v0.2.9-rc5 — P0~P2 CRUD repair & E2E validation

v0.2.9-rc5

Highlights

• P0 .stignore 目录排除修复：skills/、tools/、ignored/ 等带斜杠规则现在正确排除目录
• P1 重命名检测 + 本地复制优化：scanner 检测重命名，puller 优先本地复制而非重新下载
• P2 FileInfo 字段兼容：modified_by、blocks_hash、no_permissions 完整映射
• BEP 桥接修复：LocalIndexUpdated 事件正确转发为 IndexUpdate 到所有已连接设备
• E2E CRUD 测试：新增 5 个端到端测试（create / modify / delete / rename / stignore），全部通过

Assets

Platform	Binary	Size
Windows x64	syncthing-v0.2.9-rc5-x86_64-pc-windows-msvc.exe	~13 MB
Linux x64	syncthing-v0.2.9-rc5-x86_64-unknown-linux-gnu	~14 MB

Verification

cargo test --workspace  # 319 passed, 0 failed
cargo clippy --workspace --all-targets  # 0 warnings

Full report: docs/reports/CRUD_REPAIR_E2E_2026-05-22.md

syncthing-rust v0.2.8 — Scanner 元数据排除 + CI/文档清理

syncthing-rust v0.2.8

🔧 Fixes

• Scanner 默认排除 syncthing 元数据 (D-1): 自动排除 .stfolder、.stversions、.stignore、config.json、cert.pem、key.pem、db、logs 及 .syncthing.tmp 后缀，防止同步目录与配置目录重合时的递归灾难。
• 修复 pre-existing clippy error (clippy::only_used_in_recursion).
• 修复 6 个 pre-existing rustdoc warnings.
• 修复 CI workflow 中重复的 file-size job.
• cargo fmt --all 清理 4 个 pre-existing 格式问题文件.

✅ Verification

• CI 全部 11 个 job 通过（含 Windows Clippy/Test/Release）
• 本地 cargo test --workspace: 309 passed, 0 failed

📦 Assets

Platform	Binary
Windows x64	syncthing-x86_64-windows.exe
Linux x64	syncthing-x86_64-linux

Release main

Auto-built from main

v0.2.7 — Windows Block Transfer Fix + Config UX

Bug Fixes: Bug-1 connected_devices empty, Bug-2 double-dot tmp, Bug-3 Scanner paths. New: init wizard, URL addresses, REST hot-add, config validation, single-instance lock. CI: release/doc/e2e checks. Verified: 41 tests pass, WSL2-Windows e2e sync OK.

v0.2.6 — Runtime Safety Hardening + Cross-Version Interop

Releases

v0.2.6 — Runtime Safety Hardening (2026-05-14)

🔒 Security & Stability Hotfix

This is a pure-fix release with zero new features. It addresses the root causes identified in post-incident review INC-20260514-001 (config hot-reload watcher entered a tight loop under notify event storms, producing 21 GB of logs in 19 hours).

Six hardening patches applied (H-1 through H-6):

Patch	File(s)	What changed
H-1	daemon_runner.rs, config.rs	Config hot-reload now has a 500 ms debounce (resetting timer) and mtime comparison. Hot-reload logs downgraded from info! → debug!.
H-2	main.rs	Daemon log rotation switched from daily → hourly / 100 MB size cap, preventing single-day log explosion.
H-3	syncthing-net, syncthing-sync, daemon_runner	All 7 unbounded_channel() sites replaced with bounded channel(1024). Senders use try_send; full queues drop and emit warn!.
H-4	relay_listener, tcp_transport, bep_adapter	Cleaned up dropped-receiver patterns that leaked memory.
H-5	events.rs, block_cache, derp/server, registry	All panic! / unreachable!() on external input converted to error! + Err(...).
H-6	daemon_runner, discovery_tasks, relay_listener, events	Pure interval.tick() loops now have select! { _ = shutdown.changed() => break } for graceful termination.

🔬 Cross-Version Interoperability Verified

Rust v0.2.6 ↔ Go Syncthing v2.1.0 fully verified on local loopback (2026-05-14):

• TLS 1.3 handshake (TLS_AES_128_GCM_SHA256) ✅
• BEP Hello exchange ✅
• ClusterConfig exchange ✅
• Index broadcast / receive ✅
• Block Request / Response ✅
• File materialization on receiver ✅

Automation: scripts/cross_version_test.sh (281 lines, Linux/Windows dual-platform).

📊 Quality Metrics

cargo test --workspace        → 296 passed, 0 failed, 1 ignored (e2e_sync)
cargo clippy --workspace      → 0 warnings (-D warnings)
cargo machete                 → 0 unused deps (4 removed in this cycle)
criterion benches             → 4 skeletons (device_id, scanner, hash_parallel, encode_decode)

📦 Binaries

Platform	Target	File	Size
Windows	x86_64-pc-windows-msvc	syncthing-v0.2.6-x86_64-pc-windows-msvc.exe	~12 MB
Linux	x86_64-unknown-linux-gnu	syncthing-v0.2.6-x86_64-unknown-linux-gnu	~11 MB
Stress test	x86_64-pc-windows-msvc	stress_test-v0.2.6-x86_64-pc-windows-msvc.exe	~6.6 MB

Build from source (any platform with Rust 1.85+):

cargo build --release -p syncthing

⚠️ Known Limitations

• §1 ClusterConfig race: First handshake may stall ~10 s; auto-reconnect succeeds on second cycle. Mitigated by 60 s session health check. e2e_sync integration test marked #[ignore] under parallel load.
• 72 h endurance: Scripts ready (scripts/72h_*.sh), Linux deployment pending.
• Tailscale field validation: Core discovery integration complete; real-world CGNAT/MTU testing deferred.

🔗 References

• Full incident write-up: docs/plans/NEXT_STEPS_2026-05-14.md §0
• Known issues registry: docs/KNOWN_ISSUES.md
• Tuning plan: docs/plans/TUNING_PLAN_2026-05-11.md
• Cross-version test script: scripts/cross_version_test.sh

v0.2.5 — End-to-End sync chain verified

syncthing-rust v0.2.5

Released: 2026-05-13
Codename: "End-to-End"
Theme: 核心 sync 链路打通 + 工程债务清理

---

🎯 Headline

v0.2.4 把连接层修稳了（9h+ 压测、0 死锁、0 panic）；
v0.2.5 把"实际把文件传过去"补完了。

e2e_sync 测试在 12.02s 内完成两节点单文件端到端同步：

TLS → Hello → ClusterConfig → Index → Block → 文件落地

此前一直 #[ignore] 的诊断测试现在是回归套件一员。

---

🐛 关键修复 — T2.6 (P0)

Runtime-added folders never synchronized

症状：通过 REST API POST /rest/config/folders、TUI、或测试 harness 在
运行时添加的文件夹永不同步——必须重启进程才能生效。

根因：SyncManager::add_folder 创建 FolderModel 但未 spawn
scan/pull/watcher 三个 tokio 任务循环。start_folder_loops() 仅在
SyncManager::start() 时遍历当时已存在的文件夹。运行时新增的 folder
对应的 pull_notify.notify_one() 通知到没有 awaiter 的 Notify，静默丢失。

修复：1 行 — 在 add_folder 末尾无条件调用幂等的 start_folder_internal。
该函数早就有 folder_tasks.contains_key() 检查，调用安全。

验证：

• cargo test --test e2e_sync → 12.02s passed（此前 #[ignore]）
• 295 unit tests 全过
• 0 clippy warnings

详细 RCA 见 docs/KNOWN_ISSUES.md §2。

---

🧹 重构 — T2.3 (RFC-001)

service/mod.rs 由 695 行拆分为 60 行 + 4 个职责子模块：

service/
├── mod.rs              (60)  struct + FolderTaskHandles + mod 声明
├── lifecycle.rs        (197) new/with_*/start/stop/run + 内部 helper
├── sync_manager.rs     (225) impl SyncManager for SyncService
├── network_bridge.rs   (119) BEP 网络层回调
└── sync_model.rs       (164) impl SyncModel for SyncService (FFI 边界)

纯位置移动，0 API 变化、0 运行时语义变化。

---

🛠 工具链 — T2.2

• stress_test 日志滚动：tracing_appender::rolling 每日切片，保留 7 天。
  防止多日长跑日志膨胀到 GB 级。新增 --log-dir <PATH> CLI 选项。
• CSV ISO 8601 时间戳：替换坏掉的 fmt_system_time（产出
  "20585T05:07:55Z" 这种串），现在 CSV 报告完全可被机器解析。

---

📚 新增文档

文档	用途
docs/KNOWN_ISSUES.md	已知缺陷登记 + 修复追溯
docs/reports/STRESS_TEST_REPORT_2026-05-13.md	9h11m 压测分析（暴露 T2.5 缺口）
docs/operations/TAILSCALE_GUIDE.md	与 Tailscale 协同部署的 3 条路径
docs/operations/PROXY_GUIDE.md	SOCKS5/HTTP_PROXY 代理使用
docs/drafts/RFC-001-service-split.md	T2.3 拆分架构决策

---

🗂 累计工程清理（Phase A，T1.1–T1.6）

6/6 文件拆分完成。所有原本 > 550 行的核心文件现在都 ≤ 500：

原文件	拆分后	Commit
dialer.rs 621	mod 452 + tests 168	90e6d3b
block_cache.rs 556	mod 322 + tests 234	5fd42b0
types/mod.rs 639	mod 477 + folder.rs 176	f21c8fe
daemon_runner.rs 596	mod 476 + 2 helpers 176	0140851
traits.rs 574	mod 463 + transport.rs 127	66589c2
service/mod.rs 695	mod 60 + 4 children 705	T2.3 (this)

manual_let_else clippy nursery lint 全工程 10 处修复（921cf0f）。

---

⚠️ 已知限制（未变）

• §1 ClusterConfig 首次握手 10s 超时：首轮连接约 12s（自动重连第二轮兜底）。
• 72h 压测：Windows 桌面 sleep 杀进程，需 Linux 平台重跑。
• Go Syncthing 跨版本互通：仅 2026-04-11 手测一次，无自动化。

---

📦 二进制下载

平台	体积	文件	SHA-256
Windows x86_64	12 MB	syncthing-v0.2.5-x86_64-windows.exe	06b4451c...e9be07
Linux x86_64	11 MB	syncthing-v0.2.5-x86_64-linux	76fcde6c...3ecb2

完整 SHA-256（参见随附 SHA256SUMS）：

06b4451cfa747c43de3265cb20888797c5e147fcd1eaf437f7c91acbbbe9be07  syncthing-v0.2.5-x86_64-windows.exe
76fcde6cc958fc301b82e445960b7e8e7a3f0801d56348b06cff47d6fad3ecb2  syncthing-v0.2.5-x86_64-linux

二进制为 release 优化，无运行时依赖（OpenSSL、libc 由 rustls + 静态链接处理）。

首次运行：

# Windows
.\syncthing-v0.2.5-x86_64-windows.exe tui    # 交互式 TUI
.\syncthing-v0.2.5-x86_64-windows.exe run    # headless

# Linux
chmod +x syncthing-v0.2.5-x86_64-linux
./syncthing-v0.2.5-x86_64-linux tui

默认端口：BEP 22001，REST API 8385。

---

🔗 链接

• Source: https://github.com/juice094/syncthing-rust
• Changelog: CHANGELOG.md
• Architecture: docs/design/ARCHITECTURE_DECISIONS.md
• Known issues: docs/KNOWN_ISSUES.md

---

🙏 致谢

T2.6 根因诊断由 BEP bridge harness (T2.5) 触发——没有这个诊断测试，
add_folder spawn 缺失在生产中会一直作为"REST API 加文件夹后必须重启进程"的
偏门怪事存在。

v0.2.4 — Critical Deadlock Fix + Performance Baseline

v0.2.4 — Critical Deadlock Fix + Performance Baseline

13 commits since v0.2.3, headlined by a critical concurrency bug fix that previously made stress tests freeze at ~3 minutes.

🐛 Critical Fix: BEP Race Resolution Deadlock (T-F1)

ConnectionManager::register_connection was holding a DashMap write guard (RefMut) across .await on conn.close(). When multiple connections raced for the same device_id shard (common in BEP incoming/outgoing race), the internal parking_lot::RwLock blocked other tokio workers, eventually freezing the entire runtime at ~T+180s.

Refactoring: Introduced RegisterAction enum to separate decision-making (under lock) from execution (lock released).

Verification:

• All 86 syncthing-net unit tests pass
• 72h stress test: previously 100% froze at T+180s → now stable past T+8h at v0.2.4 ship time
• Memory: 12 → 24 MB over 8 hours (~1.5 MB/hr, no leak)

📄 Full RCA: docs/reports/STRESS_TEST_DEATH_INVESTIGATION_2026-05-12.md

🚀 Performance Baseline (T-A1, T-B1)

Complete Criterion benchmarks established as regression baseline:

Benchmark	Throughput @1 MiB	Notes
Scanner SHA-256	1.49 GiB/s	Hardware SHA limit per core
Puller (memory)	1.46 GiB/s	Equal to scanner, no overhead
BEP encode/decode	507 MiB/s	Prost protobuf

T-B1 Rayon Validation (20-core machine):

Size	Serial	Parallel	Speedup
16 MiB	2.04 GiB/s	19.02 GiB/s	9.32x
64 MiB	1.96 GiB/s	20.69 GiB/s	10.56x
256 MiB	2.01 GiB/s	23.17 GiB/s	11.51x

Near-linear scaling confirms rayon parallelization is essential and well-implemented.

📄 Full data: docs/reports/BASELINE_2026-05-12.md

🛡️ Code Quality (T-F2)

• Unwrap audit complete: Production code has zero .unwrap(), only 15 documented .expect("invariant") calls remain.
• Runtime panic risks fixed:
  • parse().unwrap() for "0.0.0.0:0" → SocketAddr::new(IpAddr::V4(Ipv4Addr::UNSPECIFIED), 0) constructor
  • relay_url.as_ref().unwrap() → let Some(...) else continue pattern
  • TcpTransport::start() double-call panic → SyncthingError::config(...) Result
  • store.rs zero-capacity panic → cache_capacity.max(1)

📄 Audit: docs/reports/UNWRAP_AUDIT_2026-05-12.md

📐 Code Structure (T-E1)

Six 700+ line files split into mod.rs + tests.rs pattern:

• messages.rs (910) → 697 + 213
• types.rs (882) → 805 + 76 + extracted connection.rs (175) → 639 prod
• connection.rs (770) → 695 + 74
• service.rs (715) → 684 + 30
• session.rs (979) → 606 + 366

🔄 CI Hardening (T-G2)

• New bench-smoke job: All benchmarks compile and run smoke-tested per push
• clippy::await_holding_lock lint enabled: Prevents T-F1-class deadlocks at compile time

📄 Lock audit: docs/reports/LOCK_AWAIT_AUDIT_2026-05-12.md

🧹 Dead Code Cleanup

• Removed ConnectionManager::from_arc() method (28 lines, never called)
• Removed PORT_MAP_SERVICE_TIMEOUT constant (never used)
• Fixed 4 redundant_clone instances (EventBus, FolderWatcher, BlockServer)
• Fixed 1 redundant_field_names warning

Stats

• 295 unit tests passing across 12 crates
• 0 clippy warnings (including new await_holding_lock)
• 0 fmt diffs
• 6 CI jobs all green (fmt, clippy×2, test×2, audit, file-size, bench-smoke)

Stress Test Status @ Release

• Started 2026-05-12 13:07
• Currently T+8h+ (running)
• Target: 72h
• Will publish complete report after test completion

Breaking Changes

None. v0.2.4 is fully backward compatible with v0.2.3.

v0.2.3

What's Changed

🧹 Code Quality

• Workspace-wide formatting: Applied cargo fmt to 118 Rust source files ensuring consistent style.
• Clippy compliance: Auto-fixed 10 collapsible_match lints in TUI event handling and stress test binary.

🔧 Infrastructure

• CI Matrix: All 7 jobs now pass on both ubuntu-latest and windows-latest:
  • ✅ Formatting
  • ✅ Clippy (Ubuntu + Windows)
  • ✅ Tests (Ubuntu + Windows) — 308 passed
  • ✅ Security Audit (cargo audit)
  • ✅ File Size Check (soft limit 600 lines)

📦 v0.2.2 → v0.2.3 Diff

This patch release contains no functional changes; it is a stability guarantee that the repository builds cleanly with zero warnings on both platforms.

Full Changelog: v0.2.2...v0.2.3

v0.2.2

What's Changed

Infrastructure & Quality

• 72h Stress Test Infrastructure: stress_test binary now supports unattended Windows execution with Scheduled Task auto-resume (fault injection every 30 min, file injection every 5 min). See scripts/register-stress-task.ps1.
• GitHub Community Health: Added .github/ISSUE_TEMPLATE/, .github/PULL_REQUEST_TEMPLATE.md, CODE_OF_CONDUCT.md.
• CI Matrix Upgrade: GitHub Actions now runs fmt, clippy, test, audit, and file-size checks in parallel across ubuntu-latest and windows-latest.
• README Maintenance: Fixed stale "72h not started" status, removed encoding artifacts, added Stress Test quick-start block.

Documentation

• Tuning plan published: docs/plans/TUNING_PLAN_2026-05-11.md — cross-cut performance / stability / architecture-debt roadmap.

Full Changelog: v0.2.1...v0.2.2

v0.2.1 — Block-level concurrency, REST write-path closure, CI health checks

What's Changed

🚀 Performance

• Block-level concurrency: Single-file downloads now pipeline block requests (default 16 concurrent), fixing 15MB file taking ~10min.
• Pre-allocated encoding: bep-protocol message encoder now uses BytesMut::with_capacity(msg.encoded_len()), eliminating realloc hot path.

🛠️ API & Sync

• /rest/db/override and /rest/db/revert implemented for ReceiveOnly folders (previously 501 stubs).
• POST /rest/db/scan sub parameter and device pause/resume body already active; write-path now fully closed.

🧪 Testing & Quality

• Added 14 new tests (workspace total: 308 passed / 0 failed / 3 ignored).
• bep-protocol: ClusterConfig encode/decode roundtrip.
• syncthing-core: Vector compare/dominates/increment, IndexID roundtrip, FolderSummary sync percent.
• syncthing-fs: .stignore real-world scenario audit + documented known gaps.

🔧 Infrastructure

• .github/workflows/ci.yml: automated test / clippy / audit / file-size checks on PR and push.
• scripts/check-health.ps1: local Windows health validation script.
• .github/workflows/release.yml: Windows x64 release binary now built automatically alongside Linux.

📋 Audit & Debt

• .cargo/audit.toml: 3 upstream unmaintained deps explicitly accepted with ADR.
• syncthing-fs/src/ignore.rs coverage audit: ~92% common scenarios covered.

Assets

File	Platform	Size
syncthing-x86_64-linux	Linux x64	~8 MB
syncthing-x86_64-windows.exe	Windows x64	~12 MB

Known Limitations

• Pre-release. 72h stress test delegated to remote peer execution.
• .stignore intermediate double-star (a/**/b) matches single directory level only.
• Go Syncthing full interoperability: handshake verified, file-sync pending validation.

Full Changelog: v0.2.0...v0.2.1