Files
prometeu-runtime/discussion/lessons/DSC-0072-headless-memcard-reopen-io-progress/runtime-handoff.md
T
bquarkz c3589be1ad
Intrepid/Prometeu/Runtime/pipeline/head There was a failure building this commit
fix(headless): wait for memcard bootstrap IO progress
Coordinate bounded native publication progress during frame advance while preserving watchdogs, guest budgets and resource ownership. Add controlled and packaged reopen regressions, clarify the runtime specs, and retain qualification evidence.
2026-11-10 23:26:40 +00:00

74 lines
3.3 KiB
Markdown

# Runtime qualification: reopened memcard first-frame progress
Owner: runtime/headless. Provider correction and native qualification complete.
## Qualified installation
Use the extracted installation at:
```text
/Users/niltonconstantino/personal/workspace.personal/intrepid/prometeu/runtime/dist-staging/qualification/PLN-0202
```
Invoke its dispatcher as `prometeu headless --stdio`. It advertises the existing
protocol 1.2 with explicit `memcard` and `log_read` negotiation. No protocol schema,
guest ABI, guest budget or replay policy was changed.
SHA-256 of `prometeu-headless`:
```text
18423f798ab28bbd38c12a7ee46459f3ef3d08876cc64327952f57be041fc5df
```
Retained archive: `prometeu-cli-aarch64-apple-darwin.zip` in the installation above.
Archive SHA-256:
```text
05b0e38f797bc6d24cd69aeb7f6cad9ae224cb2308390a084634dfebe5e8f2f7
```
[Complete fingerprints](evidence/qualification/distribution.json) identify the
dispatcher, runtime binary, schemas and packaged spec. The selected stable
installation has the same binary fingerprints as before this work.
## Corrected behavior
When successful bootstrap is blocked on its native publication, the headless
owner waits for lane progress outside firmware/VM ticks, within the original
advance. Requested waits are at most 5 ms, with an immutable outer guard of 2 s.
Native cancellation at 500 ms and escalation at 1 s keep their original request
admission time and take precedence. Native media action and original errors retain
their existing classification and resource ownership.
The 100,000 firmware-tick watchdog remains. Ordinary asynchronous gameplay saves
do not invoke this bootstrap wait. Boot still stops before guest execution;
frame advancement and cancellation retain their published boundaries.
## Native evidence
- Controlled old optimized host: watchdog expires while presentation write is held.
- Corrected host: controlled completion, native error, queued publication,
blocked IO with retained lock, immutable outer deadline and unrelated gameplay
commit cases pass.
- Neutral writer creates/commits data; a new-process reader publishes presentation
and completes exactly one first `advance(1)` with correct prior payload, paused
state, no guest fault and canonical capture. Same/changed app versions pass.
- Original no-presentation reader still verifies readback without image mutation.
- Workspace: 1,125 passed, 16 ignored; scoped release: 89 passed, 4 ignored;
separately executed packaged cases: 4 passed, zero ignored/failures.
- Formatter and strict workspace clippy pass.
See [summary](evidence/summary.json), [packaged results](evidence/qualification/packaged.log)
and [source delta](evidence/qualification/implementation.patch). Qualification used
an uncommitted checkout based on `49d7fca0b725a0a261cda09bf5e924102c84fe31`;
fingerprints identify the qualified delivery independently of a commit.
## Consumer ownership
The Studio AI owns selecting this qualification installation in its own context
and rerunning its unchanged consumer pilot and runner/UI/MCP gates. Those gates
were not executed or declared passed by the runtime delivery. No Studio/game
sources, consumer configuration or selected stable install were changed here.
Provider closure is independent of consumer acceptance. This handoff is a local
artifact; no external message was sent automatically.