-
-
Notifications
You must be signed in to change notification settings - Fork 1
Testing Requirements
Recorder tests with a fake stateful device verify per-tick chip-state capture and that Snapshot deep-copies (a later capture does not corrupt an earlier snapshot). VIC/SID/CIA/PLA CaptureState/DecodeState round-trip exercised end to end via GetChipStateAtTick.
Acceptance Criteria:
- OnInstructionCompleted captures the device state into the tick
- Snapshot deep-copies chip state across ring rotation
Unit tests covering ExecutedCycles increments per executed (non-stolen) cycle, per-CPU rate = executed/elapsed/target, reset zeroes counters, and the status surface lists host and peripheral CPU rates distinctly.
Each upgrade slice S1-S8 exits on: dotnet restore + Release build with TreatWarningsAsErrors + slice-focused test filter + the full filtered TestHarness suite at 100 percent green (0 failed, 21 skipped baseline, >= 2594 passed; filter Category!=Determinism&Category!=AiReview&Category!=ParityPending&Category!=ParityLegacy). S3 adds an interactive Avalonia smoke; S6 proves nuget.org-only restore for aiUnit/RemoteControl; S7 proves PackNuget locally; S8 proves the source generator loads under the raised SDK floor (Roslyn 5.6).
Acceptance Criteria:
- Every upgrade slice exits with the full filtered suite at 0 failed and stable skips, plus its slice-specific proof. (evidence: Local gates S1-S8 all 0 failed / 21 skipped; CI runs 1086, 1087, 1088, 1089, 1091, 1092, 1093, 1094, 1095 all succeeded.)
Focused integration tests shall prove D64 directory and file operations complete over the IEC bus and produce bus activity telemetry.
Acceptance Criteria:
- Tests prove mounted D64 directory and file operations execute through IEC bus line activity.
- Tests prove the same D64 operations complete with correct directory and file data over the IEC path.
- Focused IEC/D64 tests run with zero failed and zero skipped tests.
After SetMotor(true) and 300,000 Tick() calls, MotorRotationCycles>0. Before 300,000, stays at 0. Motor off/on resets ramp. ReadSector(18,0) returns BAM bytes 0x12/0x01. 5 pass.
Drive DOS ROM setting VIA2 PB3 sets the drive LedOn; per-drive status DTO reflects it; the card LED view-model binds it.
Attach creates+clocks+bus-connects the drive instance; detach unregisters from clock and disconnects from bus.
VM: TrueDrive toggles per drive (default off). Runtime: with TrueDrive on, the session uses the coordinator true-drive 1541 path (LOAD works); with it off, the simulated drive is used and existing tests stay green.
After setting IecBus.Atn=false and calling Tick() up to 985 times, Clock==false and Data==false. After ATN release and Tick() up to 985 times, both return true. 5 pass.
Canonical captured sequences (LOAD command frame, directory read) decode to the expected IEC event list.
With the live electrical model + true-drive, normal C64 idle CIA2 PA reads $47 and C64GS reads $07; native CIA2-IO lockstep stays green.
Tests that a drive attached mid-run answers on the bus, a detached drive removes its pulls and recomputes line states, and a renumbered drive answers its new device number.
LOAD"*",8,1 and LOAD"$",8 complete over IEC against a real D64 on a single-system C64.
Scope view-model binds trace + decoded events; host trace-delta contract returns samples since cycle N; step/rewind move the cursor.
Snapshot reports idle all-high/no-talkers; single puller -> line low + talker; multi-puller wired-AND lists all; release restores high; snapshot never mutates bus. DONE (5/5 green).
Edge capture order + cycle stamps; step-boundary marks; ring bound; rewind re-derivation equals original trace.
FrameSequenceCapture unique-dedup, drop-on-mismatch, and host frames=unique routing.
WAV header/data correctness, after-stop ignore, concurrent-stop lock safety; CaptureAudioTap clamp/forward.
ffprobe-verified video+audio streams; Start failure cleanup; BackgroundByteWriter ordering/fault.
NativeResidueDiagTests: probe 01 diffs a fresh machine's 200-cycle boot state before/after a demo .vsf resume on a scratch machine; probe 02 replicates the post-activity-reset divergence (boot + full PAL frame + reset + per-cycle walk 140-260 sampling DD00/DD01, CIA1+CIA2, VIC beam, CPU). Guards the PLAN-NATIVERESIDUE-001 mechanism fix: vice_machine_create_model re-baselines Drive{8..11}TrueEmulation to the VICE default 1 via resources_set_int so a has_tde=0 DRIVE8 snapshot module cannot disable True Drive Emulation process-wide.
Acceptance Criteria:
- Fresh-machine boot state is bit-identical before and after a .vsf snapshot resume in the same process. (evidence: TEST-NATIVE-RESIDUE-01 green post-fix; poison combo 36 passed / 0 failed / 1 skipped in one process.)
- Post-activity-reset per-cycle trace (DD00/CIA/VIC/CPU, cycles 140-260) is bit-identical before and after a .vsf resume. (evidence: TEST-NATIVE-RESIDUE-02 green post-fix; full suite one process 0 failed / 2594 passed.)
A machine created after a scratch machine attaches/detaches a disk (or resumes a has_tde=0 .vsf) must present drive-8 attach_clk/detach_clk/attach_detach_clk == 0. Guards drive-snapshot.c uninitialized-stack + shim create-time re-baseline. Code: NativeResidueDiagTests.FreshMachine_DriveAttachDetachClocks_AreZero_AfterResidue (label TEST-NATIVE-RESIDUE-03).
A machine created after a scratch machine resumes a true-drive .vsf and runs activity must present drive-8 cycle_accum == 0. Guards drivecpu_reset_clk cycle_accum omission. Code: NativeResidueDiagTests.FreshMachine_DriveCycleAccum_IsZero_AfterSnapshotResumeActivity (label TEST-NATIVE-RESIDUE-04).
PacingStrategySelectionTests (factory + live SetStrategy), SettingsServiceHostTests (UpdateSettings applies PacingStrategy to the pump and round-trips it), and AttachPanelViewModelTests (pacing change flags pending, not restart).
Acceptance Criteria:
- SetStrategy on a not-started pump switches GateName immediately; unknown defaults to Semaphore
- UpdateSettings with a VICE pacing strategy sets the pump gate to VICE and round-trips vice
The benchmark harness builds a real-ROM C64 PAL machine through ArchitectureBuilder, measures IMachine.RunFrame after warmup over the required 600-frame window, reports median and p95 frame time, and proves the measured hot path allocates zero bytes on the current thread.
Acceptance Criteria:
- Benchmark/probe builds Commodore 64 PAL through ArchitectureBuilder with the real ROM provider. (evidence: BenchmarksSmokeTests.C64PalRunFrameBenchmark_UsesRealC64Pal passed)
- RunFramePerfProbe 60/600 reports median <= 18 ms, p95 <= 22 ms, and 0 allocated bytes. (evidence: median=1.575ms; p95=2.753ms; allocated=0 bytes)
- Focused BasicBus/C64MemoryMap/VideoRenderer/VideoSurface/SID and Lockstep/Checkpoint gates pass with 0 failed and 0 skipped tests. (evidence: focused=182 passed; lockstep/checkpoint=333 passed)
- BenchmarkDotNet C64PalRunFrameBenchmark completes and reports no managed allocation. (evidence: BenchmarkDotNet mean=2.262ms median=2.255ms; Allocated reported none)
Focused unit, smoke, and benchmark-probe coverage shall verify the Pub/Sub event bus contract, including typed/raw/packed delivery, unsubscription, deterministic delivery order, collision isolation, route growth, message pool exhaustion/reset, frame arena payloads, zero publish allocation, and release probe performance.
Acceptance Criteria:
- Focused Pub/Sub test suite and benchmark smoke tests pass. (evidence: dotnet test tests/ViceSharp.TestHarness/ViceSharp.TestHarness.csproj --no-restore --filter FullyQualifiedName
LockFreePubSubTests|FullyQualifiedNameBenchmarksSmokeTests.PubSub => Passed: 22, Failed: 0) - Release quick probe proves the Pub/Sub performance targets without managed allocation. (evidence: dotnet run -c Release --project tests/ViceSharp.Benchmarks -- --pubsub-probe 1000000 => publish-one=43.78ns publish-three=58.14ns publish-packed=35.62ns pool-rent-return=16.80ns arena-alloc=3.20ns allocated=0 bytes)
- The full solution builds after the Pub/Sub implementation. (evidence: dotnet build ViceSharp.slnx --no-restore => Build succeeded)
Tests prove the RemoteControl integration is off by default (no host started when VICESHARP_REMOTECONTROL_ENABLE is unset) and fails closed when enabled without a token; and that, when enabled with a token, the configured IRemoteControlRootProvider returns the live MainWindow. App-launch gate: connect the RemoteControl client tool and confirm the visual tree is readable.
RewindCycle/RewindFrame restore exact prior state; step-then-rewind round-trips; ring eviction bound honored.
Focused SID tests verify signed waveform centering and PCM equivalency; runtime capture verifies Pieces of Light at PAL pacing before and after Space exits the first segment.
SidClockRateTests drives the SID via the SystemClock and asserts OSC3 = 0x10 after 8192 master cycles at voice freq 0x8000 (was 0x01). SidAudioPumpTests recalibrated to ClockDivisor 1 (PalTicksPerFrame 19656) with sample counts unchanged.
Acceptance Criteria:
- SidPhase_AdvancesAtPhi2Rate_WhenClockedBySystemClock passes (OSC3 == 0x10)
- SidAudioPumpTests sample-count assertions still pass after recalibration
AttachPanelViewModelTests.CollapseExpander_DockAndGlyph_TrackAnchorSide asserts the expander dock side and chevron glyph follow the panel anchor and raise PropertyChanged.
Acceptance Criteria:
- Anchored Left yields CollapseExpanderDock Right and glyph the left chevron
- Anchored Right yields CollapseExpanderDock Left and glyph the right chevron, with PropertyChanged raised for both
Restore then run N cycles equals continuous run N (full MachineState + memory equality) at multiple cut points, for C64 and C64+1541.
ViceGateSoundRegulatorTests (13) cover EvaluateSound outcomes and boundary, gate regulator selection, back-pressure blocking advance, warp precedence, and the SID IsAudioTimingSource/QueuedSampleCount wiring.
Acceptance Criteria:
- Back-pressure when queue at/over high-water advances zero cycles; below advances a chunk
- Warp selects Warp regulator even when audio buffer is full
Tests that the drive CPU advances on its own clock (not host lockstep), IEC bus edges are observed by other systems before a read, per-system pacing sustains clock under load, and the true-drive LOAD parity test still passes.
With tape inserted, MotorEnabled=true, PlayPressed=true: TryReadNextPulse returns false during first 12,808 Tick() calls; returns true after 32,000 Tick() calls. Motor off/on resets ramp. 4 pass.
SenseLine==false when PlayPressed or RecordPressed, true otherwise. TryWritePulse returns true only when MotorEnabled && RecordPressed, incrementing RecordedPulseCount. 7 pass.
TickHistoryRecorderTests (ring order/capacity, write bundling, reconstruction), TickHistoryCaptureTests (bus event + pump capture + GetTickHistory/ReadMemoryAtTick), TickHistoryViewModelTests (refresh newest-first, paused inspect, close).
Acceptance Criteria:
- Reconstruct at the newest tick yields current memory; at an earlier tick yields the pre-write byte
- InspectAsync opens the debug screen with a memory dump only when paused
Verify the emulation pump does not capture when recording is disabled (default) and that GetTickHistory arms recording.
Acceptance Criteria:
- Pump_DoesNotCaptureHistory_WhenRecordingDisabled - default off yields empty history after pumping.
- GetTickHistory_ArmsHistoryRecording - calling the RPC sets HistoryRecordingEnabled true.
Focused ViewModel and protocol tests shall prove IEC activity appears in both peripherals and status surfaces from host telemetry.
Acceptance Criteria:
- ViewModel tests prove peripherals panel drive entries show IEC active and idle states from host telemetry.
- ViewModel tests prove status bar IEC activity uses the same source while preserving status fields.
- Focused UI tests run with zero failed and zero skipped tests.
VM: toggling DockSide flips Left<->Right and IsPaneOpen opens/closes; default state correct. App-launch: flyout opens/closes, docks both sides via the single icon button, no Left/Right buttons.
VM: menu commands invoke the existing actions/host services (attach/eject, reset, snapshot, warp, true-drive toggle, swap joysticks). App-launch: menu structure matches the plan and commands work.
VM: AttachSlotViewModel exposes status/RO/activity/TrueDrive/SupportsTrueDrive and Attach/Eject route to host. App-launch: Drive8/Drive9/Tape/Cartridge all render via the one PeripheralCardView with correct per-slot controls.
VM: settings VM exposes machine/video/audio/input/limiter/resource selections + apply. App-launch: SettingsView renders + applies settings.
Managed PAL frame periodic raster/cycle-counter (2 Facts), native screen-RAM roundtrip and one-frame DMA read-only (2 ViceFacts), sprite-3 DMA window for 5 models (5 ViceTheory). All 9 pass.
After writing DEN=1/YSCROLL=0/1 and advancing to specific rasterLine/rasterX, CurrentRowCounter and IsGraphicsIdle match VICE viciisc/vicii-cycle.c:541-563 expectations. All 11 pass.
SnapshotResumeSpikeTests.ExternalX64ScVsf_FullyResumes_CpuMatchesSnapshot loads the supplied ready-c64sc-truedrive.vsf through the shim and asserts rc==0 and that the resumed MAINCPU registers equal the snapshot's MAINCPU module. Evidence: PASS (rc=0, snapshot_last_error=0, A=C2 X=02 Y=00 SP=FA PC=5769 both sides); ShimRoundTrip + SID parity 8/8; X64ScVariantLockstep 306 pass / 0 fail / 1 skip.
Acceptance Criteria:
- dotnet test ...TestHarness --filter ExternalX64ScVsf_FullyResumes => passed; rc=0, err=0, MAINCPU regs match.
Generated from MCP requirements wiki export.
- Home
- Getting Started
- Architecture
- Requirements
- Reference