motedb-python 0.15.1__tar.gz → 0.16.0__tar.gz

motedb_python-0.16.0/.github/FUNDING.yml

@@ -0,0 +1,11 @@
+# MoteDB is maintained by one person, on evenings and weekends.
+# If MoteDB saved you time, or if you depend on it to run a robot in
+# production, a small GitHub Sponsors contribution is what keeps it going.
+#
+# See the community page for everything else:
+# https://github.com/motedb/motedb/blob/main/community/README.md
+
+github: [motedb]
+# Once a second funding route exists (or if you prefer not to use GitHub
+# Sponsors), replace the line above with:
+# custom: ["<your-funding-url>"]

{motedb_python-0.15.1 → motedb_python-0.16.0}/.github/workflows/ci.yml

@@ -32,6 +32,15 @@ jobs:
       - name: cargo clippy (advisory — warnings do not fail the job)
         run: cargo clippy --all-targets --all-features --no-deps
         continue-on-error: true
+      # e3ebb24 shipped a mid-match wildcard in BOTH bindings: every concrete
+      # arm after it became unreachable (a warn-only lint) and Python returned
+      # None for every value. The root-crate clippy above never sees the
+      # binding crates — they are not workspace members. Deny the lint here.
+      - name: bindings check — unreachable_patterns denied
+        run: |
+          for m in bindings/python/Cargo.toml bindings/node/Cargo.toml; do
+            RUSTFLAGS="-D unreachable_patterns" cargo check --manifest-path "$m" --quiet
+          done
 
   # ── Supply-chain: vulnerability + license gate (cargo-deny) ──────────────
   # Fails on any NEW RustSec advisory (vulnerability/unsound). Four known
@@ -131,6 +140,7 @@ jobs:
         #     hardware-speed assumptions shared runners can't meet. Tracked in
         #     BENCHMARK.md instead.
         run: |
+          set +e
           cargo test --workspace --all-features -- \
             --test-threads=1 \
             --skip bench_ --skip stress_ --skip perf_ \
@@ -150,13 +160,29 @@ jobs:
             --skip test_scalability_latency \
             --skip test_scalability_memory \
             --skip test_durability_large_dataset_checkpoint_recovery \
-            --skip test_performance_guarantees
+            --skip test_performance_guarantees \
+            > /tmp/ci-test.log 2>&1
+          rc=$?
+          if [ $rc -ne 0 ]; then
+            # 🔑 Anonymous-readable failure surface: the trailing `Running
+            # tests/…` line before the failure block names the failing
+            # binary. (The separate diagnostics step proved unreliable — its
+            # `{ } || …` grouping suppressed set -e and silently skipped.)
+            grep -E "Running (tests|unittests)|FAILED|panicked at|^failures:" /tmp/ci-test.log \
+              | tail -8 | while IFS= read -r line; do
+                echo "::error title=integration-fail::$line"
+              done
+            echo "---- tail of test log ----"
+            tail -30 /tmp/ci-test.log
+          fi
+          exit $rc
       - name: cargo test (workspace, hardware-threshold tests skipped) [attempt 2]
         id: ci2
         if: steps.ci1.outcome == 'failure'
         env:
           CI: "1"
         run: |
+          set +e
           cargo test --workspace --all-features -- \
             --test-threads=1 \
             --skip bench_ --skip stress_ --skip perf_ \
@@ -176,7 +202,22 @@ jobs:
             --skip test_scalability_latency \
             --skip test_scalability_memory \
             --skip test_durability_large_dataset_checkpoint_recovery \
-            --skip test_performance_guarantees
+            --skip test_performance_guarantees \
+            > /tmp/ci-test.log 2>&1
+          rc=$?
+          if [ $rc -ne 0 ]; then
+            # 🔑 Anonymous-readable failure surface: the trailing `Running
+            # tests/…` line before the failure block names the failing
+            # binary. (The separate diagnostics step proved unreliable — its
+            # `{ } || …` grouping suppressed set -e and silently skipped.)
+            grep -E "Running (tests|unittests)|FAILED|panicked at|^failures:" /tmp/ci-test.log \
+              | tail -8 | while IFS= read -r line; do
+                echo "::error title=integration-fail::$line"
+              done
+            echo "---- tail of test log ----"
+            tail -30 /tmp/ci-test.log
+          fi
+          exit $rc
       - name: verdict
         if: always()
         run: |
@@ -191,8 +232,12 @@ jobs:
           # 🔑 K-debug: cargo 顶层一次跑时，crashing binary 会以 exit 101
           # 中断整个套件且不留 FAILED 行。逐 bin 运行并把每个 bin 的退出码
           # 通过 ::error:: annotations 上报（匿名可读）。
+          # 2026-10-21 修复: 曾有一行多余的 `} || echo …`（未配对大括号）
+          # 让本步骤自身以 exit 2 崩溃，逐 bin 上报从未生效过；同时对
+          # rc=101（真实测试失败，有 FAILED 行）也提取失败测试名，不再只
+          # 处理非 101 的编译错误。
           {
-            echo "### mac integration per-binary status"
+            echo "### integration per-binary status ($RUNNER_OS)"
             BINS=$(cargo test --workspace --all-features -- --list 2>/dev/null \
               | grep -E "^Running|^ " >/dev/null; cargo test --workspace \
               --all-features --no-run 2>&1 | grep -oE "Executable tests/[a-z0-9_]+\.rs" \
@@ -213,14 +258,13 @@ jobs:
                 --skip test_performance_guarantees --no-fail-fast \
                 > /tmp/one.log 2>&1
               rc=$?
-              echo "::error title=mac-integration::$t exit=$rc"
-              if [ $rc -ne 0 ] && [ $rc -ne 101 ]; then
-                grep -E "panicked at|FAILED" /tmp/one.log | head -5 \
+              echo "::error title=integration-$RUNNER_OS::$t exit=$rc"
+              if [ $rc -ne 0 ]; then
+                grep -E "^test .* FAILED|^failures:|panicked at" /tmp/one.log | head -8 \
                   | while read -r line; do echo "::error ::$line"; done
               fi
             done
           } || echo "per-binary diagnostics done" >> "$GITHUB_STEP_SUMMARY"
-          } || echo "diagnostics pass done" >> "$GITHUB_STEP_SUMMARY"
 
 
   # ── Cross-compile: aarch64-unknown-linux-gnu (build only) ────────────────
@@ -287,7 +331,7 @@ jobs:
           args: --release
       - name: Install wheel
         run: pip install bindings/python/target/wheels/*.whl
-      - name: Run E2E workload (11 suites)
+      - name: Run E2E workload (12 suites)
         env:
           MOTE_CLI: ${{ github.workspace }}/target/release/motedb-cli
         run: python bindings/python/e2e/e2e_workload.py --root /tmp/e2e

{motedb_python-0.15.1 → motedb_python-0.16.0}/.github/workflows/node-publish.yml

@@ -256,17 +256,19 @@ jobs:
         # root loader whose optionalDependencies point at them). Auth comes
         # from setup-node's .npmrc + NODE_AUTH_TOKEN. `napi pre-publish` is
         # deliberately NOT used: it assumes a GitHub release flow we don't run.
-        # Idempotent + burst-safe: skips packages already on the registry
+        # Idempotent + burst-safe: skips versions already on the registry
         # (EPUBLISHCONFLICT would otherwise abort the loop on re-runs) and
-        # paces publishes ~3 min apart with retries — npm's anti-spam
-        # heuristics reject rapid-fire new-package creation from new accounts.
+        # paces publishes with escalating retries — npm's anti-spam
+        # heuristics reject rapid-fire new-package creation from new accounts
+        # (v0.15.1 tag run: the one brand-new platform package burned all 3
+        # fast retries; a cooldown + re-dispatch is the recovery path).
         working-directory: bindings/node
         env:
           NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
         run: |
           set -e
           pub() {
-            local d="$1" name ver try
+            local d="$1" name ver try wait
             name=$(jq -r .name "$d/package.json")
             ver=$(jq -r .version "$d/package.json")
             # Compare the PINNED version, not mere existence: after a version
@@ -276,17 +278,27 @@ jobs:
             if npm view "$name@$ver" version >/dev/null 2>&1; then
               echo "  $name@$ver already on npm — skip"; return 0
             fi
-            for try in 1 2 3; do
-              echo "::group::publish $name (attempt $try)"
+            # Escalating backoff: anti-spam blocks clear after a cooldown,
+            # not immediately — fast retries just burn the attempts.
+            for try in 1 2 3 4 5; do
+              echo "::group::publish $name@$ver (attempt $try)"
               if (cd "$d" && npm publish --access public); then
-                echo "::endgroup::"; echo "  published $name"; return 0
+                echo "::endgroup::"; echo "  published $name@$ver"; return 0
               fi
               echo "::endgroup::"
-              [ "$try" = 3 ] && break
-              echo "  retry in 180s (npm anti-spam/rate heuristics)"; sleep 180
+              [ "$try" = 5 ] && break
+              case "$try" in
+                1) wait=120 ;; 2) wait=300 ;; 3) wait=600 ;; *) wait=900 ;;
+              esac
+              echo "  retry in ${wait}s (npm anti-spam/rate heuristics)"; sleep "$wait"
             done
-            echo "::error::gave up publishing $name after 3 attempts"
+            echo "::error::gave up publishing $name@$ver after 5 attempts"
             return 1
           }
-          for d in npm/*/; do pub "$d" || exit 1; sleep 90; done
+          # 主包最优先 — 它是 `npm install motedb` 的入口; 平台包是
+          # optionalDependencies, 单个被 npm 反垃圾拦截不应拖死主包
+          # (失败让 job 标红以便跟踪, 但其余包继续发)。
           pub . || exit 1
+          rc=0
+          for d in npm/*/; do pub "$d" || rc=1; sleep 120; done
+          exit $rc

motedb_python-0.16.0/CAPABILITIES.md

@@ -0,0 +1,178 @@
+# Capability Manifest
+
+> This file is the **single source of truth** for what MoteDB can do.
+> The website, the README, blog posts and any external material should cite
+> this manifest rather than restating feature claims.
+> The point: every line below should map to an implementation and a test.
+
+**Version baseline:** v0.15.2 · **Last verified:** 2026-10-24
+
+Legend: `✅ supported` · `🟡 partial (see note)` · `❌ not supported` · `🟢 differentiator`
+
+> If you find a line here that does not match actual behaviour, please open an
+> issue — **those are the ones we triage first**.
+
+---
+
+## 1. SQL engine
+
+| Capability | Status | Notes |
+|---|---|---|
+| `CREATE TABLE` / `CREATE INDEX` / `DROP TABLE [IF EXISTS]` / `DROP INDEX` | ✅ | Column, vector, text, spatial, timestamp index variants |
+| `ALTER TABLE` | ✅ | `ADD COLUMN`, `AUTO_INCREMENT = N` |
+| `INSERT` (incl. `INSERT ... SELECT`) | ✅ | |
+| `UPDATE` / `DELETE` | ✅ | |
+| `SELECT` with `WHERE` | ✅ | |
+| `JOIN` — INNER / LEFT / RIGHT / FULL | ✅ | |
+| Subqueries in `WHERE`, derived tables (`FROM (SELECT ...)`) | ✅ | |
+| `GROUP BY` | ✅ | Columns, expressions (`id % 5`), and SELECT aliases |
+| `HAVING` | ✅ | Incl. aggregate aliases |
+| `ORDER BY` | ✅ | Multi-key, expressions, non-projected columns, `NULLS FIRST/LAST` |
+| `LIMIT` / `OFFSET` | ✅ | |
+| `DISTINCT` (rows) / `COUNT(DISTINCT col)` | ✅ | |
+| Aggregates: `COUNT` `SUM` `AVG` `MIN` `MAX` `STDDEV` `VARIANCE` | ✅ | |
+| `UNION` / `UNION ALL` | ✅ | |
+| `CASE WHEN ... THEN ... ELSE ... END` | ✅ | |
+| `WITH` / CTEs, incl. `WITH RECURSIVE` | ✅ | Semi-naive fixed-point iteration, `UNION ALL` bodies, chained CTEs |
+| Window functions: `ROW_NUMBER` `RANK` `DENSE_RANK` `NTILE` `LAG` `LEAD` `FIRST_VALUE` `LAST_VALUE` + aggregates | ✅ | `OVER (PARTITION BY … ORDER BY …)` |
+| Explicit window frames (`ROWS` / `RANGE BETWEEN <bound> AND <bound>`) | ✅ 🟢 | ROWS = physical rows; RANGE = peer groups, or value intervals (`RANGE BETWEEN 10 PRECEDING AND 10 FOLLOWING`) on a numeric/timestamp order key. Default frame is the RANGE peer-group frame. Tested in `tests/test_window_frames.rs` |
+| Scalar functions: `UPPER` `LOWER` `LENGTH` `TRIM` `SUBSTR` `REPLACE` `CONCAT` `INSTR` `COALESCE` `ROUND` `ABS` + arithmetic | ✅ | `CONCAT` skips NULL args (SQLite/Postgres semantics); `\|\|` propagates NULL; `ROUND` is binary-exact decimal rounding |
+| `EXPORT TABLE t TO 'x.parquet'` / `IMPORT PARQUET 'x' INTO t` | ✅ 🟢 | Columnar dump / batch load through the insert pipeline; pyarrow / pandas / DuckDB round-trip |
+| `EXPORT LEROBOT <table> TO 'dir/' [FPS n] [TASKS col]` | ✅ 🟢 | Writes a LeRobot v3 dataset on disk that the official `LeRobotDataset` (lerobot ≥ 0.6) loads directly |
+| `DECIMAL` / `DATE` types | ❌ | Planned — see [ROADMAP](./ROADMAP.md) |
+| Cross-statement server-side cursors | ❌ | Planned — see [ROADMAP](./ROADMAP.md) |
+
+---
+
+## 2. Multimodal types
+
+| Type | Status | Notes |
+|---|---|---|
+| `VECTOR(n)` | ✅ 🟢 | Index-accelerated KNN; `ORDER BY emb <-> '[…]'` (L2) or `<~>` (cosine) |
+| `BLOB` | 🟡 | Byte-exact round-trip, **16 MiB cap per value** |
+| `TIMESTAMP` | ✅ 🟢 | Microsecond precision; `TIMESERIES(ts)` adds time-series semantics |
+| `TEXT` | ✅ | Built-in tokenizer (jieba and others configurable) |
+| Spatial coordinates | ✅ 🟢 | R-Tree index; `ST_POINT` / `ST_WITHIN` / `ST_DISTANCE` / `ST_KNN`, and `loc <-> ST_POINT(x, y)` distance ordering |
+| Array types (`TEXT[]` / `ARRAY<FLOAT>`) | ❌ | Use BLOB, a separate table, or TEXT + JOIN |
+
+### ⚠️ Known BLOB gaps (unchanged since v0.13)
+
+- Scalar functions over BLOB columns (e.g. `LENGTH(payload)`) are **not wired through all executor paths**
+- BLOB output under `ORDER BY` is **not wired through all executor paths**
+- **Verified working:** bare projection, equality (`WHERE payload = X'…'`), time-window scans, parameter binding
+
+---
+
+## 3. Indexes
+
+| Index | Status | Notes |
+|---|---|---|
+| Column index | ✅ 🟢 | Equality / range |
+| Vector index | ✅ 🟢 | **DiskANN** (FreshDiskANN + rerank). **Not** HNSW, not IVF |
+| Full-text index | ✅ 🟢 | BM25 ranking + snippet; `MATCH(col) AGAINST('q')`; multi-word defaults to AND (SQLite FTS5-compatible), explicit `a OR b` unions |
+| Spatial index | ✅ 🟢 | R-Tree |
+| Time-series index | ✅ 🟢 | Range scan + compaction + `TIME_BUCKET` aggregation + TTL |
+
+> 📌 **Correction notice.** Older material describing the vector index as
+> "HNSW / IVF" is out of date. The current implementation is **DiskANN**.
+> This file and `docs/08-vector-index.md` are authoritative.
+
+---
+
+## 4. Transactions & consistency
+
+| Capability | Status | Notes |
+|---|---|---|
+| `BEGIN` / `COMMIT` / `ROLLBACK` | ✅ | |
+| Savepoints | ✅ | |
+| Read-your-writes inside a transaction | ✅ | Scans, aggregates, point lookups |
+| MVCC | ✅ | |
+| WAL group commit | ✅ 🟢 | |
+| `LATEST BY` on tables > 1M rows | 🟡 | Performance not yet optimized (large-table merge path) |
+
+---
+
+## 5. Retrieval
+
+| Capability | Status | Notes |
+|---|---|---|
+| Vector KNN | ✅ 🟢 | `ORDER BY emb <-> '[…]'`, index-accelerated |
+| Full-text search | ✅ 🟢 | BM25 ranking |
+| **Hybrid search (BM25 × KNN, RRF fusion)** | ✅ 🟢 | One call — you don't implement score calibration yourself |
+| Time-series window aggregation | ✅ 🟢 | `TIME_BUCKET` + TTL |
+| Spatial range / KNN | ✅ 🟢 | `ST_WITHIN` / `ST_DISTANCE` / `ST_KNN` |
+
+---
+
+## 6. Data shapes & deployment
+
+| Capability | Status | Notes |
+|---|---|---|
+| Collection: high-performance directory | ✅ 🟢 | Atomic publish via temp + fsync + rename, plus WAL group commit |
+| Transport: single-file snapshot | ✅ 🟢 | `EXPORT DB TO '…'`; per-file CRC32, optional zstd, incremental re-export |
+| Deployment: read-only container | ✅ 🟢 | All write paths rejected, multiple concurrent readers |
+| MCAP import | ✅ 🟢 | |
+| Index self-healing | ✅ 🟢 | Rebuilds a corrupt index from source data |
+| Crash recovery | ✅ 🟢 | WAL replay + segment migration + orphan segment adoption |
+| `for_edge()` low-memory profile | ✅ | Tuned for small-memory devices |
+
+---
+
+## 7. Interop & SDKs
+
+| Surface | Status | Notes |
+|---|---|---|
+| Rust crate | ✅ | `cargo add motedb` (Rust ≥ 1.87) |
+| Python | ✅ | `pip install motedb-python` — imported as `motedb`. The `motedb` name on PyPI is **not yet claimed**; use `motedb-python` |
+| Arrow | ✅ 🟢 | `query_arrow()` → canonical `pyarrow.Table`, vector columns as `FixedSizeList<float32>` |
+| pandas | ✅ | `query_pandas()` |
+| numpy bulk load | ✅ | `insert_arrays()` |
+| Node.js | 🟡 | npm package **is published** (`motedb@0.15.2`). Prebuilt binaries cover macOS (arm64/x64) and Linux (x64/arm64, glibc); **musl and Windows prebuilds are not published yet**, so `npm install motedb` on Windows currently yields no native binary |
+| Go | 🟡 | Binding lives in `bindings/go` (with `go.mod`), but is **not published as a tagged Go module** yet — the module proxy only resolves a pseudo-version off the default branch |
+
+---
+
+## 8. Engineering quality
+
+| Practice | Status | Notes |
+|---|---|---|
+| Crash-injection tests | ✅ 🟢 | Resident in CI |
+| Differential fuzzing (SQLite as oracle) | ✅ 🟢 | Permanent regression suite, including JOIN diffing |
+| Corruption-recovery tests | ✅ 🟢 | |
+| Machine-independent performance gate | ✅ 🟢 | A complexity regression turns CI red |
+| `cargo fmt --check` strict gate | ✅ 🟢 | Whole tree is fmt-clean |
+| cargo-deny supply-chain audit | ✅ 🟢 | Vulnerabilities + licenses |
+| clippy | 🟡 advisory | CI runs clippy in advisory mode (`continue-on-error`); a backlog of style lints is being cleared. **New commits must not introduce new warnings** |
+| Prebuilt binaries | 🟡 | macOS (arm64/x64) and Linux (x64/arm64) prebuilt. Windows works through `cargo add motedb` (source build); a Windows prebuild is on the way |
+
+---
+
+## 9. Explicitly not supported
+
+> Written down so you don't have to discover it the hard way.
+
+| Capability | Why |
+|---|---|
+| **Built-in inference engine / ONNX** | MoteDB is a storage and query layer, not an inference runtime. We invest in **Arrow interop** and leave inference in your own stack |
+| **Distributed / multi-node replication** | The current target is single-machine edge |
+| **Server mode / multi-process concurrent writes** | Contradicts the "embedded, zero-ops" positioning |
+| **Multi-user permission system** | In embedded deployments the host application owns that |
+| **Array types** | Use BLOB, a separate table, or TEXT + JOIN |
+| **`DECIMAL` / `DATE` types** | Planned |
+| **Cross-statement server-side cursors** | Planned |
+
+---
+
+## 10. Performance methodology
+
+Full benchmark data: [`BENCHMARK.md`](./BENCHMARK.md),
+[`PERFORMANCE_REPORT.md`](./PERFORMANCE_REPORT.md), and
+[`docs/benchmark_methodology.md`](./docs/benchmark_methodology.md).
+
+> ⚠️ **Citation rule.** Every performance number published anywhere must carry
+> its **machine configuration, data scale, comparison baseline, and measurement
+> method**. Numbers without that context should not be released.
+
+---
+
+*Found something here that does not match real behaviour? Open an issue.*

{motedb_python-0.15.1 → motedb_python-0.16.0}/CHANGELOG.md

@@ -1,5 +1,123 @@
 # Changelog
 
+## [0.16.0] — 2026-10-30
+
+> 28 轮系统性猎虫与验证后的正确性版本。**42 个缺陷修复**（含 1 个数据
+> 损坏级、多个"静默错误结果"类），SQL 兼容性大幅提升（隐式别名、复合
+> 主键验证、中文全文检索、时间戳算术），四层 API（Rust/Node/Go/Python）
+> 与 MCP 全部在当前引擎上端到端验证。全面性能 A/B 对照确认无回归
+> （ORDER BY 快 ~28%，查询内存省 18%，其余持平）。
+
+### 静默错误结果类（最关键）
+
+- **🔴🔴 数据损坏**: 向量列在 UPDATE+DELETE+段合并后读成 65535 维全零
+  向量 — 列段合并的墓碑/ALTER 占位写裸 [0xFFFF]，逐行 [dim:u16] 编码里
+  是合法非空维度且无载荷 → 顺序走读者错位、同列后续全雪崩。占位改
+  [0x0000] (dim 0 = NULL)。回归: KNN 与暴力扫描全阶段逐 id 一致
+- WHERE 未知列被静默吞掉 — 单表 SELECT/UPDATE/DELETE 拼错列名返回
+  0 行/affected_rows: 0 无提示；语句入口 schema 绑定校验立即报
+  ColumnNotFound
+- SHOW TABLES / DESCRIBE 流式分发丢弃结果行（db.query 返回 []、
+  MCP list_tables 空）——行数据流经完整路径
+- 日期函数对 DATE 列返回 NULL（YEAR/MONTH/DAY_OF_MONTH/DATEDIFF）
+- CAST(x AS TEXT) 泄漏 Rust Debug 串（`Date { days: 19753 }`）——
+  现按 SQL 文本形态渲染（新增 Timestamp::to_iso）
+- ISO 时间戳小数秒被丢弃（'…12:00:00.5' 的 .5 消失，schema-order
+  INSERT 路径）——解析到 µs，双实现去重
+- 中文 MATCH 恒不命中 — FTS 索引与逐行 MATCH 都硬编码空白分词器；
+  新增 AutoTokenizer（CJK→Jieba，其余→空白）两路径统一
+- TIMESTAMP ± INTEGER(µs) 算术 TypeError 被吞 → WHERE 静默空
+- TIMESTAMP 列 BETWEEN 文本字面量恒空（vector_exec 预解析缺口，
+  `>= AND <=` 却正常）——同一 col_lit 预解析
+- GROUP BY 表达式 + ORDER BY 同表达式乱序（快路径丢键 + full_row
+  zip 错位配对）——表达式键读分组投影值，不可解析则明确报错
+- NULLS FIRST/LAST 被四条排序路径中的三条静默忽略——全路径统一
+- 0x 十六进制字面量静默返回 0（词法器缺十六进制 + 隐式别名吞噬）
+- MCP describe 工具返回 DESCRIBE 表头而非 schema（行被丢弃）
+
+### SQL 兼容性
+
+- **SELECT 列表隐式别名**（`SELECT v x`、`SELECT COUNT(*) c`）——此前
+  全部死于误导性解析错误；别名可在 ORDER BY/GROUP BY/HAVING 引用
+- 验证钉住: 复合 PRIMARY KEY（重复拒绝/双键点查/复合 upsert/重开强制）、
+  列 DEFAULT 值、NOT NULL 双路径强制、LIMIT ?/OFFSET ? 绑定参数
+
+### 语义钉住（文档化，防止引擎间歧义）
+
+- 除零: 整型/浮点除数一律 DivisionByZero（PostgreSQL 语义）
+- LIKE 与 = 一致大小写敏感；表名大小写敏感/列名不敏感
+- CONCAT 忽略 NULL（PG 风格），|| 按标准传播 NULL；DATEDIFF = a−b 秒
+- 同进程二次 open() 附着同一引擎（共享句柄注册表，docs/14）；跨进程
+  flock 互斥
+- DDL 非事务（CREATE 留存+行回滚，事务内 DROP 被拒）
+- 圆整 ROUND(x,-n) half-away；DATE 函数表（docs/13）；已知限制:
+  WHERE 中除零静默 0 行（v0.16+ 候选修复）
+
+### 质量与验证
+
+- clippy 双 feature 集零警告（默认集 + node 最小集——后者此前从未
+  被门禁覆盖）
+- 绑定层全面验证: Node smoke 双绿、Go 6/6 首跑、Python 13/13、
+  MCP stdio e2e 6/6（round-20/22 修复的端到端确认）
+- 性能 A/B（round-19 基线 vs 0.16.0，同机背靠背）: 无回归；ORDER BY
+  快 ~28%，查询后 RSS 省 18%，其余持平
+- 新增回归: test_bug_hunt_v110~v116（57 项）、MCP e2e
+
+### 此前积累（0.15.2 后、猎虫前）
+
+- 向量索引旋钮 SQL 化: CREATE VECTOR INDEX ... WITH (metric=, max_degree=,
+  build_search_list=, search_list=)；构建期精确 f32 快照
+- ORDER BY emb <-> '带引号向量' 三拼写回归；CREATE INDEX WITH 子句
+  死代码复活
+
+## [0.15.2] — 2026-10-22
+
+### 功能测试轮修复 (2026-10-22 外部功能测试 24/27 → 全绿)
+
+- **🔴 P0 Node 浮点参数静默截断**: `arr.get::<i64>` 对 JS double 不报错而是
+  向零截断(9.9→9, 1e-7→0, 123.456→123), f64 分支不可达 — 任何 `?` 传入的
+  非整数 number 全部损坏(FLOAT/DOUBLE 列通杀)。BigInt 更写 9 读 null。
+  修复 = f64 先探(napi 严格语义对非 number 干净失败)按 fract() 分流
+  Integer/Float(≤2^53), BigInt 走 get_i64 无损转换(超范围显式报错)。
+  smoke_ci.js 回归: 9.9/-2.75/1e-7/2.0/123n 逐值断言
+- **快照往还(ENOENT / already exists)**: resolve_open_path 的目录判定标记
+  还是旧布局(lsm/||wal/) — P1 惰性布局的解包快照(首写前无 wal/)判定不中,
+  落到"兄弟 .mote"规则打开不存在的路径。Python 的 import_snapshot 同中
+  (观感即"Python 没有快照 API")。修复 = 标记集扩为 meta.bin/columnar_ms/
+  lsm/wal; unpack 接受已存在的**空** dest 目录(mkdir-then-import 自然用法),
+  非空仍拒。Node openSnapshot/Python import_snapshot/Rust open_snapshot 三路
+  往返回归全绿; Python 加 export_snapshot 别名(与 Node/Rust 命名对齐,
+  原生名 export_to 保留)
+- **SHOW INDEXES [FROM t]**: 引擎语法新增(parser 原只认 SHOW TABLES), 数据
+  源为权威的 index_registry(TableSchema::indexes 不维护); 输出
+  Table/Index/Column/Type 四列
+
+### 训练生态互操作 (Phase 1 §5/§6, docs/design/2026-10-industry-parity-plan.md)
+
+- **🔑 Parquet 双向**: `EXPORT TABLE t TO 'x.parquet'` / `IMPORT PARQUET 'x'
+  INTO t` — 全类型往返(INT/FLOAT/BOOL/TEXT/TIMESTAMP/VECTOR→FixedSizeList<
+  f32>/BLOB→Binary), NULL 保留, 按列名映射(乱序/子集/缺列补 NULL), 有损方向
+  报错; arrow-array/schema/parquet 59.3.1(MSRV 1.85), feature `parquet`
+  默认开(default-edge 可关)。pyarrow 双向互认实测。Python: `export_parquet` /
+  `import_parquet`
+- **🔑 EXPORT LEROBOT**: EPISODE 表 → LeRobot v3 数据集目录(官方 writer 逐文件
+  定标), 官方 `LeRobotDataset` 加载互认实测 — MCAP→SQL→训练管线打通。Python:
+  `export_lerobot(table, dir, fps, tasks=)`
+- **修复(数据丢失级)**: 含 NULL 向量的表, 无 WHERE 全表扫描整列静默变 NULL,
+  且 compaction 将丢失永久写回 — read_vectors/read_spatial 的 7 处只前进线性
+  归并改为 `scatter_by_row_id` 精确散射; 回归测试 ×2
+- **PyTorch DataLoader**: `from motedb.torch import MoteDataset` — 列式一次
+  载入 + 逐行零拷贝视图, fork-safe; docs/19 + examples/torch_dataloader.py
+- **🔑 MCP server (§8)**: `python -m motedb.mcp <db>` — stdio JSON-RPC,
+  6 个 tools(list_tables/describe/query/vector_search/episodes/doctor);
+  引擎级只读门(Database(read_only=True), 写被引擎拒绝而非字符串嗅探),
+  多读者共存。docs/20-mcp.md; pyo3 构造器新增 read_only kwarg
+- **🔑 多引擎公开基准 (§7)**: `make bench-industry` — MoteDB vs FAISS/
+  LanceDB/sqlite-vec/DuckDB-VSS/usearch, 同语料同暴力 ground truth,
+  N/DIM 规模参数化, 方法学随数据发布(含不利行)。220K×384 实测: MoteDB
+  DiskANN recall@10 1.0000 / p50 0.57ms。docs/benchmark_report_2026-10-21.md
+- CI: 测试步骤失败时直接发逐行注解(替代从未生效的独立诊断步骤)
+
 ## [0.15.1] — 2026-10-20
 
 发布渠道与安装体验 patch(docs/design/2026-10-industry-parity-plan.md Phase 0)。

motedb_python-0.16.0/CODE_OF_CONDUCT.md

@@ -0,0 +1,105 @@
+# Code of Conduct
+
+## Our pledge
+
+We as members, contributors, and maintainers pledge to make participation in
+this project a harassment-free experience for everyone, regardless of age, body
+size, visible or invisible disability, ethnicity, sex characteristics, gender
+identity and expression, level of experience, education, socio-economic status,
+nationality, personal appearance, race, religion, or sexual identity and
+orientation.
+
+We pledge to act and interact in ways that contribute to an open, welcoming,
+diverse, inclusive, and healthy community.
+
+---
+
+## Our standards
+
+### Behaviour that contributes to a positive environment
+
+- Demonstrating empathy and kindness toward other people
+- Being respectful of differing opinions, viewpoints, and experiences
+- Giving and gracefully accepting constructive feedback
+- Accepting responsibility, apologising to those affected by our mistakes, and
+  learning from the experience
+- Focusing on what is best for the overall community
+
+### Behaviour that is unacceptable
+
+- The use of sexualised language or imagery, and sexual attention or advances
+  of any kind
+- Trolling, insulting or derogatory comments, and personal or political attacks
+- Public or private harassment
+- Publishing others' private information, such as a physical or email address,
+  without their explicit permission
+- Other conduct which could reasonably be considered inappropriate in a
+  professional setting
+
+---
+
+## Scope
+
+This Code of Conduct applies within all community spaces, and also applies when
+an individual is officially representing the community in public spaces.
+Community spaces currently include:
+
+- The project's GitHub repository — issues, pull requests, discussions, commits,
+  and code review
+- The project's website and any space created in the project's name
+- Public events where the project is represented
+
+Any space the project opens later — a mailing list, a newsletter, or a chat
+channel — is covered by this Code of Conduct from the day it opens.
+
+---
+
+## Enforcement
+
+Instances of abusive, harassing, or otherwise unacceptable behaviour may be
+reported to the maintainers through any of these channels:
+
+- **GitHub issue or discussion**, mentioning `@meluo`
+- **GitHub private message** to [@meluo](https://github.com/meluo)
+- **Email** to <motedb@126.com> with the subject `[motedb conduct]`
+
+All complaints will be reviewed and investigated promptly and fairly. The
+project maintainer is obligated to respect the privacy and security of the
+reporter of any incident.
+
+**Our commitments:**
+
+- Reports are handled confidentially
+- The identity of the reporter is not disclosed publicly
+- The maintainer responds to reports within 48 hours
+
+---
+
+## Enforcement guidelines
+
+The project maintainers are responsible for clarifying and enforcing these
+standards, and will take appropriate and fair corrective action in response to
+any behaviour they deem inappropriate, threatening, offensive, or harmful.
+Enforcement actions may include:
+
+1. **Correction** — a private, written warning from the maintainers, providing
+   clarity around the nature of the violation and an explanation of why the
+   behaviour was inappropriate
+2. **Warning** — a warning with consequences for continued behaviour, including
+   a period of no interaction with the people involved
+3. **Temporary ban** — a temporary ban from any sort of interaction or public
+   communication with the community
+4. **Permanent ban** — a permanent ban from any sort of public interaction
+   within the community
+
+---
+
+## Attribution
+
+This Code of Conduct is adapted from the
+[Contributor Covenant](https://www.contributor-covenant.org/), version 2.1, and
+also draws on the
+[Rust Code of Conduct](https://www.rust-lang.org/policies/code-of-conduct).
+
+Questions or suggestions about this Code of Conduct belong in a discussion or
+an issue on this repository.