Module icu 

message-bundles slice 3 (#878): the ICU MessageFormat placeholder mini-parser, plus the plain-{name} template scanner (split_template) it dispatches from. Self-contained and bynk-syntax-free by design (Decision B) — the whole ICU sub-grammar ({name, plural, one {…} other {…}}, {name, select, …}, {name, number[, style]}, {name, date[, style]}) lives entirely inside a messages template’s String content, parsed here as plain &str and consumed by both the checker (bynk-emit/src/project/validate.rs) and the emitter (emit_message_entry_renderer). No bynk-syntax grammar/lexer/AST change backs this — a template stays one opaque String all the way through parsing.

Lives in bynk-check (moved here in the compiler-pipeline-review’s Wave 5, batch 5.1) rather than bynk-emit, since its only consumer besides the emitter is checker-side (validate.rs’s malformed-syntax pass) and it has no emitter-specific dependency of its own.

Quoting: a bare ' toggles a “quoted” region; '' inside either mode means a literal ' and doesn’t toggle; while quoted, {/}/,/# are inert literal text. This is a deliberately narrower rule than full ICU MessageFormat’s own quoting semantics — sufficient for this slice’s fixed subset, not a general ICU implementation (Decision B/the proposal’s named scope cuts).

Explicitly unsupported, each with its own diagnosable IcuParseErrorKind rather than silent misbehaviour: selectordinal, plural’s offset:/=N exact-value arms, arbitrary CLDR skeletons beyond the fixed style keywords below, and nesting a second {arg, …} dispatch inside a sub-message (a sub-message is literal text + # only).

Structs

IcuParseError

IcuPlaceholder

Enums

DateStyle

FormatKind

The coarse comparison unit for cross-locale format agreement (bynk.messages.format_mismatch, bynk-emit/src/project/validate.rs) — arm/style content doesn’t matter for agreement, only which of the five surface forms a placeholder uses. Plain covers the bare {name} fast path, which never calls into this parser at all.

IcuParseErrorKind

NumberStyle

PlaceholderKind

PluralCategory

SubSegment

One piece of a plural/select arm’s sub-message. Owned, not &'a str: ICU ''-unescaping can shrink byte length, so a literal sub-segment cannot always borrow from the source template.

TemplateSegment

A {name} placeholder (plain, or — message-bundles slice 3 (#878) — an ICU-dispatch placeholder whose inner text contains a ,) or a run of literal text inside a message template. offset is the byte offset in the owning template where inner/the placeholder’s content begins (right after the opening {) — used by slice 3’s icu_dispatch_placeholders to rebase parse-error spans (Decision C).

Functions

find_icu_close

Quote+depth-aware scan from just after a placeholder’s opening { (rest), used only once emission-side detection (split_template) has already decided this is an ICU-dispatch placeholder (a , precedes the naive first }). Depth starts at 1 for the still-open outer brace; returns the byte offset of the true closing } (depth reaching 0), None if it never closes.

icu_dispatch_placeholders

Every ICU-dispatch placeholder in template — (byte offset of innerwithintemplate, inner text) — for the checker’s malformed-syntax pass (bynk-emit/src/project/validate.rs). A placeholder only reaches here if split_template already decided it was ICU-dispatch (its inner contains a top-level comma); a plain {name} never does, by construction (proven in split_template’s own doc comment).

parse_icu_placeholder

Parses one placeholder’s inner text (everything between the outer {/}, e.g. "count, plural, one {# item} other {# items}") — the precondition for calling this at all is that inner contains at least one top-level comma (split_template’s own trigger for treating a placeholder as ICU-dispatch rather than a plain {name}).

placeholder_names

message-bundles slice 2 (#874): a template’s placeholder-name set, for cross-locale agreement checking (bynk-emit/src/project/validate.rs). Exposes only the name set, not TemplateSegment itself, keeping the checker’s dependency on the emitter narrow. message-bundles slice 3 (#878): an ICU-dispatch placeholder’s name is its inner text up to the first top-level comma, trimmed — a plain placeholder’s inner never contains a comma (by construction of split_template’s is_icu decision above), so it’s returned verbatim, preserving the untrimmed-name quirk documented on split_template exactly as before.

split_template

Compile-time string scan splitting a template into literal/placeholder segments (Decision D, message-bundles slice 1 — no new lexer/parser grammar; {name} is resolved by this Rust-side scan during lowering, not parsed as an expression). A { with no matching }, or an empty {}, is just literal text — malformed-placeholder checking is out of scope here. The name is taken verbatim, with no whitespace trimming: { name } is a placeholder literally named " name ", which will never match a params key and so always renders as literal text (PR #872 review) — a real rough edge, left for a future slice rather than guessed at here.

template_format_kinds

Every placeholder’s (name, FormatKind) in template, silently dropping any placeholder whose ICU parse fails — that failure is check_entry_icu_syntax‘s job to report (once), not this pure helper’s, so a caller comparing two locales’ templates never double-reports a malformed one.