>
field not named children silently loses ids.

ToolbarItemVariant serializes PascalCase on purpose. ToolbarWidget compares
item.variant === "Group" / "Separator", and that comparison is on a nested value.
ivy_node's ENUM_PROPS recasing only reaches top-level props, so a rename_all here would
produce "group" and silently break group rendering.

page is 1-based, 0 means nothing selected. That is how PaginationWidget reads it (it
tests !page), so Pagination::new(0, n) is a meaningful state, not a bug.

No BladeHeader slot and no use_blades. BladeWidget's header slot needs an Ivy.Slot
child node and Rust has no Slot widget; Ivy's UseBlades push/pop controller is the app's own
use_state. Both are recorded in blade.rs's doc comments, as the plan required.

Deviation from the plan: EventName

The plan stated that no EventName variant was needed, because "the custom-name fallback in
event_registry.rs covers it" and a browser OnItemClick "normalizes to itemclick and reaches the
handler".

That was wrong about the code as it stood. normalize used s.strip_prefix("on"), which is
case-sensitive: "OnClose" lowercased to "onclose", matched no variant, and canonicalize
returned the raw "OnClose" — never equal to the registered "close". The plan's own Tests
section requires registry.dispatch("w-0", "OnClose", Null) to fire on_close, and ivy_node's
module doc already claimed PascalCase wire names are accepted.

45bfa47 therefore:

• adds Close, Refresh, ItemClick, Select variants with as_str / from_str arms and
  extends the existing round-trip test's all array;
• makes the prefix strip accept On as well as on, gated on the following character being
  uppercase, so Online / online keep their on;
• adds test_from_str_accepts_ivy_pascal_case_wire_names.

This makes the plan's specified tests pass rather than rewriting them to match broken behaviour. It
also fixes the same latent gap for every pre-existing PascalCase event name (OnClick, OnChange,
…), which is why it is called out here rather than buried.

Verification

Verification	Result	Evidence
RustFmt	Pass	1 diff fixed (fb73f7c), re-check exit 0
RustClippy	Pass	--workspace --all-targets --no-default-features -- -D warnings, exit 0, 0 warnings
VitePlusCheck	Skipped	no src/frontend change; run anyway — 563 files formatted, 548 lint-clean, tsc -b exit 0
RustBuild	Pass	exit 0, 0 warnings
NpmBuild	Pass	tsc -b && vp build exit 0; only the pre-existing chunk-size warning
RustTest	Pass	771 passed, 0 failed, 4 ignored; 37 tests added; inventory gate: 741 → 778, nothing removed
VitePlusTest	Skipped	no src/frontend change; 1 pre-existing failure (ci-step.test.ts), proven out of scope
PlaywrightE2E	Pass	152 passed, 1 flaky, 0 failed of 153, including 10 new; check-harness-script.js passes
CheckResult	Pass	plan matched section by section

The one red test anywhere is src/frontend/src/__tests__/ci-step.test.ts, which asserts the
literal gate commands in .github/workflows/ci.yml. Those gained --no-default-features and the
assertion was never updated. git diff --name-only c528820..HEAD touches neither file, so every
input is byte-identical to main — it fails there too. Details in Verification/VitePlusTest.md,
carried into Artifacts/recommendations.md.

Environment notes

• npm ci in e2e/ fails with UNABLE_TO_GET_ISSUER_CERT_LOCALLY unless you pass Node's bundled
  CA store: NODE_OPTIONS=--use-bundled-ca npm ci works and is what the recorded run used.
  pnpm install in src/frontend needs no flag (it resolves from the local store).
• One pre-existing Playwright flake, text_input read_only › is writable when read_only is not set
  — a page.goto ERR_ABORTED inside navigateToHarness, passing on the configured retry. The
  same suite ran clean earlier in the session on the identical tree.
• Everything ran in the worktree; the original repo at /Users/rorychatt/.tendril/Repos/Rusty-Framework
  was read only. Worktree left on disk, git status clean.

---

Commits:

• 59eeb8e
• 659f5dc
• 45bfa47
• 4d079bc
• d81cc62
• fb73f7c

---

Created using Ivy Tendril.