MySQL.jl 2.0: native Julia wire-protocol client (Connector/C removed)

.github/workflows/ci.yml

@@ -4,10 +4,26 @@ on:
     branches: [main]
     tags: ["*"]
   pull_request:
+  schedule:
+    # nightly deterministic fuzz budget (plan §8.4); the smoke batch runs in every test job
+    - cron: "17 6 * * *"
+  workflow_dispatch:
 jobs:
+  cleanroom:
+    name: Native clean-room source check
+    if: github.event_name != 'schedule'
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v5
+      - uses: julia-actions/setup-julia@v2
+        with:
+          version: "1.10"
+      - run: julia --startup-file=no scripts/check_native_cleanroom.jl
   test:
     name: Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ matrix.arch }} - ${{ github.event_name }}
+    if: github.event_name != 'schedule'
     runs-on: ${{ matrix.os }}
+    timeout-minutes: 30
     strategy:
       fail-fast: false
       matrix:
@@ -17,25 +33,131 @@ jobs:
           - nightly
         os:
           - ubuntu-latest
+          - macOS-latest
+          - windows-latest
         arch:
           - x64
+        exclude:
+          - os: macOS-latest
+            arch: x64
+        include:
+          - os: macOS-latest
+            arch: aarch64
+            version: "1.10"
+          - os: macOS-latest
+            arch: aarch64
+            version: 1
+          - os: macOS-latest
+            arch: aarch64
+            version: nightly
     steps:
       - uses: actions/checkout@v5
+      # The wire-protocol tests need no server; the server integration tests run only
+      # where Docker is available (Linux runners).
       - run: docker info
+        if: runner.os == 'Linux'
       - uses: julia-actions/setup-julia@v2
         with:
           version: ${{ matrix.version }}
           arch: ${{ matrix.arch }}
       - uses: julia-actions/cache@v2
       - uses: julia-actions/julia-buildpkg@v1
+      # The cross-platform matrix runs the serverless protocol suite only (MYSQL_INTEGRATION=0):
+      # it is fast, needs no server, and the Docker integration (live lanes + leak soak +
+      # the server integration tests) runs in the dedicated `integration` job instead.
+      # Coverage is measured by the `coverage` job.
+      - uses: julia-actions/julia-runtest@v1
+        with:
+          coverage: false
+        env:
+          MYSQL_INTEGRATION: "0"
+  integration:
+    # Live lanes (mysql:8.4, mariadb:11.4), the golden behavior manifest, the leak/lifecycle
+    # soak, and the server integration tests, all against real servers via Harbor. Linux-only
+    # (Docker) and pinned to Julia 1.10, whose GC is stable under the soak's GC-thrash (1.12+
+    # can hang on shared runners).
+    name: Integration (Docker)
+    if: github.event_name != 'schedule'
+    runs-on: ubuntu-latest
+    timeout-minutes: 90
+    steps:
+      - uses: actions/checkout@v5
+      - run: docker info
+      - uses: julia-actions/setup-julia@v2
+        with:
+          version: "1.10"
+      - uses: julia-actions/cache@v2
+      - uses: julia-actions/julia-buildpkg@v1
       - uses: julia-actions/julia-runtest@v1
+        with:
+          coverage: false
+  coverage:
+    # Line-coverage gate (plan §8.15). The serverless fake-peer suite alone covers ~96% of
+    # src/, so the gate needs neither Docker nor the leak soak and stays fast; it also
+    # feeds codecov.
+    name: Native coverage
+    if: github.event_name != 'schedule'
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    steps:
+      - uses: actions/checkout@v5
+      - uses: julia-actions/setup-julia@v2
+        with:
+          version: "1"
+      - uses: julia-actions/cache@v2
+      - uses: julia-actions/julia-buildpkg@v1
+      - run: julia --project=. --code-coverage=@src --startup-file=no -e 'include("test/protocol/runtests.jl")'
       - uses: julia-actions/julia-processcoverage@v1
+      - run: julia --startup-file=no scripts/check_native_coverage.jl lcov.info 0.85
       - uses: codecov/codecov-action@v5
         with:
           files: lcov.info
           token: ${{ secrets.CODECOV_TOKEN }}
+  fuzz:
+    # Budgeted mutation fuzzing in isolated worker processes (test/protocol/fuzz.jl via
+    # scripts/fuzz.jl); scheduled nightly, or run on demand with workflow_dispatch.
+    name: Nightly fuzz
+    if: github.event_name == 'schedule' || github.event_name == 'workflow_dispatch'
+    runs-on: ubuntu-latest
+    timeout-minutes: 45
+    steps:
+      - uses: actions/checkout@v5
+      - uses: julia-actions/setup-julia@v2
+        with:
+          version: "1"
+      - uses: julia-actions/cache@v2
+      - uses: julia-actions/julia-buildpkg@v1
+      - run: julia --project=. scripts/fuzz.jl --minutes 30 --seed $(date +%Y%m%d)000000 --out fuzz_failures
+      - uses: actions/upload-artifact@v4
+        if: failure()
+        with:
+          name: fuzz-failures
+          path: fuzz_failures/
+  perf:
+    # §8.9 gates (correctness/allocation) plus the wall-clock timing report. They are too
+    # heavy for the shared PR test lanes, so they run here on the nightly schedule and on
+    # demand (MYSQL_PERF_GATES=1), with coverage off. Cross-driver comparisons (MySQL.jl 1.x,
+    # other clients) live in bench/, outside CI.
+    name: Performance gates (§8.9)
+    if: github.event_name == 'schedule' || github.event_name == 'workflow_dispatch'
+    runs-on: ubuntu-latest
+    timeout-minutes: 60
+    steps:
+      - uses: actions/checkout@v5
+      - run: docker info
+      - uses: julia-actions/setup-julia@v2
+        with:
+          version: "1"
+      - uses: julia-actions/cache@v2
+      - uses: julia-actions/julia-buildpkg@v1
+      - uses: julia-actions/julia-runtest@v1
+        with:
+          coverage: false
+        env:
+          MYSQL_PERF_GATES: "1"
   docs:
     name: Documentation
+    if: github.event_name != 'schedule'
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v5

.gitignore

@@ -42,3 +42,8 @@ deps/build.log
 Manifest.toml
 
 .vscode/
+
+# Documentation build and native-backend artifacts
+docs/build/
+fuzz_failures/
+*.cov

Project.toml

@@ -1,33 +1,38 @@
 name = "MySQL"
 uuid = "39abe10b-433b-5dbd-92d4-e302a9df00cd"
 author = ["quinnj"]
-version = "1.6.0"
+version = "2.0.0"
 
 [deps]
 DBInterface = "a10d1c49-ce27-4219-8d33-6db1a4562965"
 Dates = "ade2ca70-3891-5945-98fb-dc099432e06a"
 DecFP = "55939f99-70c6-5e9b-8bb0-5071ed7d61fd"
-Libdl = "8f399da3-3557-5675-b5ff-fb832c97cbdb"
-MariaDB_Connector_C_jll = "aabc7e14-95f1-5e66-9f32-aea603782360"
 OpenSSL_jll = "458c3c95-2e84-50aa-8efc-19380b2a3a95"
 Parsers = "69de0a69-1ddd-5017-9359-2bf0b02dc9f0"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
+Reseau = "802f3686-a58f-41ce-bb0c-3c43c75bba36"
+SHA = "ea8e919c-243c-51af-8825-aaa63cd721ce"
 Tables = "bd369af6-aec1-5ad0-b16a-f7cc5008161c"
 
 [compat]
+Chairmarks = "1.3"
 DBInterface = "2.5"
 DecFP = "0.4.9, 0.4.10, 1"
 Harbor = "1.0.3"
-MariaDB_Connector_C_jll = "3.1.12"
 OpenSSL_jll = "3"
-Parsers = "0.3, 1, 2"
+Parsers = "2"
+Random = "1"
+Reseau = "1.4.1"
+SHA = "0.7.0, 1"
 Tables = "1"
 julia = "1.10"
 
 [extras]
+Chairmarks = "0ca39b1e-fe0b-4e98-acfc-b1656634c4de"
 Harbor = "af79dbb9-1a80-47ad-8928-192a4af69376"
-Sockets = "6462fe0b-24de-5631-8697-dd941f90decc"
+Logging = "56ddb016-857b-54e1-b83d-db4d58db5568"
+Printf = "de0858da-6303-5e67-8744-51eddeeeb8d7"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [targets]
-test = ["Harbor", "Sockets", "Test"]
+test = ["Chairmarks", "Harbor", "Logging", "Printf", "Test"]

README.md

@@ -1,21 +1,41 @@
-
-# MySQL
-
-[![docs](https://img.shields.io/badge/docs-latest-blue&logo=julia)](https://mysql.juliadatabases.org/dev/)
-[![CI](https://github.com/JuliaDatabases/MySQL.jl/workflows/CI/badge.svg)](https://github.com/JuliaDatabases/MySQL.jl/actions?query=workflow%3ACI)
-[![codecov](https://codecov.io/gh/JuliaDatabases/MySQL.jl/branch/master/graph/badge.svg)](https://codecov.io/gh/JuliaDatabases/MySQL.jl)
-
-[![deps](https://juliahub.com/docs/MySQL/deps.svg)](https://juliahub.com/ui/Packages/MySQL/xeTdU?t=2)
-[![version](https://juliahub.com/docs/MySQL/version.svg)](https://juliahub.com/ui/Packages/MySQL/xeTdU)
-[![pkgeval](https://juliahub.com/docs/MySQL/pkgeval.svg)](https://juliahub.com/ui/Packages/MySQL/xeTdU)
-
-Package for interfacing with MySQL databases from Julia via the MariaDB C connector library, version 3.1.6.
-
-## Documentation
-
-[![Stable](https://img.shields.io/badge/docs-stable-blue.svg)](https://mysql.juliadatabases.org/stable)
-[![Dev](https://img.shields.io/badge/docs-dev-blue.svg)](https://mysql.juliadatabases.org/dev)
-
+
+# MySQL
+
+[![docs](https://img.shields.io/badge/docs-latest-blue&logo=julia)](https://mysql.juliadatabases.org/dev/)
+[![CI](https://github.com/JuliaDatabases/MySQL.jl/workflows/CI/badge.svg)](https://github.com/JuliaDatabases/MySQL.jl/actions?query=workflow%3ACI)
+[![codecov](https://codecov.io/gh/JuliaDatabases/MySQL.jl/branch/master/graph/badge.svg)](https://codecov.io/gh/JuliaDatabases/MySQL.jl)
+
+[![deps](https://juliahub.com/docs/MySQL/deps.svg)](https://juliahub.com/ui/Packages/MySQL/xeTdU?t=2)
+[![version](https://juliahub.com/docs/MySQL/version.svg)](https://juliahub.com/ui/Packages/MySQL/xeTdU)
+[![pkgeval](https://juliahub.com/docs/MySQL/pkgeval.svg)](https://juliahub.com/ui/Packages/MySQL/xeTdU)
+
+Package for interfacing with MySQL databases from Julia.
+
+Since 2.0, MySQL.jl implements the MySQL client/server wire protocol natively in Julia
+(built on [Reseau.jl](https://github.com/JuliaServices/Reseau.jl) for TCP/TLS) — no C
+client library. 1.x used the MariaDB Connector/C library; see the
+[migration guide](https://mysql.juliadatabases.org/dev/migration/) for the differences.
+
+## Documentation
+
+[![Stable](https://img.shields.io/badge/docs-stable-blue.svg)](https://mysql.juliadatabases.org/stable)
+[![Dev](https://img.shields.io/badge/docs-dev-blue.svg)](https://mysql.juliadatabases.org/dev)
+
+## Usage
+
+```julia
+conn = DBInterface.connect(MySQL.Connection, host, user, passwd; db="mydb", port=3306)
+cursor = DBInterface.execute(conn, "SELECT * FROM mytable")   # a Tables.jl-compatible cursor
+stmt = DBInterface.prepare(conn, "INSERT INTO mytable (a, b) VALUES (?, ?)")
+DBInterface.execute(stmt, (1, "two"))
+DBInterface.close!(conn)
+```
+
+`DBInterface.execute`/`prepare`/`executemany`/`executemultiple`, Tables.jl cursors,
+`MySQL.load`, and transactions are all supported; see the
+[documentation](https://mysql.juliadatabases.org/dev/). The transport is TCP or TLS
+(no Unix sockets, named pipes, or compression yet — these raise clear errors).
+
 ## Contributing
 
 The test suite manages its own temporary MySQL container via Harbor.jl. The only prerequisite is a working Docker daemon:

bench/README.md

@@ -0,0 +1,40 @@
+# Cross-driver benchmarks
+
+Before 2.0, `test/perf/perf_gates.jl` asserted native-vs-Connector/C timing ratios inside
+`Pkg.test`. Those ratio gates retired with the C backend; the in-repo §8.9 gates now assert
+correctness, allocation budgets, and buffer limits, and print a wall-clock timing report.
+
+This directory keeps the cross-driver comparison repeatable:
+
+```sh
+julia --project=. bench/run.jl
+```
+
+builds two temp environments — the current checkout (native) and `MySQL@1` (MariaDB
+Connector/C) from the registry — starts the same Docker fixture servers the §8.9 gates use
+(`mysql:8.4`, `--max-allowed-packet=128M`, TLS disabled), runs `bench/child.jl` once per
+environment, and prints per-benchmark seconds plus the mysql@1/native ratio (>1x means the
+native client is faster).
+
+Reference numbers (macOS ARM, Docker, Julia 1.12, this harness after the 2.0 round-trip
+perf round; ratio is mysql@1 time / native time, >1x = native faster):
+
+| benchmark | native | mysql@1 (Connector/C) | speed |
+|---|---|---|---|
+| 100k executemany | 16.97s | 23.46s | 1.38x |
+| 1M-row text scan | 0.4512s | 0.5098s | 1.13x |
+| 1M tiny/NULL rows | 0.1780s | 0.2018s | 1.13x |
+| 1M-row binary (prepared) scan | 0.2866s | 0.2978s | 1.04x |
+| 64 MiB blob fetch | 0.0591s | 0.0551s | 0.93x |
+| 10k round trips (plain) | 2.701s | 2.292s | 0.85x |
+
+Per-command costs after the round: COM_PING 3 allocs, buffered `SELECT 1` 35 allocs,
+repeated prepared execute 22 allocs (framed directly into the output buffer, cached
+statement schema, lazy name lookup). The remaining round-trip gap vs the C client is
+task-wakeup latency in the transport layer (a blocking `recv` wakes the C client directly;
+the native client parks on the poller and pays a second thread wake ≈17µs). The
+`direct_wait_ms` connect option (with a Reseau newer than 1.4.1, see Reseau PR #152)
+removes that second wake by blocking the waiting task's own OS thread in a bounded
+`poll(2)`: measured ~10-15µs less per round trip against Docker (COM_PING ~150 → ~135µs,
+20k-row executemany −13%) at the cost of other tasks on that thread waiting up to the
+budget.