N1: the m2 precedence trusted region/address whenever the keys were present,
but the AdminUI model stamps defaulted HoldingRegisters/0 on every save, so a
string-only tag (CSV import, or any editor toggle) silently became HR[0] -
reads AND writes. The resolved keys are now authoritative only when the editor
marked them ("addressResolved": true, written by Resolve); an unmarked
addressString is always resolved through the parser with the driver's family.
The model no longer stamps address fields for a string-only tag, CSV export
leaves its address cells blank, and the editor hint distinguishes string-only
from pinned tags. Tests: CSV import and editor save of a string-only tag, a
stamped-default blob, and the pinned positive control across families.
n1: docs/v2/modbus-addressing.md updated (FC23 opt-in, maxReadGap semantics,
coalescing scope, addressString/family rules, coil coercion, refusals).
n2: form round-trip tests for UseFC23ForMultiRegisterWrites /
DisableReadCoalescing. n4: the editor keeps unresolved address text when the
parent echoes back its own emit. n3: noted, no change.
Refs #741
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
I-1: a PUT/GET-disabled CPU (every read denied) stayed Faulted with
LastSuccessfulRead frozen, so the comms-loss watchdog reconnected and
re-subscribed it every 90 s forever. Both access-denied arms now also record
an answer: Faulted is kept (worst state wins), the stamp advances, and the
sticky write fault still clears only on a real success. Pinned end to end in
Runtime.Tests: the real S7Driver under the real DriverInstanceActor watchdog
is not reconnected across 2x the threshold; the control (every read timing
out) is.
Minors:
- m-1: injectable pass clock (UtcNow); "LastSuccessfulRead advanced" is now
asserted strictly, including the write-refusal case.
- m-2: the editor uses S7TagShapeRules.IsByteAnchored instead of its own list.
- m-3: a release that races an init publish also cancels the probe loop the
publish started after the release took its snapshot.
- m-4: a retained poll failing during a lifecycle operation writes no health.
- m-5: a multi-var chunk that fell back memoises its members refused alone.
- m-7: access-denied needs S7.Net's exact wording and no transport fault in
the chain.
- m-8: sync teardown clears the write-side Degraded owner; a sync Dispose
racing a running shutdown is tested.
Refs #742
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
- The background probe reported the host Stopped whenever a read pass held
_gate past the probe deadline: contention is the connection being used,
not lost. A contended tick now gives no verdict (counted for tests); a
real probe failure still reports Stopped. A tick that races a teardown no
longer reopens a connection on an uninitialised driver.
- S7DriverProbe deserialised S7DriverOptions directly, whose Timeout is a
TimeSpan no config names, so "timeoutMs" was ignored and every handshake
waited the whole Test Connect budget. The handshake and test read are now
bounded by timeoutMs, capped by the caller's timeout.
Refs #742
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
A span's block is the union of its members' ranges, so one member that runs
past the DB's real end (or over-declares a STRING) made the CPU refuse the
whole block, and every neighbour within it was Bad on every poll. The
planner's stated invariant (a grouped read cannot fault where per-tag reads
would succeed) was false; the multi-var path already fell back, spans did not.
- A refused multi-member span hands its members back; they are read one at a
time for that pass, bypassing the multi-var pass.
- _noGroup remembers the members refused on their own (or every member when
none was) and keeps them out of later plans, so the rest group again. It is
cleared by every initialize publish. Non-refusal span failures are
unchanged (every member stamped).
- The once-per-tag refusal Warning memo (R07:H4) is now also cleared by the
publish, and shutdown resets the write-side Degraded owner, as documented.
- Live gate: MW10 + MD254 (2 bytes past the 256-byte M area) merged into one
refused block; MW10 keeps reading 4097 on every pass.
Refs #742
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
No read or write loop checked _plcDead between items: on a frozen-but-
established peer the first op blew its deadline and marked the handle dead,
and every remaining span / tag / write still went to the wire and could wait
its own deadline under _gate (a 20-span pass ~100 s), with Degraded and the
comms-loss watchdog's clock deferred until the pass ended.
- The span loop, the multi-var chunk loop, the per-tag loop and the write
loop stamp BadCommunicationError without I/O once _plcDead is set.
- A multi-var chunk that dies on a transport loss (socket / IO / timeout /
ConnectionError) marks the handle dead instead of falling back to one
doomed per-tag read per member. The framing class is deliberately excluded
there: snap7/S7.Net report a multi-var with one refused item as
ReadData(WrongNumberReceivedBytes) over a healthy stream.
- Live gate via an in-test freezable TCP proxy (counts request frames, not
seconds). Mutation shows S7.Net 0.20 already self-closes on a cancelled op,
so the gate pins the end-to-end property; the driver guard itself is proven
by the unit tests (an op that ignores its token keeps the socket open).
Refs #742
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
A deterministic PLC refusal (object does not exist, address out of range,
type inconsistent...) degraded driver health, and an all-refused driver (a
wrong DB number) sat Degraded with LastSuccessfulRead frozen, so the 90 s
comms-loss watchdog reconnected it forever against a reachable PLC.
- IsRefusal / IsAccessDenied classify over the whole exception chain. Live
probe on the snap7 fixture: S7.Net's READ paths raise the item error as a
BARE Exception ("Received error from PLC: ..."), its write paths as
PlcException(WriteData). The old PlcException-only match missed the read
shape entirely, so a real read access-denied was BadCommunicationError +
Degraded instead of the Faulted PUT/GET alert; the Test Connect probe had
the same blind spot.
- A refusal (read or write) is BadDeviceFailure with a once-per-tag Warning,
no health write and no reopen; it counts as answered, so it advances
LastSuccessfulRead and restores Healthy after a comms fault. An undecodable
buffer (S7ValueRangeException) also counts as answered.
- A write batch the CPU answered without a failure advances
LastSuccessfulRead and retires a Degraded the write side itself published.
- Tests that used ReadData as their generic "device fault" now use SendData
so their Degraded leg is still exercised.
Refs #742
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
Review m-4. The status-code paragraph still traced the value path
through the deleted per-value AttributeValuePublished. It now reads
AttributeValueBatchPublished -> AttributeValueBatchUpdate ->
IOpcUaAddressSpaceSink.WriteValues, which is what OpcUaPublishActor
calls.
Refs #731#734
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
Review m-2. A Calculation driver's own maxAllocBytesPerRun /
quarantineAfterConsecutiveTimeouts used to be clamped silently by
ScriptRunGuard. ResolveScripting now sanitizes the layered options
itself and returns the adjustments, and the factory logs one Warning
per clamped value, naming the driver. docs/ScriptEditor.md "Runtime
safety" gains the key table: defaults, the 1 MB allocation floor, the
1000 caps, NaN and disable semantics, and the per-driver override.
docs/Configuration.md has no Scripting section, so these keys have no
other home.
Refs #731#734
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
The init-time tag guards THREW: one mis-authored tag (DB1.DBZ0, DB1.DBW0 typed
Float64, T5 typed Int32, a writable Bool array) faulted the whole driver and the
actor retried the same init every 10 s forever, darkening every tag on the PLC.
Neither the editor nor the deploy gate checked any of it.
- S7AddressParser moves to S7.Contracts (same namespace) and a new
S7TagShapeRules there is the single authority on servable shapes.
- S7TagDefinitionFactory.FromTagConfig applies it, so the AdminUI editor, the
deploy gate (Inspect, with the specific reason) and the driver init refuse
the same tags. Init now skips a refused tag (ISkippedTagReporter) instead of
throwing; the throwing RejectUnsupportedTagConfigs is gone.
- New rules on top of the old guards: a narrow type at the wrong access width
(every read failed) and a Bool array not starting at bit 0 (U4: DB1.DBX0.3
silently read bit 0 of bytes 0..n).
- AdminUI test fixtures that used addresses the driver could never serve
(DB1.DBS0, a browse "SomeAddress", Int32/Float32 on DBW) now use real ones.
Refs #742
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
It had no production sender: every driver data change is already a
batch, and the comms-loss fan was converted earlier. DependencyMuxActor
drops its per-value receive, DriverInstanceActor drops the record, and
DependencyMuxActorTests' 13 sends move to the ValuePublishes.Single
one-item batch (same semantics through the batch path). XML docs and
the living docs (DriverLifecycle, VirtualTags, ScriptEditor) are
reworded; historical plan docs are left as written.
Refs #731#734
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
ScriptContext.CancellationToken is backed by an AsyncLocal that
ScriptEvaluator.RunAsync scopes around the compiled delegate, so it is
exactly the token ScriptRunGuard cancels on timeout. A polling script
can now end its own abandoned run and free the slot. AsyncLocal rather
than a field because the scripted-alarm engine reuses one context per
alarm: an orphan keeps seeing its own cancelled token. The scoping
entry point is internal (IVT to Core.Scripting) so a script cannot
replace its own token.
Editor parity: ScriptAnalysisService dot-completion now walks the
receiver's base chain (instance members only, stopping before
System.Object). Before this, a member declared only on ScriptContext
compiled and published but never completed. DiagnoseTests pins the
member as allow-listed; CompletionTests pins the completion.
Refs #731#734
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
Ruling refinement on I1. Only timeouts are strikes, so a never-returning
while(true), which times out once and is then only refused, was never
quarantined, and that defeated E1. A refusal now quarantines immediately
when the orphan it was refused behind has been executing for longer than
ScriptingOptions.StuckOrphanQuarantineFactor x the script's timeout
(default 10, so 20 s at the 2 s default; 0 or less disables it). It uses
the same quarantine path: one Error, otopcua.script.quarantined with
reason=stuck, Bad, lifted on source change or redeploy. A young orphan
(a short overrun) is still a plain refusal.
The guard stamps the run's start (Stopwatch timestamp) when the pool
picks it up and clears it when the run really finishes.
Tests: the 30-trigger while(true) regression (paced) now ends
quarantined with reason "stuck", still with exactly one orphan and a
healthy sibling. Positive control: a 2x overrun is refused, not
quarantined, and evaluates again once the orphan completes. A factor of
0 disables the rule. Docs (ScriptEditor "Runtime safety", VirtualTags)
and the seam remarks are updated.
Refs #734
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
- ScriptLogRateLimiter.LimitPerWindow. The ScriptLogTopicSink and
VirtualTagActor summaries now print the configured limit, not
DefaultLimitPerWindow. Tests assert "limit N/window" for a non-default
N.
- CalculationDriver.Dispose cancels, then takes _evalGate (bounded 2 s)
and disposes the gate and _evalCts. The pump and timer ticks enter the
gate through TryEnterEvalGateAsync, which stops cleanly once the
driver is disposed. Test: Dispose with a 6 s runaway run in flight
returns promptly, disposes both, and a late trigger is harmless.
- ScriptedAlarmEngine captures _shutdownToken once and disposes
_shutdownCts in Dispose when every drained task has completed. A task
stuck past the grace leaves the CTS to the GC. Lifecycle tests assert
the disposal.
- DraftValidator: CalculationNoEffectiveTrigger is skipped for a
change-triggered tag whose scriptId does not resolve. That case is
already CalculationScriptMissing or CalculationScriptNotFound, and
there are no reads to count. Both existing tests now assert the
absence (red before the change).
- docs/VirtualTags.md: System.Threading is denied as a whole namespace,
with an allow-list; the type list is now accurate (Unsafe,
RuntimeHelpers, and the ForbiddenMembers on System.Type); the resource
limits section covers the run guard and the allocation budget.
docs/ScriptEditor.md: accurate deny-list and a new "Runtime safety"
section covering timeout, refusal, quarantine after 3 timeouts,
allocation budget, lifting, and rate-limit summaries that appear only
on the source's next entry after a burst then quiet.
Refs #734
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
The hand-rolled bearer handler is replaced by AddJwtBearer (Microsoft.AspNetCore.Authentication.JwtBearer
10.0.7, pinned on the same servicing line as the other AspNetCore packages). Validation parameters come
from JwtTokenService.BuildValidationParameters, which now pins ValidAlgorithms = HS256 for both the
bearer scheme and TryValidate; MapInboundClaims is off so the minted Role/zb:username keys resolve.
/auth/token requires the cookie scheme, so a bearer can no longer mint a fresh token for itself.
Refs #733
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
BoundedApiKeyStore's negative cache was keyed by key id, so a flood of DISTINCT well-formed
otua_<random>_x credentials missed it every time and cost one SQL query per attempt under the
SDK's SessionManager-wide lock, no credential needed (R03:3.3).
- IApiKeyStore.ListKeyIdsAsync (default null = unsupported; ConfigDbApiKeyStore lists every
ApiKeyId): the bounded store keeps a known-id set refreshed every StoreCacheSeconds
(single-flight, failed refresh keeps the previous set) and refuses an unknown id without a store
read ({outcome=unknown_id}).
- A process-wide miss budget (token bucket, Security:ApiKeys:MissBurst=100 /
MissRefillPerSecond=20, ApiKeyStoreGuardOptions) bounds lookups that reach the store and miss
while no set is available; a hit refunds its token. Per-endpoint accounting needs the
OpcUaServer impersonation seam to carry the client address — follow-up.
- docs/security.md: the new knobs, and the known-cluster gate's documented fail-open window at
central boot (R03:4.2) plus the async PSK refresh (R03:3.4).
Refs #733
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT
The #687/F4 crowding re-read doubled its room up to 8 re-reads with no cap, so a pathological single-ms
cluster on a streaming gateway was streamed several times over for one page. The room is now capped at
ServerHistorian:MaxTieClusterOverfetch (plumbed into GatewayHistorianDataSource as maxPreWindowRows,
default 65536); past it, or once a read has already seen more, the read is BadTooManyOperations with
PreWindowRows set.
Item 1: IHistorianDataSource.MaxRawValuesPerRead (Gateway: learned ceiling - 16 = L; 0 elsewhere, guard-tested).
RawPageLimit plans every raw page on top of #692's lookahead: with L > 0 the read (page P + boundary ties k
+ lookahead) is <= L, P = min(N, L - k - 1) (uncapped: L - k - 1); no room => straight to the tie-cluster
over-fetch (never an uncapped or underflowing request). One re-plan + re-read when the read taught a new
limit or was refused for crowding pre-window rows (HistoryReadLimitExceededException.PreWindowRows): the
page shrinks by them. Over-fetch bound min(MaxTieClusterOverfetch, L - 1); slice size never 0.
Review fixes: F1 a window starting at/before the epoch counts AVEVA's synthetic row as pre-window (a
ceiling-truncated read was reported complete); F2 the page path disables ReverseRawReader's own re-run
(cold end-only ceiling 20/N=3000 cost 3,023 reads, now <= 40); F3 crowded look-back re-read around, not
refused; F4 the #687 crowding re-read room at least doubles (up to 8 re-reads), so a cluster of hundreds
just before a resume point no longer fails on a default gateway; a crowding refusal that leaves no room
is BadTooManyOperations, never a tick-by-tick cursor.
Item 2: ServerHistorian:EventReadHorizon (default 00:00:00 = exact, unchanged). When set, every fresh
HistoryReadEvents window's earlier bound is raised to now - horizon (both directions, all backends).
Lossy by design; documented.
Item 3 (native TimeAverage/Total re-check) stays blocked on historiangw#27.
A time inside a gap of a bursty tag costs a 64-row jump read landing in the next
burst plus a bisection back across the gap (~6 reads), so the 64 + 4/time cap
refused a 1 h/day 1 Hz tag asked at noon on 30/100/300 consecutive days
(186/606/1,806 reads) that AVEVA's own AtTime served. The cap is a runaway guard
(HistoryReadDeadline bounds wall clock): now 64 + 16 per distinct time.
Regression tests for x30/x100/x300 (each <= 7 reads a time, values checked).
Docs: DateTime-hinted nodes are stepped but keep the gateway's Double (the
Gateway never historizes DateTime nodes; the SQL backends return DateTime).
H1: the walk-back reset its width after a full segment and discarded a full
whole-remainder read, so a dense tag that stopped reporting weeks ago gave up
after 65 reads (BadTooManyOperations). It now gallops (x4, whole-remainder read
after two empties) and then BISECTS once any stored row is known below — a full
segment's (incl. a full whole-remainder read's) last row, or an earlier window
read's last row: (mid, hi] empty => hi = mid; full => known = its last row;
complete with rows => done; one density-aimed segment after a full read until an
aimed segment comes back empty; below 2 ms the whole stretch from the known row
(never a 100 ns step). Stopped 1 h..365 d with 5 y history: 6-18 reads at
1/min, 21-25 at 1 Hz, 25-31 at 10 Hz.
M1: a jump read asks for 64 rows (1,000 times over 10 y: ~66k rows, was ~1M).
L1: per-request read cap 64 + 4 x distinct times with a clear message; a strict
ceiling below 4 rows is refused naming the ceiling.
L2: DateTime hints are stepped. L3: under an integer hint a NaN/inf or
out-of-Int64-range value is unusable, never saturated Good.
L5: conformance Int64 AtTime asserts the type. L4: docs name the backend
differences. Nit: the window read end is clamped to now + 1 day.
Reviewer harnesses (Scratch693Fuzz/Cost, not committed): 0 differential diffs
over 2,400 cases.
F1 (pre-existing): SliceTieCluster continued a forward read while next <= endUtc, but the
forward window is end-exclusive; an oversized cluster one step below EndTime resumed AT
EndTime, the resume became a start == end point read and returned the EndTime samples as
data (microsecond: EndTime 8.1 us truncated to 8.0). Continue only while next is strictly
before EndTime truncated to the backend resolution.
Tests: A_tie_cluster_just_below_the_end_time_never_resumes_onto_the_end_time (native +
microsecond; both fail on the old rule) and
A_cursor_inside_a_tie_cluster_over_fetches_without_a_normal_read_first (pins the
WithinTieCluster shortcut by per-page read count; fails with the shortcut disabled).
Docs: the oversized-cluster exception to 'every page full / no lookahead => last' (F2),
and the N + skip + 1 strict-ceiling edge note referencing #694 (F4).
Review follow-up: on the Gateway every value is a Double (AVEVA stores a Boolean
as a 0/1 Int2), so the raw-evaluated AtTime interpolated a Boolean node to 0.5.
- IHistorianDataSource gains a default-interface overload
ReadAtTimeAsync(fullReference, timestamps, TypeCode? valueType, ct) that
forwards to the existing one (SqlServer/QuestDB/Null/fakes unchanged).
- OtOpcUaNodeManager.HistoryReadAtTime passes the node's scalar DataType
(AtTimeValueType: built-in type -> TypeCode; arrays/other types => null);
Raw and UNS NodeIds alike.
- GatewayAtTimeReader converts each raw row to the hinted kind BEFORE the shared
AtTimeEvaluator runs (Boolean = value != 0 => stepped; integers => Int64 =>
interpolated, rounded half away from zero, returned as the node's integer
type; floats Double), so the evaluator's own type branch decides.
- AtTimeEvaluator boxed every interpolated Int64 as a Double (`cond ? (long)x :
v` widens to double); its "keeping its type" test passed because Shouldly
ShouldBe compares numbers across types. Fixed (affects SqlServer/QuestDB too)
and pinned with ShouldBeOfType.
Gateway HistoryRead-AtTime passed AVEVA's own interpolation through: live it
answered the previous value as Good at a stored NULL's own time, null + Bad at
a Good sample right after a NULL, a drifted 99.99999999944 for a stored 100,
and no Interpolated bit. It is now served from raw gateway reads and evaluated
with Sql.Common's AtTimeEvaluator, as the SQL backends do (Part 11 §6.5.4).
GatewayAtTimeReader: one <=1,024-row forward window read [first-1ms, last+1ms]
for the batch, a walk-back (density-seeded, x4 growth, whole-remainder shortcut,
bisects dense history) for the first time's prev, and an 18-row probe for the
last time's next; truncated window reads continue or jump with a density-sized
margin. All bounds whole ms; a row counts only strictly inside its request
(AVEVA's boundary row never a sample); Bad rows value-less (#690); never past
now + 1 day; strict-gateway ceiling learned (#691.3) without re-probing.
10 dense times = 1 read (exact) / 3 (interpolated); 10 hourly = 12.
Kill switch ServerHistorian:GatewayAtTimeFromRaw (default true; false = the
old AVEVA passthrough). Known limit: the wire carries no data type, so numeric
tags interpolate as Double (Boolean/integer nodes are not stepped/rounded as on
the SQL backends).
A tie cluster's truncated read left the width at one tick; once the gap
above it was under two milliseconds nothing floored it, so segHi
stepped down one tick per read: 30 rows at one ms 0.3 s below hi with a
carried 1 s width took 10,006 identical reads (S8). (a) The top-anchored
width is floored at the granularity, Min kept; (b) a segment no wider
than the granularity is the unit that escalates to
MaxTieClusterOverfetch and then refuses - per millisecond, exact for the
ms-resolution Gateway (documented). Both are needed: the floor alone
loops on crowded reads with a small tie budget. S8 is a repo test
(4 shapes, reads <= 30, 0 repeated requests); random-data tests now
measure the refusal unit as rows in any 1 ms window. Reviewer harness,
5,000 seeds x default/MSONLY/CROWD: worst page 70 reads (default/MSONLY),
63 (CROWD) at M=65536, p99.9 27-37; no page over 500 reads at any tie
budget.
Bisecting the gap above a truncated read halved below the 1 ms request
granularity (requests are ms-rounded, so it re-sent identical requests)
and bisected the WHOLE gap even when the burst ended right above the
row it saw: a 1,026-row burst ending 100 days ago took 46 reads. After
an empty top-anchored segment the walk now gallops up from the known
row ((known + d, top], d x4, following each truncated read), then
bisects the bracket, with halving floored at the granularity (the Min
kept; a crowded gallop falls back to top-anchored halving). Burst: 18
reads, 0 repeated requests; dense tag stopped 2 days ago: 26 (gallop
alone crawled: 94). Walk fuzzed at 4,000 seeds locally (400 in-tree).
Docs: measured counts; CLAUDE.md gateway prerequisites list
RuntimeDb:EventReadMaxRows.
GatewayHistorianDataSource.ReadEventsAsync sized its result list to
maxEvents, which is int.MaxValue for NumValuesPerNode = 0: every
uncapped Gateway events read (forward on master, and the new walked
reverse one) threw 'Array dimensions exceeded' and counted as a backend
failure. Capacity is bounded at 1024. The events walk delegate moved to
ReverseEventsReader so it is tested over the real Gateway source
(forward uncapped, reverse uncapped, reverse capped).
A reverse or end-only capped events read on a backend without
SupportsReverseEvents (Gateway, Null) was refused
BadHistoryOperationUnsupported (explicit reverse) or answered with the
OLDEST N ascending (end-only, #675). It is now served newest-first by
the same ReverseWalk as raw reads: page-sized ms-aligned segments,
never answered from a read the source flagged truncated, ties sorted by
EventId (the gateway orders by event time only - the fake shuffles
them per call), one extra event walked to signal 'more', uncapped = one
forward read reversed, and the width + proven-empty stretch carried in
the events cursor. The oversized-cluster over-fetch keeps exactly the
events at T in the same order. A tie cluster over the budget is
BadTooManyOperations. Docs note the gateway's EventReadMaxRows must be
unlimited (it truncates silently).
The last pages of a deep reverse page-through on a forward-only backend
each walked down to 1970 one x4 growth step at a time (~14-16 reads,
~2.5 s live). After two empty segments the walk now asks once per page
for the whole remainder at the page budget (a truncated answer is
discarded and the walk carries on), and the stretch the page's last
complete read proved empty rides the continuation point
(HistoryContinuationState.EmptyFromTicks/EmptyToTicks, session-held, so
no versioning) so the next page steps over it. Fuzz now pages with the
walk's own hints and with true random empty stretches.
ReverseRawReader aimed every segment at half of MaxTieClusterOverfetch
(~32k rows by default) and carried that width page to page, so a paged
'latest N' read on a dense Gateway tag fetched ~32,770 rows per page
(2-5 s live) to return N. The walk now lives in a generic ReverseWalk
core: segments are aimed at ~N+8 rows (page budget 2N+16), a discovery
segment (first without a hint / after an empty one) may read 1,024, and
the tie budget is only the ceiling a single-timestamp cluster escalates
to. A truncated read never re-enters the stretch below its newest row
until the gap above it is bisected. Tests: per-page row counts on a
dense tag (node manager + walk), randomized exactness fuzz (walk and
node-manager paging), dense-tag-below-a-long-gap read bound.
Joseph Doherty