The Scribe

xrune-fmt is the formatter for ui! { … } blocks. It’s a CLI binary, not a library — install it once and point it at any .rs file that contains casting macros.

cargo install xrune-fmt

xrune-fmt src/app.rs            # rewrite in place
xrune-fmt src/app.rs --check    # exit 1 if not formatted, leave file alone

What it does

For every ui! { … } block it finds, the scribe:

1. Locates the macro by regex (ui!\s*\{) and the matching } via brace-depth counting.
2. Hands the inside to the real parser — xrune-nexus’s DsRoot::parse — and gets back a DsTree.
3. Walks the tree and re-emits it with consistent indentation, line breaks, and spacing.
4. If the parser rejects the input, the original block is left untouched. The scribe never silently rewrites a block it cannot understand.

It only touches the body of ui! { … }. Code around the macro is preserved byte-for-byte.

Formatting rules

• Context header always multi-line, one attr per line, indented one step beyond the ui! brace.
• Widget attrs ride on a single line if they fit within MAX_LINE_WIDTH = 100; otherwise each attr goes on its own line. If the original was multi-line, the scribe keeps it multi-line even when it would now fit on one — author intent wins over column count.
• Attribute values, walk iterables, if/match scrutinees, and on bodies all run through prettyplease so embedded Rust expressions render in canonical form.
• on EventKind clauses stack between attrs and body; their bodies are indented one step beyond the surrounding widget.
• Enchants sit between attrs and body in [ … ], comma-separated.

The scribe round-trips the same forms xrune-nexus::tests exercises — positional attrs, named attrs, headerless widgets, niches, match arms, the three on shapes (Form B, Form C, callback). The xrune-fmt test suite walks every fixture in examples/example0/src/ui to confirm idempotence.

Why a separate parser-shaped consumer

The scribe is the second consumer of the same DsTree. The first is your rune (DsRune impls + decipher). The scribe doesn’t implement DsRune — it walks the tree manually, because its goal is to re-emit syntax rather than transform a tree into runtime code.

This is also why the scribe is the canary for the language: any time the parser learns a new shape, the scribe needs the same field exposed; any time the AST grows a new node, the scribe needs a new arm. If you’re considering adding to xrune-nexus, glance at the scribe to see how much downstream work the addition implies.

Source-of-truth

• CLI + ui!-block extraction: crates/xrune_fmt/src/main.rs
• Tree walking + emission: crates/xrune_fmt/src/formatter.rs