Rotorsoft / act-root / 34356651662

100%

push

github

fix(act): treat a zero-length lane lease as no pinned lease on shutdown (#1657)

The derived shutdown budget is a fold over the lanes with a cycle in
flight: each contributes its configured leaseMillis, and a lane that
pinned no lease contributes drain's own 10s fallback instead. #1617
fixed the fold's seed, which started at 0 whenever a settle was the only
thing running. This is the same fold's other operand.

Inside the loop the lease was coalesced with `??`, which treats only
null and undefined as absence. Zero is neither, so a lane configured
`leaseMillis: 0` contributed a literal 0, the maximum came out 0, and
_await_inflight returned before awaiting the cycle it had just
enumerated. The handler is abandoned, its round of work discarded and
redelivered, and an acked listener never observes it. Only the default
path is affected, which is the path dispose()() uses.

A zero-length lease expires the instant it is granted, so it says
nothing about how long a handler may hold a stream. That is the
pinned-no-lease case spelled with a number rather than an omission, and
the method's own doc comment already said such a lane takes the
fallback. The code now agrees with the prose.

leaseMillis: 0 is legal configuration, not a degenerate cast:
LaneConfigSchema admits it and #1613 already made a zero-length lease
behave correctly at the store.

Explicit graceMs is unaffected — it bypasses the derivation entirely, so
graceMs: 0 still short-circuits.

Closes #1647


Claude-Session: https://claude.ai/code/session_01QPx2Qu7ZHDTSpoWYJvX4RJ

Co-authored-by: rotorsoft <rotorsoft@outlook.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

4353 of 4353 branches covered (100.0%)

Branch coverage included in aggregate %.

1 of 1 new or added line in 1 file covered. (100.0%)

9419 of 9419 relevant lines covered (100.0%)

1186.75 hits per line