Check the allocator contract once and flatten the test suite

.github/workflows/checks.yml

@@ -26,7 +26,6 @@ jobs:
       - uses: actions/checkout@v5
         with:
           fetch-depth: 0
-      - run: sudo apt-get update && sudo apt-get install -y clang-format
       - uses: astral-sh/setup-uv@v5
       - run: uv sync --all-extras
       - run: uv run pre-commit run --all-files
@@ -48,6 +47,9 @@ jobs:
       - run: uv run python examples/02_plotting.py
       - run: uv run python examples/03_allocators.py
       - run: uv run python examples/04_sources.py
+      - run: >
+          uv run jupyter nbconvert --execute --to notebook
+          --output-dir executed_notebooks notebooks/*.ipynb
 
   test:
     runs-on: ${{ matrix.os }}
@@ -79,25 +81,14 @@ jobs:
       # Gate coverage only where every optional path runs: Hub rate limits and
       # the Linux-only minimalloc install starve the other legs of statements
       - run: >
-          uv run pytest -v
+          uv run pytest -v -n auto --dist worksteal
           ${{ matrix.os == 'ubuntu-latest'
           && matrix.python-version == env.DEFAULT_PYTHON
           && '--cov=omnimalloc --cov-fail-under=90 --cov-report=term-missing'
           || '' }}
 
-  build:
-    runs-on: ubuntu-latest
-    timeout-minutes: 60
-    steps:
-      - uses: actions/checkout@v5
-        with:
-          fetch-depth: 0
-      - uses: astral-sh/setup-uv@v5
-      - run: uv sync --all-extras
-      - run: uv build
-
-  # TODO(fpedd): clang-tidy and iwyu findings still only get reported, not
-  # enforced; compiler warnings do fail the job via -Werror.
+  # Also builds the sdist and wheel. TODO(fpedd): clang-tidy and iwyu findings
+  # still only get reported, not enforced; compiler warnings fail via -Werror.
   cpp-quality:
     runs-on: ubuntu-latest
     timeout-minutes: 60
@@ -115,7 +106,7 @@ jobs:
   checks-status:
     runs-on: ubuntu-latest
     if: always()
-    needs: [pre-commit, examples, test, build, cpp-quality]
+    needs: [pre-commit, examples, test, cpp-quality]
     steps:
       - run: exit 1
         if: contains(needs.*.result, 'failure') || contains(needs.*.result, 'cancelled')

.github/workflows/nightly.yml

@@ -42,6 +42,12 @@ jobs:
           fetch-depth: 0
       - uses: astral-sh/setup-uv@v5
       - run: uv sync --all-extras
+      - run: uv pip install ${{ env.MINIMALLOC_URL }}
+      # Fuzz every workload and allocator, then the optimality, timeout and pin oracles
+      - run: uv run python fuzz.py --deep
+        working-directory: scripts/stress
+      - run: uv run python oracles.py
+        working-directory: scripts/stress
       - run: uv run python scripts/stress_omni.py
       # Two workers keep greedy_by_all's per-process copies inside runner memory
       - run: uv run python scripts/benchmark_allocation.py --max-threads 2
@@ -51,6 +57,7 @@ jobs:
         with:
           name: nightly-figures
           path: |
+            scripts/stress/fuzz_results_stress/
             stress_results_omni/
             benchmark_results_allocation/
             benchmark_results_pressure/
@@ -108,8 +115,8 @@ jobs:
       - run: uv run ruff format --check
       - run: uv run ty check
 
-  # The examples job in checks.yml stops at 04; notebooks run nowhere else.
-  examples-notebooks:
+  # The examples job in checks.yml stops at 04: 05 is slow and hits the HF Hub.
+  examples-benchmark:
     runs-on: ubuntu-latest
     timeout-minutes: 120
     steps:
@@ -120,6 +127,3 @@ jobs:
       - run: uv sync --all-extras
       - run: uv pip install ${{ env.MINIMALLOC_URL }}
       - run: uv run python examples/05_benchmark.py
-      - run: >
-          uv run jupyter nbconvert --execute --to notebook
-          --output-dir executed_notebooks notebooks/*.ipynb

.pre-commit-config.yaml

@@ -17,7 +17,7 @@ repos:
       - id: check-merge-conflict
       - id: name-tests-test
         args: [--pytest-test-first]
-        exclude: '^tests/(markers|paths)\.py$'
+        exclude: '^tests/markers\.py$'
 
   - repo: https://github.com/pre-commit/mirrors-clang-format
     rev: v22.1.8

README.md

@@ -90,10 +90,10 @@ Run your own campaigns with the [benchmark harness](examples/05_benchmark.py).
 # Initial setup
 git clone git@github.com:fpedd/omnimalloc.git
 cd omnimalloc
-uv sync --all-extras --group dev
+uv sync --all-extras
 
 # Run tests, linting, type checking
-uv run pytest
+uv run pytest -n auto
 uv run ruff check --fix && uv run ruff format && uv run ty check
 
 # Setup pre-commit hooks (run once)

pyproject.toml

@@ -43,7 +43,6 @@ all = [
   "deap>=1.4.0",
   "huggingface-hub>=0.35.0",
   "matplotlib>=3.10.0",
-  "numpy>=1.25.0",
   "onnx>=1.19.0",
   "tqdm>=4.66.0",
 ]
@@ -86,7 +85,7 @@ test-command = "pytest {project}/tests/unit -x --no-header -q"
 
 [tool.cibuildwheel.linux]
 archs = ["auto"]
-before-all = "yum install -y cmake || apk add cmake"
+before-all = "yum install -y cmake"
 
 [tool.cibuildwheel.macos]
 archs = ["arm64"]
@@ -102,8 +101,6 @@ exclude = ["*.ipynb", "*.pyi"]
 [tool.ruff.lint]
 select = ["ALL"]
 ignore = [
-  "B019",    # cache decorator
-  "B027",    # non-abstract methods in abc
   "COM812",  # trailing comma (formatter)
   "CPY001",  # copyright notice
   "D100",    # module docstring
@@ -113,36 +110,24 @@ ignore = [
   "D104",    # package docstring
   "D105",    # magic method docstring
   "D107",    # init docstring
-  "D202",    # blank lines in docstring
   "D203",    # incompatible with D211
   "D213",    # incompatible with D212
   "D401",    # imperative mood
-  "D413",    # blank line after last section
-  "DTZ005",  # datetime.now()
   "EM101",   # string literals in exceptions
   "EM102",   # f-strings in exceptions
   "FBT001",  # boolean positional args
   "FBT002",  # boolean default args
   "FIX002",  # TODOs
   "G004",    # f-strings in logging
   "INP001",  # allow scripts without init
-  "N801",    # class names not CamelCase
   "PERF203", # try-except inside loop
-  "PGH003",  # generic type: ignore
-  "PIE790",  # ellipsis in functions
   "PLC0415", # import outside top-level
   "PLR0913", # too many arguments
   "PLR0917", # too many positional arguments
   "PLR2004", # magic value comparison
-  "PLW2901", # reassign for loop vars
-  "PYI019",  # custom TypeVar
-  "PYI034",  # return Self in dunders
-  "RET504",  # unnecessary variable before return
   "S101",    # assert
   "S311",    # pseudo-random generators
   "S603",    # allow untrusted subprocess
-  "SIM102",  # nested if
-  "SIM110",  # for loop replaceable with any()
   "T201",    # print()
   "TD003",   # missing issue link
   "TRY003",  # long exception messages
@@ -158,8 +143,6 @@ log_cli_level = "INFO"
 log_cli_format = "%(asctime)s [%(levelname)s] %(message)s (%(filename)s:%(lineno)d)"
 log_cli_date_format = "%Y-%m-%d %H:%M:%S"
 
-[tool.ty]
-
 [tool.ty.rules]
 # Ignores guarding optional imports are needed without the dep, unused with it
 unused-type-ignore-comment = "ignore"

scripts/_plot.py

@@ -0,0 +1,124 @@
+#
+# SPDX-License-Identifier: Apache-2.0
+#
+"""House style and helpers shared by the benchmark and stress scripts."""
+
+from math import isnan, nan
+from statistics import mean
+from typing import TYPE_CHECKING, Any
+
+import matplotlib.pyplot as plt
+
+if TYPE_CHECKING:
+    from matplotlib.axes import Axes
+    from matplotlib.figure import Figure
+
+SURFACE = "#fcfcfb"
+INK = "#0b0b0b"
+INK_SECONDARY = "#52514e"
+INK_MUTED = "#898781"
+GRID = "#e1e0d9"
+AXIS = "#c3c2b7"
+# Most-separated hues first; per-series markers keep identity readable
+SERIES = ("#2a78d6", "#1baf7a", "#eda100", "#e34948", "#4a3aa7", "#008300", "#e87ba4")
+MARKERS = ("o", "s", "^", "D", "v", "P", "X")
+
+Sample = dict[str, Any]
+
+
+def fmt_seconds(seconds: float) -> str:
+    if isnan(seconds):
+        return "-"
+    if seconds < 1e-3:
+        return f"{seconds * 1e6:.0f} us"
+    if seconds < 1.0:
+        return f"{seconds * 1e3:.1f} ms"
+    return f"{seconds:.2f} s"
+
+
+def series_means(
+    samples: list[Sample],
+    sizes: list[int],
+    family: str,
+    name: str,
+    key: str,
+    expected: int,
+) -> list[float]:
+    """Mean of `key[name]` per size; NaN where fewer than `expected` samples."""
+    means = []
+    for size in sizes:
+        values = [
+            s[key][name]
+            for s in samples
+            if s["family"] == family and s["size"] == size and name in s.get(key, {})
+        ]
+        means.append(mean(values) if len(values) >= expected else nan)
+    return means
+
+
+def two_panel_figure() -> "tuple[Figure, Axes, Axes]":
+    """A tall time panel over a short quality panel sharing the x axis."""
+    fig, (top, bottom) = plt.subplots(
+        2,
+        1,
+        sharex=True,
+        figsize=(8, 6.4),
+        height_ratios=(3, 1.4),
+        layout="constrained",
+    )
+    fig.set_facecolor(SURFACE)
+    return fig, top, bottom
+
+
+def style_axes(ax: "Axes") -> None:
+    ax.set_facecolor(SURFACE)
+    ax.grid(visible=True, which="major", color=GRID, linewidth=0.8)
+    ax.set_axisbelow(True)
+    ax.tick_params(colors=INK_MUTED, labelcolor=INK_SECONDARY, labelsize=9)
+    for side in ("top", "right"):
+        ax.spines[side].set_visible(False)
+    for side in ("bottom", "left"):
+        ax.spines[side].set_color(AXIS)
+
+
+def log_time_axes(ax: "Axes", xs: list[int], base: int = 10) -> None:
+    style_axes(ax)
+    ax.set_yscale("log")
+    ax.set_xscale("log", base=base)
+    ax.set_xticks(xs, [f"{x:,}" for x in xs])
+    ax.tick_params(which="minor", bottom=False)
+    ax.set_ylabel("wall time [s]", color=INK_SECONDARY, fontsize=10)
+
+
+def plot_series(
+    ax: "Axes",
+    xs: list[int],
+    values: list[float],
+    label: str,
+    color: str,
+    marker: str = "o",
+    linestyle: str = "-",
+) -> None:
+    # An all-NaN series leaves a log axis empty and crashes on save
+    if all(isnan(value) for value in values):
+        return
+    ax.plot(
+        xs,
+        values,
+        label=label,
+        color=color,
+        linestyle=linestyle,
+        linewidth=2,
+        marker=marker,
+        markersize=5.5,
+        markeredgecolor=SURFACE,
+        markeredgewidth=1,
+    )
+
+
+def titles(ax: "Axes", title: str, caption: str) -> None:
+    ax.set_title(title, loc="left", color=INK, fontsize=12, fontweight="medium", pad=22)
+    note = ax.text(
+        0.0, 1.04, caption, transform=ax.transAxes, fontsize=8.5, color=INK_MUTED
+    )
+    note.set_in_layout(False)