pika-orch-incus-cleanup-23

The core of this change is a new struct and a builder function added to model.rs.

Struct definition

#[derive(Clone, Copy, Debug, Deserialize, Serialize, Eq, PartialEq)]
pub struct StagedLinuxRustTargetPayloadSpec {
    pub role: StagedLinuxRustPayloadRole,
    pub output_name: &'static str,
    pub nix_installable: &'static str,
}

Each spec captures exactly the three pieces of data that previously lived as separate scalar fields on StagedLinuxRustTargetConfig:

StagedLinuxRustPayloadRole also gains Serialize/Deserialize derives with #[serde(rename_all = "snake_case")] so the specs can round-trip through JSON.

Constructor helper

fn staged_linux_target_payload_specs(
    workspace_deps_output_name: &'static str,
    workspace_deps_installable: &'static str,
    workspace_build_output_name: &'static str,
    workspace_build_installable: &'static str,
) -> [StagedLinuxRustTargetPayloadSpec; 2] { ... }

This private function keeps every config() match arm concise — it accepts the same four values that used to be individual fields and returns the fixed-size array.

StagedLinuxRustTargetConfig loses four fields and gains one:

pub struct StagedLinuxRustTargetConfig {
    // ... unchanged fields ...
    pub payload_specs: [StagedLinuxRustTargetPayloadSpec; 2],
    pub workspace_output_system: &'static str,
    pub shadow_recipe: &'static str,
}

Because the array is fixed-size [...; 2], the struct remains Copy and stack-allocated — no heap allocation is introduced.

All nine target variants (PreMergePikaRust, PreMergePikaFollowup, PreMergeAgentContracts, PreMergeNotifications, PreMergeFixtureRust, PreMergeRmp, PreMergePikachatRust, PreMergePikachatTypescript, PreMergePikachatOpenclawE2e) are updated uniformly. For example:

Self::PreMergePikaRust => StagedLinuxRustTargetConfig {
    // ...
    payload_specs: staged_linux_target_payload_specs(
        "ci.x86_64-linux.workspaceDeps",
        ".#ci.x86_64-linux.workspaceDeps",
        "ci.x86_64-linux.workspaceBuild",
        ".#ci.x86_64-linux.workspaceBuild",
    ),
    // ...
},

The pattern is identical across all arms — only the Nix attribute paths differ.

Three new methods appear on both StagedLinuxRustLane and StagedLinuxRustTarget:

// Return the full two-element spec array
pub fn payload_specs(self) -> [StagedLinuxRustTargetPayloadSpec; 2]

// Look up a single spec by role (panics if missing — impossible with the fixed array)
pub fn payload_spec(self, role: StagedLinuxRustPayloadRole) -> StagedLinuxRustTargetPayloadSpec

// Convenience: get the Nix installable string for a role
pub fn payload_nix_installable(self, role: StagedLinuxRustPayloadRole) -> &'static str

Existing accessors like workspace_deps_output_name() and workspace_build_output_name() on StagedLinuxRustLane now delegate to payload_spec(role).output_name instead of reading the (now-removed) config fields directly.

The previous payload_roles() method is preserved but now derives from payload_specs().map(|spec| spec.role), and payload_output_name(role) delegates to payload_spec(role).output_name.

The CLI's show-target-config subcommand previously emitted two fixed key-value lines:

workspace_deps_installable=...
workspace_build_installable=...

It now iterates over config.payload_specs and emits:

payload_workspace_deps_installable=...
payload_workspace_build_installable=...

The key name is derived from role.prepare_node_suffix(), which returns "workspace_deps" or "workspace_build". This is a minor output format change — downstream consumers that parse this output will need to match the new payload_*_installable keys.

In run.rs, the staged_linux_payload_specs function is the runtime consumer that builds StagedPreparedPayload values for each stage of the CI pipeline.

Before this change it called lane.payload_roles() and then separately looked up lane.payload_output_name(role). Now it calls lane.payload_specs() and reads output_name directly from the spec:

lane.payload_specs()
    .into_iter()
    .map(|payload_spec| StagedPreparedPayload {
        output_name: payload_spec.output_name,   // direct access
        // ...
    })

This eliminates the indirection of asking the lane to re-derive data that already lives in the spec.

Tests in both main.rs and model.rs are updated:

Serialization test (main.rs): JSON assertions now index into the payload_specs array:

assert_eq!(
    payload["payload_specs"][0]["nix_installable"],
    ".#ci.x86_64-linux.workspaceDeps"
);

Config accessor tests (model.rs): All assertions that previously read config.workspace_deps_installable or config.workspace_build_installable now call the typed accessor:

assert_eq!(
    notifications.payload_nix_installable(StagedLinuxRustPayloadRole::WorkspaceDeps),
    ".#ci.x86_64-linux.notificationsWorkspaceDeps"
);

Allowlist loop test: The test that verifies all installables appear in the strict staged remote helper now binds the installable strings via target.payload_nix_installable(role) before asserting, keeping the error messages readable.

This covers every target: notifications, followup, fixture, pikachat (Rust, TypeScript, OpenClaw), RMP, agent contracts, and core pika.