mirror of
https://github.com/razzant/ouroboros.git
synced 2026-10-03 12:18:39 +00:00
d1ed1e2fb1
Five scenarios on the wave-1 skeleton (manifest rows S6-S10, keyless mock lane, durable-artifact assertions, durable-event polling): - S6 subagent tree: ReplayModel parent drives schedule_subagent -> wait_tasks -> exact-hash tree_note disposition -> final; the child runs on its own stub slot (slot binder = model id x tool-bearing shape) so every fixture ordinal is deterministic under parent/child concurrency. Pins lineage truth in the durable rows, the swarm_fanout receipt in the parent's forked drive, quiescence (child task_done precedes parent's; clean, not children_unabsorbed), the child result reaching the parent verbatim, the tree-ledger disposition row, root with-children cost keys, and full fixture consumption. - S7 cancellation single: typed cancelled terminal with an owed answer (outbox delivery or the typed no_lineage_chat handoff row), honest cost names, self-draining cancel intent with the requested->claimed->settled forensic trail, terminal task_done, and a /proc no-orphans scan of the live server tree. - S8 cancellation cascade: live child, cascade teardown; root intent minted scope=cascade at the ingress and settled only on the cascade postcondition, descendant carries its own cascade_descendant intent, the subtree snapshot lists the child, no orphan processes. - S9 managed update ff core: a real managed install with a LOCAL managed upstream (official update URL redirected via git url.insteadOf - install config, not patched runtime), dirty-tree fast-forward applied over the live HTTP surface; stash-first insurance carries the owner's work, the server re-execs, boot-finalize consumes both markers, writes managed_update_finalized head==target, restores the dirty work and keeps the durable rescue-local stash pin. - S10 rollback contracts (subprocess driver on a second isolated install): absent marker and explicit null stamp refuse typed (no pre_update_sha; strict corrupt) with marker bytes intact; a FUTURE-schema stamp refuses typed from both rollback_managed_update and finalize_managed_update_on_boot with the marker byte-identical; a half-applied update rolls back to a byte-for-byte identical worktree with the candidate preserved on failed-update-* and the durable managed_update_rolled_back receipt. Harness: callable script/fixture steps (dynamic server-minted arguments), ReplayModel model_ids override, pids_with_env_value no-orphans oracle, terminal_deliveries / child_task_ids / tree_blackboard oracle readers; manifest gen/verify + marker pins now scan every module of the package; shared session clone moved to the package conftest. ARCHITECTURE System E2E subsection extended; ledger carries the wave section including the E2E findings table (W2-F1..F4) and lane timings (full mock lane 31 passed in ~219s).
21 lines
774 B
Python
21 lines
774 B
Python
"""Shared fixtures of the system_e2e scenario package.
|
|
|
|
One throwaway clone serves every scenario server of a session (cloning the
|
|
checkout is the expensive step); scenarios that MOVE the clone's HEAD or add
|
|
remotes (the managed-update wave) must build their own private clone instead —
|
|
sharing a mutated clone would couple scenario outcomes to execution order.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import pytest
|
|
|
|
from tests.system_e2e.harness import LANE_MOCK, clone_repo, require_lane
|
|
|
|
|
|
@pytest.fixture(scope="session")
|
|
def e2e_clone(tmp_path_factory):
|
|
"""One throwaway clone of the checkout under test, shared by every scenario
|
|
server that leaves HEAD alone."""
|
|
require_lane(LANE_MOCK)
|
|
return clone_repo(tmp_path_factory.mktemp("system_e2e_clone"))
|