Add adapter smoke CI

Author: OptHuang

.github/workflows/ci.yml

@@ -0,0 +1,47 @@
+name: CI
+
+on:
+  push:
+  pull_request:
+  schedule:
+    - cron: "23 2 * * *"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+
+    env:
+      OMP_NUM_THREADS: "2"
+      OPENBLAS_NUM_THREADS: "2"
+      MKL_NUM_THREADS: "2"
+      NUMEXPR_NUM_THREADS: "2"
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Checkout OptiProfiler
+        uses: actions/checkout@v4
+        with:
+          repository: optiprofiler/optiprofiler
+          path: optiprofiler
+          ref: main
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install -e optiprofiler
+
+      - name: Run adapter smoke and random tests
+        run: |
+          python -m unittest discover -s tests -p 'test_*.py'

README.md

@@ -28,8 +28,27 @@ set_plib_config('s2mpj', variable_size='all', test_feasibility_problems=2)
 
 You can also set the environment variables `S2MPJ_VARIABLE_SIZE` and `S2MPJ_TEST_FEASIBILITY_PROBLEMS` directly. Environment variables take precedence over `config.txt`.
 
+## Testing
+
+The `CI` workflow runs daily and on pushes. It checks the OptiProfiler adapter layer by:
+
+- selecting a small set of representative `u`, `b`, `l`, and `n` problems;
+- loading each selected problem through `s2mpj_load`;
+- evaluating `fun`, `cub`, and `ceq` at the initial point;
+- sampling a few additional small problems each day with at most two numerical-library threads.
+
+Locally, from this repository:
+
+```bash
+python -m unittest discover -s tests -p 'test_*.py'
+```
+
 ## Maintenance
 
 This repository is **automatically synchronized** with the upstream `GrattonToint/S2MPJ` repository via GitHub Actions. It checks for updates daily to ensure the problem set remains current.
 
+## Provenance and Citation
+
+The files under `src/` are a filtered Python subset of the upstream [S2MPJ](https://github.com/GrattonToint/S2MPJ) repository. This repository adds only the OptiProfiler adapter, metadata, and maintenance workflows. Please follow the upstream S2MPJ citation and license guidance when using the problem collection.
+
 For the full collection or other languages, please visit the [original repository](https://github.com/GrattonToint/S2MPJ).

s2mpj_tools.py

@@ -176,20 +176,24 @@ def s2mpj_load(problem_name, *args):
 
     # The linear constraints are hidden in the cJx method output.
     # cx = jx @ x0 - bx
-    buf = io.StringIO()
-    with redirect_stdout(buf):
-        try:
-            cx, jx = p.cJx(p.x0)[:2]
-            if hasattr(cx, 'toarray'):
-                cx = cx.toarray()
-            if hasattr(jx, 'toarray'):
-                jx = jx.toarray()
-            cx = cx.flatten()
-            bx = jx @ x0 - cx
-        except Exception as err:
-            _warn_s2mpj_evaluation_failure(name, 'the linear constraint Jacobian', err)
-            jx = np.full((getattr(p, 'm', 0), getattr(p, 'n', x0.size)), np.nan)
-            bx = np.full(getattr(p, 'm', 0), np.nan)
+    if getattr(p, 'm', 0) == 0:
+        jx = np.empty((0, getattr(p, 'n', x0.size)))
+        bx = np.empty(0)
+    else:
+        buf = io.StringIO()
+        with redirect_stdout(buf):
+            try:
+                cx, jx = p.cJx(p.x0)[:2]
+                if hasattr(cx, 'toarray'):
+                    cx = cx.toarray()
+                if hasattr(jx, 'toarray'):
+                    jx = jx.toarray()
+                cx = cx.flatten()
+                bx = jx @ x0 - cx
+            except Exception as err:
+                _warn_s2mpj_evaluation_failure(name, 'the linear constraint Jacobian', err)
+                jx = np.full((getattr(p, 'm', 0), getattr(p, 'n', x0.size)), np.nan)
+                bx = np.full(getattr(p, 'm', 0), np.nan)
 
     nonlincons = np.setdiff1d(np.arange(p.m), p.lincons)
     idx_eq = np.intersect1d(np.arange(p.nle, p.nle + p.neq), idx_cl_finite)
@@ -607,6 +611,8 @@ def _gethess(p, is_feasibility, problem_name, x):
     return h
 
 def _getcx(p, problem_name, x):
+    if getattr(p, 'm', 0) == 0:
+        return np.empty(0)
     buf = io.StringIO()
     with redirect_stdout(buf):
         try:
@@ -619,6 +625,8 @@ def _getcx(p, problem_name, x):
     return c
 
 def _getJx(p, problem_name, x):
+    if getattr(p, 'm', 0) == 0:
+        return np.empty((0, getattr(p, 'n', x.size)))
     buf = io.StringIO()
     with redirect_stdout(buf):
         try:
@@ -630,6 +638,8 @@ def _getJx(p, problem_name, x):
     return j
 
 def _getHx(p, problem_name, x):
+    if getattr(p, 'm', 0) == 0:
+        return []
     buf = io.StringIO()
     with redirect_stdout(buf):
         try:

tests/test_s2mpj_python.py

@@ -0,0 +1,101 @@
+from __future__ import annotations
+
+from datetime import date
+from pathlib import Path
+import math
+import os
+import random
+import sys
+import unittest
+
+import numpy as np
+
+
+REPO_DIR = Path(__file__).resolve().parents[1]
+
+op_candidates = [
+    REPO_DIR / "optiprofiler" / "python",
+    REPO_DIR.parents[1] / "optiprofiler" / "python",
+]
+for op_path in op_candidates:
+    if (op_path / "optiprofiler").is_dir():
+        sys.path.insert(0, str(op_path))
+        break
+
+sys.path.insert(0, str(REPO_DIR))
+
+from s2mpj_tools import s2mpj_load, s2mpj_select
+
+
+REPRESENTATIVE_PROBLEMS = ["ALLINITU", "ALLINIT", "ALSOTAME", "ALLINITA"]
+
+
+def _as_array(value):
+    if value is None:
+        return np.empty(0)
+    return np.asarray(value)
+
+
+def _assert_problem_contract(testcase, problem_name):
+    problem = s2mpj_load(problem_name)
+    testcase.assertEqual(problem.name.split("_")[0], problem_name.split("_")[0])
+    testcase.assertGreaterEqual(problem.n, 1)
+    testcase.assertEqual(problem.x0.size, problem.n)
+
+    fx0 = problem.fun(problem.x0)
+    testcase.assertTrue(math.isfinite(float(fx0)) or math.isnan(float(fx0)))
+
+    cub0 = _as_array(problem.cub(problem.x0))
+    ceq0 = _as_array(problem.ceq(problem.x0))
+    testcase.assertEqual(cub0.ndim, 1)
+    testcase.assertEqual(ceq0.ndim, 1)
+
+    # Evaluate twice to catch wrappers that mutate S2MPJ state unexpectedly.
+    fx1 = problem.fun(problem.x0)
+    testcase.assertTrue(math.isfinite(float(fx1)) or math.isnan(float(fx1)))
+
+
+class S2MPJPythonAdapterTests(unittest.TestCase):
+    def test_select_representative_problem_types(self):
+        selected = s2mpj_select(
+            {
+                "ptype": "ubln",
+                "maxdim": 5,
+                "maxb": 20,
+                "maxlcon": 20,
+                "maxnlcon": 20,
+                "maxcon": 20,
+            }
+        )
+        for problem_name in REPRESENTATIVE_PROBLEMS:
+            self.assertIn(problem_name, selected)
+
+    def test_load_representative_problem_types(self):
+        for problem_name in REPRESENTATIVE_PROBLEMS:
+            with self.subTest(problem=problem_name):
+                _assert_problem_contract(self, problem_name)
+
+    def test_daily_random_small_problem_sample(self):
+        seed = int(os.environ.get("OP_RANDOM_SEED", date.today().strftime("%Y%m%d")))
+        candidates = s2mpj_select(
+            {
+                "ptype": "ubln",
+                "maxdim": 5,
+                "maxb": 20,
+                "maxlcon": 20,
+                "maxnlcon": 20,
+                "maxcon": 20,
+            }
+        )
+        self.assertGreaterEqual(len(candidates), 4)
+
+        rng = random.Random(seed)
+        sample = rng.sample(candidates, k=min(4, len(candidates)))
+        print(f"S2MPJ Python random sample seed={seed}: {sample}")
+        for problem_name in sample:
+            with self.subTest(problem=problem_name):
+                _assert_problem_contract(self, problem_name)
+
+
+if __name__ == "__main__":
+    unittest.main()