Phase 4. Records (structs)

Field	Value
MEP	MEP-53 §Phases
Status	LANDED
Started	2026-05-28 (GMT+7)
Landed	2026-05-29 07:35 (GMT+7)
Tracking issue	— (umbrella)
Tracking PR	#22499
Commit	a7b47981ff

Gate

TestPhase4Records walks tests/transpiler3/rust/fixtures/phase04-records/ (14 fixtures) and asserts byte-equal stdout. Coverage: record literal, field access, field update (functional, via struct update syntax { ..base, field: new }), nested records, anonymous-record types, records in lists / maps, structural equality, print via Debug.

Lowering decisions

record User { id: int, name: string } and type Pair = { a: int, b: int } both lower to:

#[derive(Clone, Debug, PartialEq, Default)]
struct User {
    id: i64,
    name: String,
}

Both surface forms route through lowerStruct in transpiler3/rust/lower/struct.go. The Default derive lets spawn AgentType() (phase 9) construct zero-valued instances. The PartialEq derive lets == work as expected. The Clone derive is required because field updates are functional ({ ..base, field: new } consumes base by value unless cloned). The Debug derive is required for print(user) to lower to mochi_runtime::io::print_str(format!("{:?}", user)).

Field updates lower to:

User { id: 42, ..base.clone() }

The .clone() on base is gated by the colour pass; in phase 4 it always clones, in phase 6 it elides where the base is dead.

Anonymous-record types declared at use site (let p: { a: int, b: int } = { a: 1, b: 2 }) get a synthesised struct name (__Anon_<sha8>) hoisted to module scope; deduplication is by structural-hash of the field set, so two uses of the same anonymous shape collapse to the same struct.

Files changed

File	Purpose
transpiler3/rust/lower/struct.go	Record + anonymous-record lowering
transpiler3/rust/build/phase04_test.go	14-fixture gate
tests/transpiler3/rust/fixtures/phase04-records/*.mochi + .out	14 fixtures

Test set

• TestPhase4Records/<fixture> for each .mochi in the fixture directory (14 fixtures).

Closeout notes

Printing records via Rust's {:?} Debug format does not match vm3's pretty-printer exactly (vm3 separates fields with , , Debug with ,; vm3 elides type name on anonymous records). Phase 4 fixtures all stay within field-by-field printing rather than whole-struct printing to dodge the gap; a follow-up sub-phase could emit a custom Display derive that matches vm3 byte-for-byte.