Skip to content

fix(runtime): complete the nif_child_tasks to engine_tasks rename - #27

Open
JoshuaBearup wants to merge 1 commit into
ablative-io:mainfrom
JoshuaBearup:fix/complete-engine-tasks-rename
Open

fix(runtime): complete the nif_child_tasks to engine_tasks rename#27
JoshuaBearup wants to merge 1 commit into
ablative-io:mainfrom
JoshuaBearup:fix/complete-engine-tasks-rename

Conversation

@JoshuaBearup

Copy link
Copy Markdown

bed958b0 renamed crates/aion/src/runtime/nif_child_tasks.rs to engine_tasks.rs, and the type ChildTaskRuntime to EngineTaskRuntime, but the module declaration and every reference were left behind. main does not compile as a result.

On main today:

  • crates/aion/src/runtime/mod.rs:27 declares mod nif_child_tasks; — that file no longer exists, and engine_tasks is never declared.
  • nif_child_engine.rs and nif_child_watch.rs carry 10 references to nif_child_tasks::ChildTaskRuntime.

This PR declares engine_tasks, drops the dead nif_child_tasks declaration, and points both call sites at engine_tasks::EngineTaskRuntime. Rename completion only — no behaviour change, no logic touched.

Verified by building the release aion binary from this change and running it as our deployed engine.

bed958b renamed crates/aion/src/runtime/nif_child_tasks.rs to
engine_tasks.rs and the type ChildTaskRuntime to EngineTaskRuntime, but
left the module declaration and every reference behind, so main does not
compile:

  mod nif_child_tasks;                       # runtime/mod.rs:27 — file is gone
  use crate::runtime::nif_child_tasks::ChildTaskRuntime;   # x2
  Arc<ChildTaskRuntime> and ChildTaskRuntime::new()        # x8

This declares engine_tasks, drops the dead nif_child_tasks declaration,
and points nif_child_engine.rs and nif_child_watch.rs at
engine_tasks::EngineTaskRuntime. Rename only — no behaviour change.

Verified by building the release aion binary from this change and running
it as the deployed engine.
tomWhiting added a commit that referenced this pull request Aug 8, 2026
…open questions put for ruling

The last paper before the #27 build is dispatched: folds Tom's four
2026-08-08 direction threads (beamr supervision framing, pre-built worker
packaging, AWL worker-authoring UX, ACP v2) into the decided ground of
RUNTIME-OPERATIONS R6-R10 and WORKER-LIFECYCLE-BUILD W-0..W-6, states what
the tree has grown since those papers (the containment core already runs
under every declared body but nothing triggers it; drain/stop routes exist
but are registry-only; the census landed; the harness adapters are the open
orphan surface), and answers the two questions WORKER-LIFECYCLE-BUILD held
for the operator: artifact posture (builtin verb-deployments + additive
.aion worker kind, no separate artifact object) and the phase plan (W-0..W-6
with three tree-fact amendments). ACP verified v1-not-v2 with the upgrade
sized S. Waits at Waffles' seat for review before any build unit moves.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
tomWhiting added a commit that referenced this pull request Aug 8, 2026
…he binary it runs

The foundation unit of worker lifecycle through the server. A WorkerDeployment
is a durable, type-erased record (builtin-only artifact, deploy-time binary
identity captured from BuildIdentity + SHA-256 of the running executable) with
a WorkerDeploymentStore contract implemented by the in-memory, libSQL, and
haematite backends under ONE shared conformance suite — restart survival over
genuinely reopened storage, replacement that preserves the record's memory
(created_at, status history + a stable "replaced" entry, last_spawn_binary),
poisoned rows visible and removable without taking the listing or the cluster
channel down, and PUT returning the exact record persisted so no read races a
delete. Workers MAY carry a tag-7 instance identity on RegisterWorker (exact
bytes pinned); association to a deployment is observational tri-state
(Known/Absent/Unchecked) — never an admission gate. Five deploy-granted HTTP
routes behind the deploy surface's own switch; cluster events and snapshots
move together through to the console's strict decoder. NOTHING SPAWNS: a
lexical source fence with positive controls and truncation anchors pins the
scope fence over every W-0 site.

Honestly incomplete, by dated amendment in the brief: the liminal wire has no
instance-identity field in published liminal-rs, so that half is deferred to a
named lane under #27; cross-binary comparison is W-2's.

Review chain: Opus R1 REFUSED (3B/7M/7m) → fix lane → R2 REFUSED (3M/8m) →
fix lane → R3 REFUSED (1M/2m) → operator fixes → R4 CLEAR. Every finding from
every round discharged, none waved. Mutation probes red→green with named-test
green legs and hash-verified restorations; planted positive re-run at the
final call-site form; the namespace_restart red attributed to #85 with an
independent base rate. Evidence: docs/evidence/worker-lifecycle-w0-durable-
record-wire-identity.md + gate-logs/w0-worker-deployment/ (sealed, 103 files).
Accepted debt recorded: four pre-existing over-500-line files grew ~610 lines
of integration seams; owner #27, due before the lifecycle units complete.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant