4 Commits

Author SHA1 Message Date
Joseph Doherty 1d7a4b7a53 test: pass TestContext.Current.CancellationToken at xUnit1051 call sites (analyzer code fix, 451 sites) (lmxopcua#725)
Claude-Session: https://claude.ai/code/session_01GT67AEHd2Acv1YPwGfXEoV
2026-11-09 13:19:34 -05:00
Joseph Doherty 459cbb444e fix(abcip): review fix round 1 - BOOL-array fallback, local refusals, minors
I1: the U11 Bits[n] -> Bits[n/32] rewrite was unconditional. The native
runtime now keeps the original element path until the first reply settles the
layout: 0xD3 confirms DWORD addressing (bit n%32); any other reply type (e.g.
a 0xC1 per-element BOOL array) or a refusal of the DWORD index falls back to
Bits[n] and re-reads, and a write settles the layout by a read first, so it can
never decode or write element n/32 of an element-addressed array.

I2: libplctag ErrorNoMatch/ErrorBadDevice no longer fold into NotFound (they
now read as BadConnection: comms, Degraded, LastSuccessfulRead preserved).
ErrorNotImplemented - and the native runtime's own pre-send refusals - raise
AbCipLocalRefusalException: BadNotSupported, no health, LastSuccessfulRead NOT
advanced, no backoff record, not reachable for the probe or Test Connect.

M1: the bit RMW's parent read/write status words follow the health contract
(refusal: per-tag Bad + Warning; comms: evict the parent + Degraded).
M2: device Unsupported read refusals return Refused (advance
LastSuccessfulRead); device write refusals call RecordWriteAnswered (not the
pre-send native type gate).
M3: a good write clears a Degraded a write comms fault recorded (a read-side
Degraded is still left for the read path).
M4: a superseded init throws AbCipInitializationSupersededException (not an
OCE); ShutdownAsync stamps Unknown only if no newer init published meanwhile.
M5: the sync-Dispose timing assert is 2 s against a 4 s probe read; the
unencodable-array test asserts health.

Refs #737
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
2026-10-22 14:53:06 -04:00
Joseph Doherty aabe027d10 test(abcip): native end-to-end proof a NotFound tag causes no lease churn
L4 evidence over the real native runtime and a fake CIP target: a renamed tag
that is the device's only lease holder is created once and Forward-Opened once
across five polls, with no Forward Close, every poll BadNodeIdUnknown and the
driver Healthy.

Refs #737
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
2026-10-22 14:32:31 -04:00
Joseph Doherty f087f01d2b fix(abcip): classify every controller refusal as per-tag Bad, keep the runtime
H1/L4: one classification (AbCipRefusals.IsDeviceRefusal) of the statuses
that prove the controller answered - NotFound, OutOfBounds, NotAllowed,
TooLarge, TooSmall, TypeMismatch, Unsupported - shared by the data path, the
Test-Connect probe (IsReachableStatus widened) and the native runtime.

- A refused read/write (thrown or as a status word) is per-tag Bad with a
  once-per-tag Warning, keeps the runtime (no evict, so no lease/Forward-Open
  churn for a sole lease holder), writes no Degraded and, since the device
  answered, advances LastSuccessfulRead without touching the state.
- A refused cold create records the backoff as reachable, so siblings are no
  longer throttled for up to 30 s by one Arr[999].
- The native runtime completes InitializeAsync on any refusal (it used to on
  TypeMismatch/Unsupported only) and reports it on every later read.
- A refused READ of NotAllowed reads BadNotReadable, not BadNotWritable.
- Test Connect names the refusal ("probe tag refused: OutOfBounds").

0x1C/0x20/0xFF-0x2104 still fold into BadReply inside Driver.Ab.Transport
(Lane K follow-up); they remain comms faults until mapped there.

Refs #737
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
2026-10-22 14:07:50 -04:00