Generate Rust enum shells from spec literalUnions

[lorisleiva]

This PR teaches the generator to emit Rust enum shells from spec literalUnion TypeExprs, parallel to the enumeration support added in the previous PR. The sole literalUnion in v1 is isSigner: [true, false, "either"], referenced by instructionAccountNode.isSigner and instructionRemainingAccountsNode.isSigner (two value-identical inline copies); both consumers collapse to a single generated type. Because literalUnion is anonymous and inline in the spec — there's no name registry, unlike enumerations — the type name is derived from the referencing attribute (pascalCase('isSigner') → IsSigner). Distinct value-sets become distinct types; the same value-set referenced by two differently-named attributes throws as ambiguous.

The renderer uses a deliberately narrower derive set than enumPage does: Debug, PartialEq, Eq, Clone, Copy only — no Serialize/Deserialize, no Default, no rename_all. A literalUnion's wire form is heterogeneous (real JSON booleans for the boolean variants, a string for the "either" case) and can't be expressed via serde's declarative attributes; the bespoke Serialize/Deserialize visitor pair, the From<bool> conversion, and the Default choice all stay hand-written in a companion file. This is the same generated-shell / hand-written-impls split established by Number and re-used for the enumeration Default impls in the previous PR.

Reconciling the Rust crate required one breaking rename: IsAccountSigner → IsSigner, since the generator derives the type name from the attribute name. ~45 occurrences across codama-nodes (2 node files + shared/) and codama-attributes (the FromMeta impl + account_directive tests). The hand-written shared/is_account_signer.rs (104 lines, enum + impls + tests) becomes shared/is_signer.rs (impls + tests only — 122 lines, +2 tests covering From<bool> and default_is_false that weren't previously covered). One small wiring note: attributeBodyLine.ts resolves the literalUnion field type itself (because getTypeExprFragment has no access to the attribute name), while typeExpr.ts's literalUnion case continues to throw — only nested-position literalUnions would reach it, and those don't occur in v1.

Workspace-wide: 1006 → 1008 cargo tests passing (+2 from the new hand-written tests), 119 → 130 JS tests passing (+11 from the new discovery + renderer + integration tests), fmt and clippy stay clean, pnpm generate round-trips deterministically. The override surface (3 maps, FIELD_TYPE_OVERRIDES with 1 entry) is unchanged.

[lorisleiva]

• spec-generators: map float to crate::Number; remove FIELD_TYPE_OVERRIDES #104
• Generate Rust type nodes #103
• Generate Rust topLevel nodes (account, constant, error, event, instruction*, pda, program, root, …) #102
• Generate Rust enum shells from spec literalUnions #101 👈 (View in Graphite)
• Generate Rust enum shells from spec enumerations #100
• Generate contextualValue nodes; align Rust variant names with the derived rule #99
• Generate RegisteredNodes category unions; remove HAND_WRITTEN_UNIONS #98
• Derive inline-union allowlist from the spec; remove INLINE_UNIONS #97
• Rename Rust types to match spec; remove UNION_NAME_OVERRIDES + ENUMERATION_NAME_OVERRIDES #96
• Generate value nodes; introduce field-type override + hand-written-union skip #95
• Add inline-union support and complete pdaSeed generation #94
• Generate count and discriminator nodes #93
• Add JS spec generator for codama-nodes link nodes #92
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[lorisleiva]

Merge activity

• Jun 10, 2:13 PM UTC: A user started a stack merge that includes this pull request via Graphite.
• Jun 10, 2:31 PM UTC: Graphite rebased this pull request as part of a merge.
• Jun 10, 2:31 PM UTC: @lorisleiva merged this pull request with Graphite.