68 lines
3.0 KiB
Markdown
68 lines
3.0 KiB
Markdown
|
---
|
||
|
|
created_at: 2026-10-22T21:22:48Z
|
||
|
|
generated_by:
|
||
|
|
harness: opencode
|
||
|
|
provider: openai
|
||
|
|
model: openai/gpt-6-astra
|
||
|
|
git_ref: dc618229bd8c9c90aa3160e26677b553a7d472c7
|
||
|
|
scope: code
|
||
|
|
substantive: true
|
||
|
|
raw_file: 20261022T212248Z_dc618229_ib_recovery_prompt_io.raw.md
|
||
|
|
---
|
||
|
|
|
||
|
|
## Prompt
|
||
|
|
|
||
|
|
The human approved trying the bounded candidate: back out the IB
|
||
|
|
quote-context cache, import adapter-only reset/boundary fixes and
|
||
|
|
the missing EMS subscription fix, preserve source authorship and
|
||
|
|
commit boundaries, and exercise the real Tractor bridge. They
|
||
|
|
required an explicit keep/backout audit and clarification that
|
||
|
|
the evaluated downstream lineage lacked the broadcast fix.
|
||
|
|
|
||
|
|
## Response summary
|
||
|
|
|
||
|
|
Verified identical stable patch IDs for original 4af885d9 and
|
||
|
|
landing replay 635314c7. Created `wkt/land_ib_recovery` at dc618229
|
||
|
|
in the main checkout. Applied source-equivalent adapter backports
|
||
|
|
from 0846cbd4 and 1da9b760 and the EMS code from 12b0d66e, with
|
||
|
|
source attribution recorded in `ai/ib-recovery-candidate.md`.
|
||
|
|
|
||
|
|
The real-runtime quote test first reproduced the reported fatal
|
||
|
|
TrioTaskExited propagation with caching enabled. The direct-owner
|
||
|
|
backout restores restart and subsequent RPC survival, but a strict
|
||
|
|
callback-cleanliness test remains red. Real bridge history pacing
|
||
|
|
recovery and real broadcaster dark clearing pass. No live broker,
|
||
|
|
VNC, orders or TSDB operations were performed by the agent.
|
||
|
|
|
||
|
|
On 2026-10-23 the human supplied successful live startup/backfill,
|
||
|
|
sequential reattachment and overlapping chart-close/reopen results,
|
||
|
|
plus the still-failing daemon SIGINT teardown. The human accepted
|
||
|
|
deferring teardown from #94, requested loud partial-backport
|
||
|
|
attribution and required downstream rebase rehearsals before landing.
|
||
|
|
The agent removed only the explicit single-null debugger stop,
|
||
|
|
retaining its legacy result. Downstream `976db6c3` instead rewrites
|
||
|
|
null grouping; the local cleanup does not claim that algorithmic fix.
|
||
|
|
|
||
|
|
## Files changed
|
||
|
|
|
||
|
|
- `piker/brokers/ib/feed.py`: source-equivalent backout/backports.
|
||
|
|
- `piker/clearing/_ems.py`: original EMS subscription patch.
|
||
|
|
- `tests/test_ib_broadcast.py`: removed with the reverted behavior.
|
||
|
|
- `tests/test_ib_history.py`: original adapter regressions only.
|
||
|
|
- `tests/test_ems_broadcast.py`: imported and real-broadcaster tests.
|
||
|
|
- `tests/test_ib_quote_restart.py`: real Tractor boundary tests.
|
||
|
|
- `piker/tsp/_anal.py`: local single-null breakpoint removal only.
|
||
|
|
- `ai/ib-recovery-candidate.md`: provenance and qualification limits.
|
||
|
|
|
||
|
|
## Human edits
|
||
|
|
|
||
|
|
The human required original fixes and authorship, rejected broad
|
||
|
|
speculative integration work, required real concurrency boundaries
|
||
|
|
in tests, and corrected the interpretation of live qualification.
|
||
|
|
They performed the live tests, chose to defer known teardown defects
|
||
|
|
from #94, permitted the minimal debugger cleanup, and requested
|
||
|
|
explicit downstream discrepancies and rebase compatibility checks.
|
||
|
|
The agent applied those directed edits; no direct manual source edits
|
||
|
|
are known in this candidate. The raw partner retains the originally
|
||
|
|
captured output; this structured record includes later contributions.
|