docs: record reviewed close ownership and native attribution
This commit is contained in:
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,41 @@
|
||||
# Candidate close and weak-root control
|
||||
|
||||
The final bounded control at `564f10914357fc34db303b2faf030907663e6779`
|
||||
passed: one test executed, one passed, no skips; runner elapsed 10.082 seconds.
|
||||
It creates one abandoned browse point and six abandoned history points through
|
||||
the same encrypted canonical fixture helpers. Before close, the retained server
|
||||
session has exactly `(1,6)` points; SDK2's stronger post-close `(0,0)` check remains.
|
||||
|
||||
The old authentication token returns exactly `BadSessionIdInvalid` (`0x80250000`).
|
||||
A fresh session rejects both old continuation tokens with
|
||||
`BadContinuationPointInvalid`; invalid history does not call the historian.
|
||||
Fresh reads and the session's own browse/history roundtrips succeed around the
|
||||
denials, so a generally broken channel cannot satisfy this control.
|
||||
|
||||
After the session helper and its completed async state machine leave the isolated
|
||||
boundary, **all 23 old client/session/point/value/state targets are collectible
|
||||
while the same server is still running**, in each of five fixed GC observations.
|
||||
After normal host/application disposal and the fixture boundary, all 13 observed
|
||||
application/configuration/manager/validation-core/store/server/node-manager/
|
||||
backend targets are collectible in every round. Independent retained/collectible
|
||||
observer controls behave as expected. Both SDK2 certificate managers are disposed,
|
||||
cached stores are empty and actual PKI descriptors are zero.
|
||||
|
||||
Earlier one-history-point control receipts remain recorded: `b5986c61` passed
|
||||
with the original two-status auth whitelist; `6737e367` passed after narrowing to
|
||||
the observed exact status. Six affected history tests also passed at `6737e367`.
|
||||
Their launcher mistakenly expected seven tests. The original false guard receipt
|
||||
is retained; a separate source inventory and TRX review confirms four session
|
||||
history tests plus two event-paging tests, all executed/passed. No seventh source
|
||||
test is missing and the tests were not rerun to correct that bookkeeping error.
|
||||
|
||||
[Machine-readable observations, commands and raw hashes](close-ownership-candidate.json)
|
||||
retain all four executions. The current build has zero warnings/errors. Astra
|
||||
reviewed the design and actual code before each diagnostic execution; it verified
|
||||
all three close-control TRX files, the aligned 23/13 weak-target result and the
|
||||
six-test inventory correction independently.
|
||||
|
||||
This establishes bounded candidate close semantics and managed collectibility.
|
||||
It does **not** establish full canonical operation/host parity with the original
|
||||
standalone adapter, a steady RSS denominator, a resource-budget waiver, macOS
|
||||
native-retention resolution or G4 acceptance. No production code changed.
|
||||
Reference in New Issue
Block a user