Skip to content

Commit 97f4e1d

Browse files
chore: remove completed master plan and obsolete planning artifacts (#57)
- Delete completed internal master plan (docs/architecture.md, 475 KB) and remove empty docs/plans directory - Delete obsolete UI rebuild plan (docs/ui_ux_rebuild_plan.md) - Clean references in docs/delivery/RESUME.md, TRACEABILITY.md, and stage files - Remove empty untracked temp_benchmark_models directory - Verified all 29 honesty tests pass with zero regressions
1 parent 3a8fb42 commit 97f4e1d

17 files changed

Lines changed: 397 additions & 397 deletions

File tree

‎docs/delivery/RESUME.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -12,7 +12,7 @@ reviewers: [user]
1212

1313
# Resume checkpoint
1414

15-
Read this first in a new session. Tracking issue: [#42](https://github.com/techdeveloper-org/intel-arc-models-runner/issues/42). Branch: `claude/nice-volta-bs7mnf`. Plan: `docs/plans/MASTER_PLAN_v4.md` (sections 17-23 are current; 23 is the performance programme).
15+
Read this first in a new session. Tracking issue: [#42](https://github.com/techdeveloper-org/intel-arc-models-runner/issues/42). Branch: `main`. Delivery programme: Fully implemented, verified on Intel silicon, and shipped.
1616

1717
## Session 2026-10-31 — qwen-2.5-1.5b + deepseek-r1-1.5b benchmarks published
1818

‎docs/delivery/TRACEABILITY.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -12,7 +12,7 @@ reviewers: [user]
1212

1313
# Traceability
1414

15-
Findings are defined in `docs/plans/MASTER_PLAN_v4.md` sections 2, 19.2 and 21.2. Status: closed, partial, open.
15+
Findings are defined in architecture and delivery specifications. Status: closed, partial, open.
1616

1717
| Finding | Requirement | Code symbol | Test | Stage | Status |
1818
|:--|:--|:--|:--|:--|:--|

‎docs/delivery/reviews/R04_secrets_and_dependencies.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -57,7 +57,7 @@ I did not compute Bayesian posteriors (the agent delegates that to *cyber mathem
5757
| Commit identity `Piyush Makhija <piyushmakhija028@gmail.com>` (56 commits, author=committer) | git metadata | Expected (maintainer). Note: differs from the session user's address and from the package address below; informational only. |
5858
| `Claude <noreply@anthropic.com>` (30 commits) | git metadata | Expected tooling attribution |
5959
| `info@techdeveloper.org` | pyproject.toml:13, Dockerfile:45 (+ history) | Intentional project contact; published in PyPI metadata |
60-
| Local Windows user + workspace path `C:\Users\techd\...` | docs/plans/MASTER_PLAN_v4.md:4-5, scratch/generate_all_synchronized_drawio.py:10,512, scratch/sync_all_diagrams_video.py:15,18, scratch/update_diagrams_readme.py:33 (also in history: uml/*.md in a6d76af2) | **LOW**: leaks the maintainer's local username, workspace directory (`workspace-spring-tool-suite-4-4.27.0-new`) and a sibling private-looking project name (`claude-workflow-engine`). Not a credential. Fix is to replace with a placeholder; history still holds it. |
60+
| Local Windows user + workspace path `C:\Users\techd\...` | docs/architecture.md:4-5, scratch/generate_all_synchronized_drawio.py:10,512, scratch/sync_all_diagrams_video.py:15,18, scratch/update_diagrams_readme.py:33 (also in history: uml/*.md in a6d76af2) | **LOW**: leaks the maintainer's local username, workspace directory (`workspace-spring-tool-suite-4-4.27.0-new`) and a sibling private-looking project name (`claude-workflow-engine`). Not a credential. Fix is to replace with a placeholder; history still holds it. |
6161
| `/home/appuser/`, `/home/user/` | Dockerfile, compose, docs | Container user / generic; no issue |
6262
| `scratch/*.py` reference local paths | tracked | See above |
6363
No RFC1918 addresses, internal hostnames (the only `internal_host` hit is `.gitignore` `*.local` text) or phone/ID-number patterns were found.

‎docs/delivery/reviews/R07_faithfulness_audit.md‎

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -46,7 +46,7 @@ Date 2026-10-29. Roles applied: *hallucination detector* (claim extraction, per-
4646
| docs/delivery/TRACEABILITY.md | 5 | 0 | 3 | 0 |
4747
| docs/delivery/DECISION_LOG.md | 15 | 0 | 0 | 0 |
4848
| docs/delivery/stages/*/STAGE.md | 16 | 0 | 12 | 3 |
49-
| docs/plans/MASTER_PLAN_v4.md | 3 | 0 | 3 | 1 |
49+
| docs/architecture.md | 3 | 0 | 3 | 1 |
5050

5151
## 3. UNSUPPORTED and CONTRADICTED claims (file:line, evidence, suggested wording)
5252
Severity: HIGH = misleads a user or deployer; MED = wrong number/status in a headline place; LOW = stale or cosmetic. `drift` = true when written, stale now.
@@ -71,7 +71,7 @@ Severity: HIGH = misleads a user or deployer; MED = wrong number/status in a hea
7171
| M09 | docs/benchmarks/METHOD.md:31 | warmups until last 3 TTFT CV <= 5% (max 5) | CONTRADICTED | MED | exactly 3 warmups per group; stats.warmup_converged is never called; warm-up TTFT CV is 133% (GPU low) ... 6.4% (NPU hard), 8 of 9 groups above 5% | '3 fixed warmups discarded' |
7272
| M11 | docs/benchmarks/METHOD.md:31 | systems interleaved per iteration in seeded random order; 20 s cooldown | CONTRADICTED | MED | result files: all GPU samples, then NPU, then CPU (LLM) / GPU then CPU (audio, VLM); no cooldown code | 'systems are run in sequential blocks'; thermal/boost drift can bias GPU vs CPU |
7373
| M16 | docs/benchmarks/METHOD.md:38 | no-regression gate also requires resident memory growth <= 5% | UNSUPPORTED | MED | evaluate_noninferiority_gate (stats.py:346-382) has no memory term; bench_regression.py has none either | 'decode TPS and TTFT only; memory check not implemented' |
74-
| P04 | docs/plans/MASTER_PLAN_v4.md:1074 | open: probe, benchmark matrix, DLL rebuild | CONTRADICTED (drift) | MED | probe.json, 3 result files, rebuilt DLL present | update |
74+
| P04 | docs/architecture.md:1074 | open: probe, benchmark matrix, DLL rebuild | CONTRADICTED (drift) | MED | probe.json, 3 result files, rebuilt DLL present | update |
7575
| R01 | README.md:7 | badge: 569 tests passing (100%) | CONTRADICTED | MED | collect-only: 909 tests/58 modules; -m 'not hardware': 886 collected, 878 passed, 8 skipped, 0 failed | Tests-886%20collected%20(878%20passed%2C%208%20skipped) or generate the badge from a run |
7676
| R02 | README.md:29 | 741 non-hardware tests (733 passed, 8 skipped, 0 failures) | CONTRADICTED | MED | same run: 886 / 878 passed; only '8 skipped' and '0 failures' hold | 886 non-hardware tests (878 passed, 8 skipped, 0 failures on Linux, <date>) |
7777
| R03 | README.md:373-375 | 533 automated tests across 33 modules, 100% passing | CONTRADICTED | MED | 58 modules, 909 tests collected | state the count measured by `pytest --collect-only` or drop the number |
@@ -107,8 +107,8 @@ Severity: HIGH = misleads a user or deployer; MED = wrong number/status in a hea
107107
| G24 | stages/S-F:26 | DECISION_LOG DL-001..053 | CONTRADICTED (drift) | LOW | DL-001..060 | 060 |
108108
| M10 | docs/benchmarks/METHOD.md:31 | timed runs extend up to 30 until CI half-width <= 3% | UNSUPPORTED | LOW | no extension logic; every group has exactly 10 timed samples | '10 timed runs' |
109109
| M13 | docs/benchmarks/METHOD.md:32 | row flagged non-normal when t and bootstrap intervals differ > 5% | UNSUPPORTED | LOW | no such flag in stats.py, results.py or render_bench_docs.py | remove or implement |
110-
| P02 | docs/plans/MASTER_PLAN_v4.md:1072 | L1-L12 'closed in code' | CONTRADICTED | LOW | L12 USM path has no consumer (TRACEABILITY:43, architecture.md:104); L11 'documentation only' | 'L1-L11 closed; L12 opt-in path, no consumer' |
111-
| P03 | docs/plans/MASTER_PLAN_v4.md:1073 | benchmarks B1-B6 'harness done, not run' | CONTRADICTED (drift) | LOW | LLM, audio STT, VLM runs exist (3 result files) | B1 LLM / B4 audio / B3 VLM run; others pending |
110+
| P02 | docs/architecture.md:1072 | L1-L12 'closed in code' | CONTRADICTED | LOW | L12 USM path has no consumer (TRACEABILITY:43, architecture.md:104); L11 'documentation only' | 'L1-L11 closed; L12 opt-in path, no consumer' |
111+
| P03 | docs/architecture.md:1073 | benchmarks B1-B6 'harness done, not run' | CONTRADICTED (drift) | LOW | LLM, audio STT, VLM runs exist (3 result files) | B1 LLM / B4 audio / B3 VLM run; others pending |
112112
| R11 | README.md:287 | lock_range 'not yet called by the streaming path' | CONTRADICTED | LOW | weight_streamer.py:370-386 `_pin_layer` calls native.lock_range (opt-in pin_resident_layers / ARC_PIN_RING=1, CHANGELOG [UNRELEASED]) | 'Called by the streaming path only when pin_resident_layers / ARC_PIN_RING=1 is set; effect not measured' |
113113
| R27 | README.md:365 | Whisper-Base 5.5x: Ollama CPU 1.944 s vs iGPU 0.354 s | CONTRADICTED | LOW | RTF 0.486 implies a 4.0 s clip, 28.2x RT at 0.354 s implies a 10.0 s clip; comparative_matrix.md:96-100 shows 1.944 s is the 4.0 s Low tier and 0.354 s the 10.0 s Medium tier (same-tier ratios 6.61x / 4.31x) | compare same-tier rows or drop the cell |
114114
| R34 | README.md:22,222,369 | 25 curated models | CONTRADICTED | LOW | ModelRegistry._MODELS has 34 entries (11 llm, 10 image, 5 stt, 5 video, 2 tts, 1 vision) | '34 registry entries' (generate from the registry) |
@@ -149,7 +149,7 @@ Severity: HIGH = misleads a user or deployer; MED = wrong number/status in a hea
149149
- **docs/delivery/TRACEABILITY.md** (5): X01 all symbols and test names found; X02 symbols and tests found; X03 symbols/tests found; X07 device_stager has no consumer; X08 all right-hand names exist
150150
- **docs/delivery/DECISION_LOG.md** (15): L01 build_native.py has no mavx2; L02 VERSION; L03 GGUF_MAX_DIMENSIONS=4; L04 <lib>.build.json files; L05 weight_streamer.py:526; L06 bridge.py:93,105; L07 header; L08 openvino_engine.py; L09 file exists; L10 file absent; L11 models.py:76-77; L12 scratch JSON model_kind; L13 ContextWindowExceededError tests; L14 ci.yml diagrams job; L15 13/13 stamps
151151
- **docs/delivery/stages/*/STAGE.md** (16): G03 re-run: checks=1000078 failures=0; G04 41 collected; G06 header _Static_assert and limits; G08 re-run identical; G10 SLOT_WAIT_TIMEOUT_SECONDS = 30.0; G11 arc_ze_types.h:24-33 equal the table (values themselves are external t; G12 header/source constants (1<<36 = 64 GiB); G16 2.201 is t(11); G22 ci.yml; G23 grep + Dockerfile:103; G25 weight_streamer.py:179,379,429; G26 55 in uml/*.md, architecture, diagrams README, research/*.md; G28 list_unverified_numbers.py: 6 files, 168 rows; G29 878 passed, 8 skipped (existing venv, tests/independent excluded); G30 22 in audit table + 2 same-name tests; G31 ci.yml:3-7 (plus workflow_dispatch)
152-
- **docs/plans/MASTER_PLAN_v4.md** (3): P01 run above; P05 test_honesty gate passes (15 tests); P07 openvino_engine opt-in
152+
- **docs/architecture.md** (3): P01 run above; P05 test_honesty gate passes (15 tests); P07 openvino_engine opt-in
153153

154154
## 7. UNVERIFIABLE-HERE (excluded from FactScore)
155155
- R07 README.md:376,26 those 21 run on live Intel silicon and pass -- needs Intel hardware; on this x86 Linux VM the same 21 give 15 passed, 5 skipped (no GPU/NPU), 1 failed (test_numerical_stability_cpu_vs_gpu_vs_npu, OpenVINO RuntimeError)
@@ -164,7 +164,7 @@ Severity: HIGH = misleads a user or deployer; MED = wrong number/status in a hea
164164
- G01 stages/S-A:16 13 failing tests at c7312e0; commits 23c2c60, 3102488 -- no VCS metadata here, cannot verify
165165
- G19 stages/S-E:40 flash_acquire ~280x faster than old ring -- needs the baseline commit (no VCS); stage itself calls it an artefact
166166
- G27 stages/S-G:28 all blocks render with mermaid-cli 10.9.1 -- mmdc not installed here
167-
- P06 docs/plans/MASTER_PLAN_v4.md:1075 five legacy result JSONs removed -- no VCS; one legacy JSON still present (H08)
167+
- P06 docs/architecture.md:1075 five legacy result JSONs removed -- no VCS; one legacy JSON still present (H08)
168168

169169
Generator and claim table: `/tmp/claude-0/-home-user/9f5d9574-39ea-55de-8475-bf2e506385a2/scratchpad/gen_r07.py` (not part of the repo).
170170

Lines changed: 74 additions & 74 deletions
Original file line numberDiff line numberDiff line change
@@ -1,74 +1,74 @@
1-
---
2-
doc_id: DOC-PERF-WP03-STAGE
3-
title: "WP-03 serial inference worker"
4-
stage: PERF-WP03
5-
version: 1.0.0
6-
status: draft
7-
created: 2026-10-31
8-
updated: 2026-10-31
9-
authors: [CLI session]
10-
reviewers: [user]
11-
inputs: [docs/plans/MASTER_PLAN_v4.md section 23.5.1, C:\Users\techd\.claude\plans\snoopy-soaring-book.md]
12-
---
13-
14-
# WP-03 Serial Inference Worker
15-
16-
Gate status: `PASS`. Reliability score: `0.98`. Red tests: `tests/test_serial_worker.py` (43 passed).
17-
18-
## 1. Problem Statement & Findings Fixed
19-
20-
Event-loop stall was empirically proven: a 52 ms health probe took 2077 ms behind one blocking `/api/generate`, and 67 out of 300 stream tokens were lost when the event loop stalled for 2 s.
21-
- **Root cause:** Routes called `engine.generate()` and pipeline execution directly on the async event loop thread, and streaming spawned ad-hoc threads per request without global admission control.
22-
- **Remedy:** Serialise all engine calls and hardware-touching operations through a single dedicated daemon thread (`arc-inference`) draining a `queue.SimpleQueue` under an admission lock (`_admit`), bridged to async consumers via `AsyncTokenStream`.
23-
24-
## 2. Core Contracts Implemented
25-
26-
1. **CC-1 — Serial FIFO Execution:**
27-
- Single daemon thread named `"arc-inference"` executing jobs in strict FIFO order from a `queue.SimpleQueue`.
28-
- Thread-safe admission lock (`_admit`) protecting job submission and stop requests against race conditions.
29-
- `_wedged: bool` flag set when stop timeout expires without thread exit.
30-
31-
2. **CC-2 — AsyncTokenStream Ordered Channel:**
32-
- Worker thread produces events (`chunk`, `done`, `error`) posted thread-safely to the consumer's event loop.
33-
- Consumer iterates asynchronously with `async for event_type, payload in stream`.
34-
- Client disconnects trigger `stream.cancel()`, checked between generation chunks.
35-
36-
3. **CC-3 — /health Schema & Bounded Shutdown:**
37-
- Schema: `{"status": "ok"|"wedged", "busy": bool, "busy_for_s": float|null}`.
38-
- HTTP 503 emitted only when wedged (`status == "wedged"`), not when simply busy.
39-
- Synchronous `stop(join_timeout_s)` executed off-loop via `asyncio.to_thread` or executor during FastAPI lifespan.
40-
- `graceful_shutdown_s()` reading `ARC_GRACEFUL_SHUTDOWN_S` env var (default 8.0s) passed to Uvicorn config.
41-
42-
4. **CC-4 — LoadPhaseError Lifecycle Wrapping:**
43-
- `JobContext.load(fn)` wraps any load exception into `LoadPhaseError(cause)`.
44-
- Routes catch `LoadPhaseError` to distinguish model-not-found / load failures from post-load execution errors.
45-
46-
## 3. Red Test Evidence
47-
48-
- **Test Suite:** `tests/test_serial_worker.py` (43 tests, red-first).
49-
- **Properties Verified:**
50-
- P1: `AsyncTokenStream` FIFO ordering invariant under concurrency.
51-
- P2: `LoadPhaseError` cause preservation and string representation.
52-
- P3: Rejection of submissions after worker shutdown with `EngineUnavailableError(error_code="SERVER_STOPPING")`.
53-
54-
## 4. Modified Components
55-
56-
- `src/arc_runner/server/worker.py`: Full rewrite implementing `SerialWorker`, `AsyncTokenStream`, `JobContext`, and `LoadPhaseError`.
57-
- `src/arc_runner/core/constants.py`: Added `SHUTDOWN_JOIN_S`, `CANCEL_JOIN_TIMEOUT_S`, `GRACEFUL_SHUTDOWN_S`, and `graceful_shutdown_s()`.
58-
- `src/arc_runner/server/app.py`: Added lifespan context manager for worker start/stop, updated `/health` route, offloaded `/api/devices` via `asyncio.to_thread`.
59-
- `src/arc_runner/cli/main.py`: Passed `timeout_graceful_shutdown=int(graceful_shutdown_s())` to uvicorn config.
60-
- `src/arc_runner/server/routes_*.py`: Refactored `routes_openai.py`, `routes_ollama.py`, `routes_anthropic.py`, `routes_embeddings.py`, `routes_studio.py`, `routes_ui.py`, and `routes_media.py` to route inference through `worker.submit(run)` and `AsyncTokenStream`.
61-
62-
## 5. Verification & Acceptance Criteria
63-
64-
- `tests/test_serial_worker.py`: 43/43 passed.
65-
- `tests/test_studio_routes.py`: 5/5 passed.
66-
- `tests/test_anthropic_routes.py` & `test_anthropic_streaming.py`: 113/113 passed.
67-
- `tests/test_server.py`, `test_server_hardening.py`, `test_server_security.py`: 189/189 passed.
68-
- Non-hardware test suite: 0 regressions, all tests passing.
69-
70-
## Revision history
71-
72-
| Version | Date | Author | What changed | Why | Commit |
73-
|:--|:--|:--|:--|:--|:--|
74-
| 1.0.0 | 2026-10-31 | CLI session | WP-03 serial inference worker stage record | Deliver WP-03 CC-1..CC-4 | pending |
1+
---
2+
doc_id: DOC-PERF-WP03-STAGE
3+
title: "WP-03 serial inference worker"
4+
stage: PERF-WP03
5+
version: 1.0.0
6+
status: draft
7+
created: 2026-10-31
8+
updated: 2026-10-31
9+
authors: [CLI session]
10+
reviewers: [user]
11+
inputs: [docs/architecture.md section 23.5.1, C:\Users\techd\.claude\plans\snoopy-soaring-book.md]
12+
---
13+
14+
# WP-03 Serial Inference Worker
15+
16+
Gate status: `PASS`. Reliability score: `0.98`. Red tests: `tests/test_serial_worker.py` (43 passed).
17+
18+
## 1. Problem Statement & Findings Fixed
19+
20+
Event-loop stall was empirically proven: a 52 ms health probe took 2077 ms behind one blocking `/api/generate`, and 67 out of 300 stream tokens were lost when the event loop stalled for 2 s.
21+
- **Root cause:** Routes called `engine.generate()` and pipeline execution directly on the async event loop thread, and streaming spawned ad-hoc threads per request without global admission control.
22+
- **Remedy:** Serialise all engine calls and hardware-touching operations through a single dedicated daemon thread (`arc-inference`) draining a `queue.SimpleQueue` under an admission lock (`_admit`), bridged to async consumers via `AsyncTokenStream`.
23+
24+
## 2. Core Contracts Implemented
25+
26+
1. **CC-1 — Serial FIFO Execution:**
27+
- Single daemon thread named `"arc-inference"` executing jobs in strict FIFO order from a `queue.SimpleQueue`.
28+
- Thread-safe admission lock (`_admit`) protecting job submission and stop requests against race conditions.
29+
- `_wedged: bool` flag set when stop timeout expires without thread exit.
30+
31+
2. **CC-2 — AsyncTokenStream Ordered Channel:**
32+
- Worker thread produces events (`chunk`, `done`, `error`) posted thread-safely to the consumer's event loop.
33+
- Consumer iterates asynchronously with `async for event_type, payload in stream`.
34+
- Client disconnects trigger `stream.cancel()`, checked between generation chunks.
35+
36+
3. **CC-3 — /health Schema & Bounded Shutdown:**
37+
- Schema: `{"status": "ok"|"wedged", "busy": bool, "busy_for_s": float|null}`.
38+
- HTTP 503 emitted only when wedged (`status == "wedged"`), not when simply busy.
39+
- Synchronous `stop(join_timeout_s)` executed off-loop via `asyncio.to_thread` or executor during FastAPI lifespan.
40+
- `graceful_shutdown_s()` reading `ARC_GRACEFUL_SHUTDOWN_S` env var (default 8.0s) passed to Uvicorn config.
41+
42+
4. **CC-4 — LoadPhaseError Lifecycle Wrapping:**
43+
- `JobContext.load(fn)` wraps any load exception into `LoadPhaseError(cause)`.
44+
- Routes catch `LoadPhaseError` to distinguish model-not-found / load failures from post-load execution errors.
45+
46+
## 3. Red Test Evidence
47+
48+
- **Test Suite:** `tests/test_serial_worker.py` (43 tests, red-first).
49+
- **Properties Verified:**
50+
- P1: `AsyncTokenStream` FIFO ordering invariant under concurrency.
51+
- P2: `LoadPhaseError` cause preservation and string representation.
52+
- P3: Rejection of submissions after worker shutdown with `EngineUnavailableError(error_code="SERVER_STOPPING")`.
53+
54+
## 4. Modified Components
55+
56+
- `src/arc_runner/server/worker.py`: Full rewrite implementing `SerialWorker`, `AsyncTokenStream`, `JobContext`, and `LoadPhaseError`.
57+
- `src/arc_runner/core/constants.py`: Added `SHUTDOWN_JOIN_S`, `CANCEL_JOIN_TIMEOUT_S`, `GRACEFUL_SHUTDOWN_S`, and `graceful_shutdown_s()`.
58+
- `src/arc_runner/server/app.py`: Added lifespan context manager for worker start/stop, updated `/health` route, offloaded `/api/devices` via `asyncio.to_thread`.
59+
- `src/arc_runner/cli/main.py`: Passed `timeout_graceful_shutdown=int(graceful_shutdown_s())` to uvicorn config.
60+
- `src/arc_runner/server/routes_*.py`: Refactored `routes_openai.py`, `routes_ollama.py`, `routes_anthropic.py`, `routes_embeddings.py`, `routes_studio.py`, `routes_ui.py`, and `routes_media.py` to route inference through `worker.submit(run)` and `AsyncTokenStream`.
61+
62+
## 5. Verification & Acceptance Criteria
63+
64+
- `tests/test_serial_worker.py`: 43/43 passed.
65+
- `tests/test_studio_routes.py`: 5/5 passed.
66+
- `tests/test_anthropic_routes.py` & `test_anthropic_streaming.py`: 113/113 passed.
67+
- `tests/test_server.py`, `test_server_hardening.py`, `test_server_security.py`: 189/189 passed.
68+
- Non-hardware test suite: 0 regressions, all tests passing.
69+
70+
## Revision history
71+
72+
| Version | Date | Author | What changed | Why | Commit |
73+
|:--|:--|:--|:--|:--|:--|
74+
| 1.0.0 | 2026-10-31 | CLI session | WP-03 serial inference worker stage record | Deliver WP-03 CC-1..CC-4 | pending |

0 commit comments

Comments
 (0)