0006 — String props use AttrValue

• Status: accepted
• Date: 2026-07-11
• Supersedes: 0001

Context

ADR 0001 fixed NavLink’s class / active_class as &'static str, trading runtime composition away for zero-allocation literals. The same type then spread by imitation to every string-ish prop in the crate: id, text, toggle_text, href, variant, role, aria_label.

Two things changed the balance:

1. The zero-cost argument no longer discriminates. Yew’s AttrValue is a two-variant handle: AttrValue::Static(&'static str) — the exact same pointer ADR 0001 optimized for — or Rc<str> for runtime strings. html! converts string literals into AttrValue::Static with no allocation, so the 99% literal case costs the same as before. Yew’s own documentation recommends AttrValue over String/&str for props.
2. The restriction turned out to bite in practice. PageLink’s href: Option<&'static str> makes a pagination link’s URL impossible to build at runtime — format!("/page/{n}") cannot become &'static str without leaking. What was a stylistic trade for class slots is a functional hole for URL slots.

Decision

All string-valued props take AttrValue (Option<AttrValue> when optional). Defaults use AttrValue::Static, keeping the literal path allocation-free. Class-list props that benefit from composition and deduplication stay Classes, as before.

Consequences

Positive

• Runtime values work everywhere they should — most importantly PageLink::href.
• Props match the idiom the Yew book teaches, so consumers coming from other Yew code find no surprises.
• Cloning props stays cheap: AttrValue clones are a pointer copy (Static) or an Rc refcount bump.

Negative

• The type-level guarantee that a class name is a compile-time constant is gone; a consumer can now funnel a runtime string into a class slot. The escaping behaviour of the DOM layer still prevents HTML injection.
• One-time semver break: struct-literal constructions of props must wrap literals in AttrValue::Static (in html! nothing changes).

Cost to reverse

Moderate and symmetric to ADR 0001’s estimate: signature changes across public props, one-line fixes per call site.