Design 47 H2: H-A-481 transfers finish or are taken over; H-A-482 a missing output is said

This commit is contained in:
2026-11-22 17:02:11 +01:00
parent 364552d2b2
commit c2712b5cd1
@@ -1005,3 +1005,11 @@ The lead's one-line answers to the questions H2's develop and review raise (impl
- **H-A-479 (third addendum, 18:45)** A rerun that would reopen an ended pipeline while another open pipeline carries
one of its movers at the same revision is refused with the same text ("its pipeline <key> is running"), so one entry
is never built side by side; keying reservations by pipeline is the next milestone's shape. P4.
- **H-A-481 (a transfer is finished or taken over, never left open, 19:00)** An artifact store's transfer (one open
publish, named by the artifact and its digest) ends when nothing is lacked: an opener that opens again with nothing
lacked finishes it; another caller with the same digest finishes it when nothing is lacked and takes it over when
the transfer's last part is older than two minutes; the idle bound is armed before the finish too, so a cut finish
never leaves a transfer open. The tool that lists transfers also shows each one's idle end. P-failure-loud, P3.
- **H-A-482 (a missing output is said as missing, 19:00)** A build module's output tool says "the ask kept no output"
when the output object is absent (an empty read is not a header fault), and a failed build step keeps its whole
standard error as its output object so the cause is readable after the fact. P-failure-loud.