Fixes #284 (plan S1).
A first-entry placement that could not be prepared returned
RetrySetupUnavailable and was re-Advanced every pump forever. Nothing counted
it, nothing named its cause, and nothing distinguished "waiting for something
that will arrive" from "waiting for something that never can". That is why
#281's 43 test failures presented as four unrelated symptoms across App and
Runtime instead of one cause, and why a stuck entity in the live client simply
never appears with no log line to follow.
Worse, the two causes were conflated: 670f307c's missing-world-frame park
reported itself as RetrySetupUnavailable, sending anyone diagnosing it to the
prepared-asset pipeline rather than to the absent local-player Create that
actually publishes the frame.
- RetryWorldFrameUnavailable splits the two causes. Call sites now ask
IsRetryable() instead of comparing against one reason, so a future retry
reason cannot be silently reclassified as a hard rejection - the exact way
this class of bug hides.
- The operation retains its RuntimeSetPositionParkReason, and
RuntimeSetPositionOwnershipSnapshot reports parked work by cause
(ParkedAwaitingSetupCollisionCount / ParkedAwaitingWorldFrameCount /
ParkedPlacementCount), so parked placements appear wherever ledgers are
already asserted.
- ObserveLocalPlayerCreate records the accepted local-player Create even when
it carries no landblock - precisely the case where no frame is ever
published - and ThrowIfWorldFrameUnreachable makes that contradiction
terminal. Waiting is legitimate only while that Create is outstanding; after
it, no later pump can supply the frame. Same shape as 01f4791e, which made a
violated receipt-ledger invariant terminal rather than resumable.
This is observability plus fail-fast. There is no timeout, no retry cap, and
no grace period anywhere in it; retryable work still retries exactly as before
and no placement behaviour changed.
The parked counts are deliberately NOT folded into IsConverged: #277 documents
a far Create legitimately parking for a whole session, so a parked entry at
teardown is not automatically a defect. Wiring them into the connected gates
is carried with #277's service-window conversion, where "legitimately parked"
becomes definable.
Runtime 1,012/1,012. Complete Release solution: 10,834 passed / 4 skipped /
0 failed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
671 lines
32 KiB
C#
671 lines
32 KiB
C#
using AcDream.Content;
|
|
using AcDream.Runtime.Entities;
|
|
using AcDream.Runtime.Physics;
|
|
|
|
namespace AcDream.Runtime.Gameplay;
|
|
|
|
/// <summary>
|
|
/// Typed yields for <see cref="RuntimeLocalPlayerFirstEntryState.Advance"/>.
|
|
/// Mirrors the executor's own <c>RuntimeInitialCreateExecutionStatus</c>
|
|
/// shape (terminal Completed/RejectedToken/RejectedAuthority plus named
|
|
/// retryable yields) rather than inventing a parallel vocabulary.
|
|
/// </summary>
|
|
internal enum RuntimeLocalPlayerFirstEntryStatus : byte
|
|
{
|
|
/// <summary>
|
|
/// The underlying <see cref="RuntimeInitialCreateContinuationExecutor.Execute"/>
|
|
/// call reported <c>Completed</c>: residence consumed, initial tail and
|
|
/// FIFO drained, ExecutorCompleted receipt dispatched on the placement
|
|
/// stream. Terminal; the conductor's own progress entry is removed.
|
|
/// </summary>
|
|
Completed,
|
|
|
|
/// <summary>
|
|
/// The authored-mover Setup read
|
|
/// (<see cref="RuntimeSetPositionState.TryPrepareAuthoredMover"/>) is not
|
|
/// yet available (<c>RetrySetupUnavailable</c>) — the prepared-asset
|
|
/// package has not landed. Retry <see cref="RuntimeLocalPlayerFirstEntryState.Advance"/>
|
|
/// with the same arguments once it has; no Runtime state changed.
|
|
/// </summary>
|
|
AwaitingCollisionSource,
|
|
|
|
/// <summary>
|
|
/// <see cref="RuntimeLocalPlayerPhysicsPublicationState.EvaluateActivation"/>
|
|
/// or <see cref="RuntimeLocalPlayerPhysicsPublicationState.CommitActivation"/>
|
|
/// yielded <c>DeferredCell</c> or a retryable <c>RejectedPlacement</c> —
|
|
/// the destination cell's collision generation is not ready, or the
|
|
/// placement needs re-evaluation after some other change. Retry later
|
|
/// (e.g. on a collision-generation wake); the dormant activation lease
|
|
/// itself remains intact and is re-driven from the same stage.
|
|
/// </summary>
|
|
AwaitingActivation,
|
|
|
|
/// <summary>
|
|
/// The activation committed but its <c>Place</c> projection has not been
|
|
/// acknowledged yet — either <see cref="RuntimeSetPositionState.AcknowledgeProjection"/>
|
|
/// was not (yet) the exact FIFO head, or the subsequent
|
|
/// <see cref="RuntimeInitialCreateContinuationExecutor.Execute"/> call
|
|
/// still observed <c>PendingPlacement</c> from
|
|
/// <see cref="RuntimeInitialCreateResidenceState.Complete"/>. Retry the
|
|
/// same stage.
|
|
/// </summary>
|
|
AwaitingReceiptAcknowledgement,
|
|
|
|
/// <summary>
|
|
/// Passthrough of the executor's own <c>AwaitingContinuationPlacement</c>
|
|
/// — a later FIFO continuation (a Position update accepted while this
|
|
/// entity's initial placement was in flight) needs its own authored
|
|
/// placement prepared/submitted/acknowledged before the drain can
|
|
/// finish. Entirely the executor's own concern from this point forward;
|
|
/// the conductor's job (residence -> publication -> Execute) is done as
|
|
/// soon as it reaches this yield.
|
|
/// </summary>
|
|
AwaitingContinuationPlacement,
|
|
|
|
/// <summary>
|
|
/// A reentrant <see cref="RuntimeLocalPlayerFirstEntryState.Advance"/>
|
|
/// call for the SAME entity arrived while an outer call for it was still
|
|
/// on the stack (mirrors the executor's own <c>_executing</c> fail-closed
|
|
/// guard). Not a Runtime-state rejection — retry once the outer call has
|
|
/// returned.
|
|
/// </summary>
|
|
Contention,
|
|
|
|
/// <summary>The residence/placement token no longer matches anything tracked.</summary>
|
|
RejectedToken,
|
|
|
|
/// <summary>
|
|
/// An authority-shaped failure (stale epoch/session/identity, deleted or
|
|
/// replaced record, GUID reuse, disposed identity, or an inner
|
|
/// currency check failing during a reentrant callback). Abandoned: the
|
|
/// conductor's own in-flight publication candidate/activation (if any)
|
|
/// is discarded through the shared choke points and its progress entry
|
|
/// is removed. The caller must begin a fresh first-entry sequence
|
|
/// (a new residence lease) rather than retry this exact call.
|
|
/// </summary>
|
|
RejectedAuthority,
|
|
}
|
|
|
|
internal readonly record struct RuntimeLocalPlayerFirstEntryOwnershipSnapshot(
|
|
int ActiveCount)
|
|
{
|
|
internal bool IsConverged => ActiveCount == 0;
|
|
}
|
|
|
|
/// <summary>
|
|
/// The dormant, resumable Runtime transaction that dissolves C3's Finding B:
|
|
/// the local player's initial residence lease opens its SetPosition operation
|
|
/// at Create time, but nothing wires the mover-preparation ->
|
|
/// <see cref="RuntimeLocalPlayerPhysicsPublicationState"/> body/controller
|
|
/// attach -> placement acknowledgement -> FIFO drain sequence together into
|
|
/// one driveable state machine. This class ORCHESTRATES the existing,
|
|
/// already-tested residence
|
|
/// (<see cref="RuntimeInitialCreateResidenceState"/>), publication
|
|
/// (<see cref="RuntimeLocalPlayerPhysicsPublicationState"/>), mover
|
|
/// (<see cref="RuntimeSetPositionState.TryPrepareAuthoredMover"/>), and
|
|
/// executor (<see cref="RuntimeInitialCreateContinuationExecutor"/>)
|
|
/// machinery — it reimplements none of their validation and bypasses none of
|
|
/// their staged semantics.
|
|
///
|
|
/// Required order (campaign handoff route-1,
|
|
/// docs/research/2026-07-31-remaining-physics-campaign-handoff.md:280-292):
|
|
/// residence Begin (already done at registration, before this class is ever
|
|
/// invoked) -> authored-mover preparation (Setup read + PrepareMover, which
|
|
/// MUST precede publication Prepare —
|
|
/// <see cref="RuntimeLocalPlayerPhysicsPublicationState.Prepare"/>'s own
|
|
/// <c>CanPrepare</c> gate requires
|
|
/// <see cref="RuntimeSetPositionState.IsExactPreparedPlacementCurrent"/> to
|
|
/// already be true) -> publication Prepare -> publication Commit (the
|
|
/// <see cref="RuntimeSetPositionState.PrepareDormantLocalActivationOwnership"/>
|
|
/// seam attaches the body) -> activation Evaluate/Commit/Finalize ->
|
|
/// Place-receipt acknowledgement -> <see cref="RuntimeInitialCreateContinuationExecutor.Execute"/>
|
|
/// (FIFO drain) -> ExecutorCompleted receipt.
|
|
///
|
|
/// Once <see cref="RuntimeSetPositionState.PrepareDormantLocalActivationOwnership"/>
|
|
/// sets an operation's <c>DormantLocalActivation</c> flag,
|
|
/// <see cref="RuntimeSetPositionState.SubmitPreparedPlacement"/> (and the
|
|
/// fused <see cref="RuntimeSetPositionState.TryPrepareAndSubmitAuthoredPlacement"/>)
|
|
/// must never be called against it again — see <c>RetryDeferred</c>'s "must
|
|
/// never bypass that path through the ordinary remote CommitCanonical tail"
|
|
/// comment — so this class calls the mover-only
|
|
/// <see cref="RuntimeSetPositionState.TryPrepareAuthoredMover"/> half instead
|
|
/// and never the fused method.
|
|
///
|
|
/// PRODUCTION-DRIVEN since the C3c flip: <see cref="RuntimeEntityObjectLifetime"/>
|
|
/// fully constructs and wires this class (construction, publication binding,
|
|
/// retirement fan-out, bulk session-clear cleanup, ownership fold) exactly
|
|
/// like every other owner it builds, and the host first-entry drive
|
|
/// (<c>RuntimeFirstEntryDriveController</c>, pumped by the graphical
|
|
/// hydration/frame-retry cadence and the headless spawn/position/tick
|
|
/// cadence) calls <see cref="Advance"/> for every local-player
|
|
/// initial-create residence.
|
|
/// </summary>
|
|
internal sealed class RuntimeLocalPlayerFirstEntryState
|
|
{
|
|
private enum Stage : byte
|
|
{
|
|
/// <summary>No progress yet, or the mover has not been prepared.</summary>
|
|
AwaitingMoverPreparation,
|
|
|
|
/// <summary>Mover command in hand; publication Prepare+Commit not run yet.</summary>
|
|
MoverPrepared,
|
|
|
|
/// <summary>
|
|
/// Publication Prepare+Commit succeeded (the body/controller are
|
|
/// attached to the canonical record); Evaluate+CommitActivation not
|
|
/// yet reached <c>Committed</c>. Also the retry point for
|
|
/// <c>DeferredCell</c>/<c>RejectedPlacement</c>.
|
|
/// </summary>
|
|
PublicationCommitted,
|
|
|
|
/// <summary>
|
|
/// CommitActivation reached <c>Committed</c>; the Place projection is
|
|
/// known but not yet acknowledged.
|
|
/// </summary>
|
|
ActivationCommitted,
|
|
|
|
/// <summary>
|
|
/// The Place projection has been acknowledged. Only
|
|
/// <see cref="RuntimeInitialCreateContinuationExecutor.Execute"/>
|
|
/// remains; re-acknowledging the same (already-consumed) token would
|
|
/// fail, so this stage is never re-entered by the acknowledgement
|
|
/// step.
|
|
/// </summary>
|
|
Acknowledged,
|
|
}
|
|
|
|
private sealed class Progress
|
|
{
|
|
internal required ulong LeaseId { get; init; }
|
|
internal Stage Stage { get; set; } = Stage.AwaitingMoverPreparation;
|
|
internal RuntimeSetPositionCommand PreparedCommand { get; set; }
|
|
internal RuntimeLocalPlayerPhysicsPublicationToken PublicationToken
|
|
{ get; set; }
|
|
internal RuntimeLocalPlayerPhysicsActivationToken ActivationToken
|
|
{ get; set; }
|
|
internal RuntimePlacementProjectionToken Projection { get; set; }
|
|
}
|
|
|
|
private readonly RuntimeInitialCreateResidenceState _residences;
|
|
private readonly RuntimeInitialCreateContinuationExecutor _executor;
|
|
private readonly RuntimePhysicsState _physics;
|
|
private RuntimeLocalPlayerPhysicsPublicationState? _publication;
|
|
private readonly Dictionary<RuntimeEntityKey, Progress> _progress = [];
|
|
private readonly HashSet<RuntimeEntityKey> _executing = [];
|
|
|
|
/// <summary>
|
|
/// F2: <see cref="RuntimeEntityObjectLifetime"/> constructs this class
|
|
/// (alongside the residence and executor it also owns) BEFORE
|
|
/// <see cref="RuntimeLocalPlayerPhysicsPublicationState"/> exists —
|
|
/// <c>GameRuntime</c> creates <c>RuntimeLocalPlayerMovementState</c> and
|
|
/// attaches its physics publication only after the entity-object lifetime
|
|
/// is already built. This mirrors the SAME late-bind pattern already
|
|
/// used throughout this class family (<c>BindGeneration</c>,
|
|
/// <c>BindRetirementNotification</c>, <c>BindLiveInputs</c>,
|
|
/// <c>RuntimeLocalPlayerMovementState.PhysicsPublication</c>'s own
|
|
/// throws-if-unbound accessor) rather than requiring the caller to
|
|
/// construct things out of their natural order.
|
|
/// </summary>
|
|
internal RuntimeLocalPlayerFirstEntryState(
|
|
RuntimeInitialCreateResidenceState residences,
|
|
RuntimeInitialCreateContinuationExecutor executor,
|
|
RuntimePhysicsState physics)
|
|
{
|
|
_residences = residences
|
|
?? throw new ArgumentNullException(nameof(residences));
|
|
_executor = executor
|
|
?? throw new ArgumentNullException(nameof(executor));
|
|
_physics = physics
|
|
?? throw new ArgumentNullException(nameof(physics));
|
|
}
|
|
|
|
internal void BindPublication(
|
|
RuntimeLocalPlayerPhysicsPublicationState publication)
|
|
{
|
|
ArgumentNullException.ThrowIfNull(publication);
|
|
if (_publication is not null)
|
|
{
|
|
throw new InvalidOperationException(
|
|
"The local-player first-entry conductor's publication owner is already bound.");
|
|
}
|
|
_publication = publication;
|
|
}
|
|
|
|
private RuntimeLocalPlayerPhysicsPublicationState Publication =>
|
|
_publication ?? throw new InvalidOperationException(
|
|
"The local-player first-entry conductor's publication owner is not yet bound.");
|
|
|
|
/// <summary>
|
|
/// One resumable step. Callers pass the SAME arguments on every retry;
|
|
/// this method re-reads currency from the owning states on every entry
|
|
/// rather than trusting anything cached beyond its own stage cursor and
|
|
/// the exact token/receipt/projection structs each owning method itself
|
|
/// requires as arguments — there is no other source for those; they are
|
|
/// the "exact keys" this class carries, not a second copy of any owning
|
|
/// state's internal record.
|
|
/// </summary>
|
|
internal RuntimeLocalPlayerFirstEntryStatus Advance(
|
|
RuntimeEntityRecord record,
|
|
in RuntimeInitialCreateResidenceToken residenceToken,
|
|
PlayerMovementConstructionOptions options,
|
|
in RuntimeLocalPlayerPhysicsActivationPreparation activationPreparation,
|
|
IPreparedCollisionSource collisionSource,
|
|
double gameTime,
|
|
in RuntimeInitialCreateExecutionInputs inputs,
|
|
out RuntimeInitialCreateExecutionReceipt receipt)
|
|
{
|
|
ArgumentNullException.ThrowIfNull(record);
|
|
ArgumentNullException.ThrowIfNull(collisionSource);
|
|
receipt = default;
|
|
if (!residenceToken.IsValid || record.Key is not { } key)
|
|
return RuntimeLocalPlayerFirstEntryStatus.RejectedToken;
|
|
|
|
// Mirrors RuntimeInitialCreateContinuationExecutor.Execute's own
|
|
// _executing.Add(key) guard: a synchronous reentrant call for the
|
|
// SAME entity (e.g. from a collision-report/placement observer
|
|
// invoked mid-Advance) fails closed rather than interleaving two
|
|
// drains of the same stage machine.
|
|
if (!_executing.Add(key))
|
|
return RuntimeLocalPlayerFirstEntryStatus.Contention;
|
|
try
|
|
{
|
|
return AdvanceCore(
|
|
record,
|
|
residenceToken,
|
|
options,
|
|
activationPreparation,
|
|
collisionSource,
|
|
gameTime,
|
|
inputs,
|
|
key,
|
|
out receipt);
|
|
}
|
|
finally
|
|
{
|
|
_executing.Remove(key);
|
|
}
|
|
}
|
|
|
|
private RuntimeLocalPlayerFirstEntryStatus AdvanceCore(
|
|
RuntimeEntityRecord record,
|
|
in RuntimeInitialCreateResidenceToken residenceToken,
|
|
PlayerMovementConstructionOptions options,
|
|
in RuntimeLocalPlayerPhysicsActivationPreparation activationPreparation,
|
|
IPreparedCollisionSource collisionSource,
|
|
double gameTime,
|
|
in RuntimeInitialCreateExecutionInputs inputs,
|
|
RuntimeEntityKey key,
|
|
out RuntimeInitialCreateExecutionReceipt receipt)
|
|
{
|
|
receipt = default;
|
|
|
|
// H2: fail transactionally, before any state mutation, if
|
|
// Publication has not been bound yet. Without this upfront check,
|
|
// an unbound call could still get as far as authored-mover
|
|
// preparation (which mutates RuntimeSetPositionState's own
|
|
// _preparedMovers) and creating THIS class's own Progress entry
|
|
// (stored into _progress) before the first Publication dereference
|
|
// (inside the MoverPrepared stage below) throws — leaving a
|
|
// poisoned Progress entry that a later, unrelated Discard/DiscardAll
|
|
// call (from a retirement notification or session-clear fan-out)
|
|
// would ALSO throw on. Referencing the accessor here throws
|
|
// immediately with nothing yet mutated.
|
|
_ = Publication;
|
|
|
|
_progress.TryGetValue(key, out Progress? progress);
|
|
// ABA/GUID-reuse guard, exactly like the executor's own Progress
|
|
// reconciliation: an existing entry for a DIFFERENT (older or
|
|
// reused) lease id can never be resumed by this call.
|
|
if (progress is not null && progress.LeaseId != residenceToken.LeaseId)
|
|
{
|
|
Discard(key);
|
|
progress = null;
|
|
}
|
|
|
|
if (progress is null || progress.Stage is Stage.AwaitingMoverPreparation)
|
|
{
|
|
if (!_residences.TryGetCurrent(
|
|
record,
|
|
out RuntimeInitialCreateResidenceLease lease)
|
|
|| lease.Token != residenceToken)
|
|
{
|
|
// No progress was ever tracked for this key under this exact
|
|
// lease — mirrors RuntimeInitialCreateResidenceState.Complete's
|
|
// own convention (a token matching neither its active nor its
|
|
// completed table is RejectedToken, not RejectedAuthority).
|
|
// Only abandon (RejectedAuthority) when THIS class was
|
|
// actually tracking in-flight publication/activation state
|
|
// that must now be discarded.
|
|
if (progress is null)
|
|
return RuntimeLocalPlayerFirstEntryStatus.RejectedToken;
|
|
Discard(key);
|
|
return RuntimeLocalPlayerFirstEntryStatus.RejectedAuthority;
|
|
}
|
|
|
|
if (!lease.Route.PerformsSetPosition)
|
|
{
|
|
// Parented/PickedUp residence — never true for a real
|
|
// login, but kept for structural completeness: no
|
|
// SetPosition operation exists at all, so there is nothing
|
|
// for the publication chain to attach a body to. Skip
|
|
// straight to Execute (idempotent/retryable on its own).
|
|
progress ??= new Progress { LeaseId = residenceToken.LeaseId };
|
|
progress.Stage = Stage.Acknowledged;
|
|
_progress[key] = progress;
|
|
return RunExecute(record, residenceToken, inputs, key, out receipt);
|
|
}
|
|
|
|
RuntimeSetPositionMoverPreparationStatus moverStatus = _physics
|
|
.SetPosition.TryPrepareAuthoredMover(
|
|
record,
|
|
lease.Placement,
|
|
lease.Route.OperationKind,
|
|
lease.Route.SetPositionFlags,
|
|
collisionSource,
|
|
gameTime,
|
|
out RuntimeSetPositionCommand command);
|
|
// #284: every retryable reason resumes on a later pump. Comparing
|
|
// against one reason would silently reclassify a new one as a
|
|
// hard rejection.
|
|
if (moverStatus.IsRetryable())
|
|
{
|
|
return RuntimeLocalPlayerFirstEntryStatus
|
|
.AwaitingCollisionSource;
|
|
}
|
|
if (moverStatus != RuntimeSetPositionMoverPreparationStatus.Prepared)
|
|
{
|
|
if (progress is not null)
|
|
Discard(key);
|
|
return RuntimeLocalPlayerFirstEntryStatus.RejectedAuthority;
|
|
}
|
|
|
|
progress ??= new Progress { LeaseId = residenceToken.LeaseId };
|
|
progress.PreparedCommand = command;
|
|
progress.Stage = Stage.MoverPrepared;
|
|
_progress[key] = progress;
|
|
}
|
|
|
|
if (progress.Stage is Stage.MoverPrepared)
|
|
{
|
|
if (!_residences.TryGetCurrent(
|
|
record,
|
|
out RuntimeInitialCreateResidenceLease lease)
|
|
|| lease.Token != residenceToken)
|
|
{
|
|
Discard(key);
|
|
return RuntimeLocalPlayerFirstEntryStatus.RejectedAuthority;
|
|
}
|
|
|
|
RuntimeLocalPlayerPhysicsPublicationStatus prepareStatus =
|
|
Publication.Prepare(
|
|
record,
|
|
lease.Placement,
|
|
progress.PreparedCommand,
|
|
options,
|
|
activationPreparation,
|
|
out RuntimeLocalPlayerPhysicsPublicationToken pubToken);
|
|
if (prepareStatus
|
|
!= RuntimeLocalPlayerPhysicsPublicationStatus.Prepared)
|
|
{
|
|
// Prepare mutates nothing canonical on rejection (its own
|
|
// second CanPrepare recheck discards any just-built
|
|
// candidate itself); there is nothing further for this
|
|
// class to undo beyond dropping its own progress entry.
|
|
Discard(key);
|
|
return RuntimeLocalPlayerFirstEntryStatus.RejectedAuthority;
|
|
}
|
|
|
|
progress.PublicationToken = pubToken;
|
|
|
|
RuntimeLocalPlayerPhysicsPublicationStatus commitStatus =
|
|
Publication.Commit(
|
|
pubToken,
|
|
out RuntimeLocalPlayerPhysicsActivationToken activationToken);
|
|
if (commitStatus != RuntimeLocalPlayerPhysicsPublicationStatus.Committed)
|
|
{
|
|
Discard(key);
|
|
return commitStatus
|
|
is RuntimeLocalPlayerPhysicsPublicationStatus.RejectedToken
|
|
? RuntimeLocalPlayerFirstEntryStatus.RejectedToken
|
|
: RuntimeLocalPlayerFirstEntryStatus.RejectedAuthority;
|
|
}
|
|
|
|
progress.ActivationToken = activationToken;
|
|
progress.Stage = Stage.PublicationCommitted;
|
|
}
|
|
|
|
if (progress.Stage is Stage.PublicationCommitted)
|
|
{
|
|
// A single combined retry point for Evaluate+CommitActivation.
|
|
// Every publication test that hits DeferredCell/RejectedPlacement
|
|
// WITHOUT an intervening CommitActivation call chains both calls
|
|
// together and retries both together; running EvaluateActivation
|
|
// again before every CommitActivation retry is safe even for
|
|
// CommitActivation's own internal AwaitingFinalShadowPreparation
|
|
// resumption (its top-of-method check re-validates
|
|
// activation.Receipt == receipt, which a fresh Evaluate call
|
|
// satisfies, before consulting the untouched stored
|
|
// PendingFinalCommit).
|
|
//
|
|
// EvaluateActivation's DeferredCell status is overloaded: once a
|
|
// PRIOR CommitActivation call has already registered this lease
|
|
// as awaiting a specific cell/collision generation
|
|
// (IsDormantLocalActivationAwaitingCell), a REPEATED
|
|
// EvaluateActivation call that is still not ready returns
|
|
// DeferredCell WITHOUT ever populating receipt (it stays
|
|
// default/invalid) — see
|
|
// DeferredCommitWaitsThenRearmsSameLeaseAfterExactGenerationWake
|
|
// in the publication test suite, which asserts exactly
|
|
// `waiting.IsValid == false` on that repeat call and never feeds
|
|
// it to CommitActivation. Calling CommitActivation with that
|
|
// invalid receipt would hit its own `!receipt.IsValid` guard and
|
|
// incorrectly report RejectedAuthority instead of "still
|
|
// waiting" — so this class must check validity first and simply
|
|
// yield AwaitingActivation again without calling CommitActivation
|
|
// at all in that case.
|
|
RuntimeLocalPlayerPhysicsActivationStatus evalStatus =
|
|
Publication.EvaluateActivation(
|
|
progress.ActivationToken,
|
|
out RuntimeLocalPlayerPhysicsActivationReceipt evalReceipt);
|
|
if (evalStatus is
|
|
RuntimeLocalPlayerPhysicsActivationStatus.RejectedToken
|
|
or RuntimeLocalPlayerPhysicsActivationStatus.RejectedAuthority)
|
|
{
|
|
Discard(key);
|
|
return evalStatus
|
|
is RuntimeLocalPlayerPhysicsActivationStatus.RejectedToken
|
|
? RuntimeLocalPlayerFirstEntryStatus.RejectedToken
|
|
: RuntimeLocalPlayerFirstEntryStatus.RejectedAuthority;
|
|
}
|
|
if (!evalReceipt.IsValid)
|
|
{
|
|
// DeferredCell with nothing to commit — the destination
|
|
// cell/collision generation genuinely is not resolvable yet.
|
|
return RuntimeLocalPlayerFirstEntryStatus.AwaitingActivation;
|
|
}
|
|
|
|
RuntimeDormantSetPositionCommitStatus commitActivationStatus =
|
|
Publication.CommitActivation(
|
|
evalReceipt,
|
|
out RuntimePlacementProjectionToken projection);
|
|
switch (commitActivationStatus)
|
|
{
|
|
case RuntimeDormantSetPositionCommitStatus.Committed:
|
|
progress.Projection = projection;
|
|
progress.Stage = Stage.ActivationCommitted;
|
|
break;
|
|
case RuntimeDormantSetPositionCommitStatus.DeferredCell:
|
|
case RuntimeDormantSetPositionCommitStatus.RejectedPlacement:
|
|
// Stage stays PublicationCommitted — retry re-runs both
|
|
// EvaluateActivation and CommitActivation next Advance.
|
|
return RuntimeLocalPlayerFirstEntryStatus.AwaitingActivation;
|
|
default:
|
|
Discard(key);
|
|
return RuntimeLocalPlayerFirstEntryStatus.RejectedAuthority;
|
|
}
|
|
}
|
|
|
|
if (progress.Stage is Stage.ActivationCommitted)
|
|
{
|
|
if (!_physics.SetPosition.AcknowledgeProjection(
|
|
progress.Projection))
|
|
{
|
|
// A failed acknowledge is retryable ONLY while nothing has
|
|
// moved authority out from under this exact projection — a
|
|
// genuinely later entity simply sitting ahead of ours in the
|
|
// FIFO. It is NOT automatically retryable: a mid-flight
|
|
// delete (TryAcceptDelete -> CompleteProjectionRetirement ->
|
|
// Physics.SetPosition.Forget -> CancelCoreDeferred) rewrites
|
|
// the SAME pending slot from Place to Discard with a bumped
|
|
// Revision, so the exact struct this class cached in
|
|
// progress.Projection can never match the FIFO head again —
|
|
// without this check, AcknowledgeProjection would fail
|
|
// forever and this progress entry would never converge.
|
|
if (!IsAcknowledgementStillPending(
|
|
record, residenceToken, progress.Projection))
|
|
{
|
|
Discard(key);
|
|
return RuntimeLocalPlayerFirstEntryStatus.RejectedAuthority;
|
|
}
|
|
return RuntimeLocalPlayerFirstEntryStatus
|
|
.AwaitingReceiptAcknowledgement;
|
|
}
|
|
progress.Stage = Stage.Acknowledged;
|
|
}
|
|
|
|
return RunExecute(record, residenceToken, inputs, key, out receipt);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Re-validates authority after a failed acknowledge. C3b review M2:
|
|
/// the mechanism (residence-lease currency + exact-head-token match; a
|
|
/// DIFFERENT entity's head stays retryable) is shared verbatim with the
|
|
/// remote conductor via
|
|
/// <see cref="RuntimeFirstEntryAcknowledgement.IsStillPending"/> — one
|
|
/// body, so the abandonment fix that stops a delete-rewritten Discard
|
|
/// head from producing an infinite AwaitingReceiptAcknowledgement retry
|
|
/// cannot regress independently in either conductor. Full rationale on
|
|
/// the shared helper.
|
|
/// </summary>
|
|
private bool IsAcknowledgementStillPending(
|
|
RuntimeEntityRecord record,
|
|
in RuntimeInitialCreateResidenceToken residenceToken,
|
|
in RuntimePlacementProjectionToken expected) =>
|
|
RuntimeFirstEntryAcknowledgement.IsStillPending(
|
|
_residences,
|
|
_physics.SetPosition,
|
|
record,
|
|
residenceToken,
|
|
expected);
|
|
|
|
private RuntimeLocalPlayerFirstEntryStatus RunExecute(
|
|
RuntimeEntityRecord record,
|
|
in RuntimeInitialCreateResidenceToken residenceToken,
|
|
in RuntimeInitialCreateExecutionInputs inputs,
|
|
RuntimeEntityKey key,
|
|
out RuntimeInitialCreateExecutionReceipt receipt)
|
|
{
|
|
RuntimeInitialCreateExecutionStatus executeStatus = _executor.Execute(
|
|
record, residenceToken, inputs, out receipt);
|
|
switch (executeStatus)
|
|
{
|
|
case RuntimeInitialCreateExecutionStatus.Completed:
|
|
_progress.Remove(key);
|
|
return RuntimeLocalPlayerFirstEntryStatus.Completed;
|
|
case RuntimeInitialCreateExecutionStatus.PendingPlacement:
|
|
return RuntimeLocalPlayerFirstEntryStatus
|
|
.AwaitingReceiptAcknowledgement;
|
|
case RuntimeInitialCreateExecutionStatus.AwaitingContinuationPlacement:
|
|
return RuntimeLocalPlayerFirstEntryStatus
|
|
.AwaitingContinuationPlacement;
|
|
case RuntimeInitialCreateExecutionStatus.RejectedToken:
|
|
_progress.Remove(key);
|
|
return RuntimeLocalPlayerFirstEntryStatus.RejectedToken;
|
|
default:
|
|
_progress.Remove(key);
|
|
return RuntimeLocalPlayerFirstEntryStatus.RejectedAuthority;
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Discards any in-flight publication candidate/activation this class
|
|
/// owns for <paramref name="key"/> and drops its own progress entry.
|
|
/// Safe to call at every stage: <see cref="RuntimeLocalPlayerPhysicsPublicationState.Discard"/>
|
|
/// and <see cref="RuntimeLocalPlayerPhysicsPublicationState.DiscardActivation"/>
|
|
/// are both no-ops against a default/invalid token (an unreached stage's
|
|
/// token field holds exactly that) AND against a token whose activation
|
|
/// has already reached <c>FinalizeActivation</c>'s terminal success path
|
|
/// — <c>FinalizeActivation</c> itself nulls the publication's own
|
|
/// tracked activation the instant the commit succeeds, because the
|
|
/// controller is now genuinely live/published, not a discardable
|
|
/// in-progress candidate. Calling this once <see cref="Stage.ActivationCommitted"/>
|
|
/// (or later) has been reached is therefore correctly a no-op on the
|
|
/// controller/body — an abandoned Place acknowledgement never
|
|
/// retroactively un-publishes an already-live entity; that is ordinary
|
|
/// entity teardown's job (the SAME path any other live entity's delete
|
|
/// already runs), not this class's. The residence and executor own
|
|
/// their own convergence independently (their existing retirement/reset
|
|
/// paths are untouched by this class).
|
|
/// </summary>
|
|
private void Discard(RuntimeEntityKey key)
|
|
{
|
|
if (!_progress.TryGetValue(key, out Progress? progress))
|
|
return;
|
|
_progress.Remove(key);
|
|
// H2: post-H2, a Progress entry can only exist at all if Advance's
|
|
// own upfront check already found Publication bound — this should
|
|
// therefore be structurally unreachable. Guarded anyway
|
|
// (belt-and-suspenders) so a future caller shape can never turn an
|
|
// already-surfaced Advance failure into a SECOND throw from inside
|
|
// an unrelated retirement/session-clear teardown fan-out.
|
|
if (_publication is null)
|
|
return;
|
|
Publication.Discard(progress.PublicationToken);
|
|
Publication.DiscardActivation(progress.ActivationToken);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Cleanup for one key this class is tracking. <see cref="RuntimeEntityObjectLifetime"/>
|
|
/// binds this into <see cref="RuntimeInitialCreateResidenceState"/>'s
|
|
/// multicast retirement notification (alongside the executor's own
|
|
/// <c>DiscardProgress</c>), so any residence retirement path — delete,
|
|
/// reset, generation replacement, a host discovering staleness — reaps
|
|
/// this class's progress automatically. The notification always carries
|
|
/// the exact <see cref="RuntimeEntityKey"/> the residence itself tracked
|
|
/// internally, so it converges correctly even after
|
|
/// <see cref="RuntimeEntityRecord.Key"/> has gone null (e.g. post-delete
|
|
/// teardown released the local id) — unlike re-deriving a key from the
|
|
/// record, which <see cref="Advance"/> cannot do once that happens.
|
|
/// Still exposed directly for a caller that captured a key before a
|
|
/// teardown this class was not notified about (e.g. constructed
|
|
/// standalone in a test without the lifetime's fan-out).
|
|
/// </summary>
|
|
internal void Forget(RuntimeEntityKey key) => Discard(key);
|
|
|
|
/// <summary>
|
|
/// Bulk cleanup mirroring <see cref="RuntimeInitialCreateContinuationExecutor.DiscardAll"/>
|
|
/// — wired into the same session-clear sequence
|
|
/// (<see cref="RuntimeEntityObjectLifetime.BeginSessionClear"/>). Discards
|
|
/// every tracked key's in-flight publication candidate/activation before
|
|
/// dropping the whole progress table, exactly like a per-key
|
|
/// <see cref="Forget"/> for each entry.
|
|
/// </summary>
|
|
internal void DiscardAll()
|
|
{
|
|
// H2: same belt-and-suspenders tolerance as Discard above — a
|
|
// structurally unreachable case post-H2, guarded so bulk session
|
|
// clear can never throw from an unbound Publication either.
|
|
if (_publication is not null)
|
|
{
|
|
foreach (Progress progress in _progress.Values)
|
|
{
|
|
Publication.Discard(progress.PublicationToken);
|
|
Publication.DiscardActivation(progress.ActivationToken);
|
|
}
|
|
}
|
|
_progress.Clear();
|
|
}
|
|
|
|
internal RuntimeLocalPlayerFirstEntryOwnershipSnapshot CaptureOwnership() =>
|
|
new(_progress.Count);
|
|
}
|