Use Gurobi as the default solver

.github/workflows/ci.yml

@@ -22,4 +22,6 @@ jobs:
           python -m pip install --upgrade pip
           pip install -e ".[dev]"
       - name: Test
+        env:
+          OVERFLOW_SOLVER: glpk
         run: pytest -v --durations=10 -m "not integration"

README.md

@@ -18,9 +18,10 @@ kept in [`legacy_matlab/`](legacy_matlab/).
 pip install -e ".[dev]"
 ```
 
-Python 3.11 or later. A linear programming solver is required; the analysis
-scripts are developed against Gurobi, and the test suite runs on the GLPK
-solver that ships with cobrapy.
+Python 3.11 or later. Gurobi (`gurobipy` with a licence) is the default solver for
+the analysis and the tests. Another solver can be chosen with `--solver`, or with
+the `OVERFLOW_SOLVER` environment variable, which also applies to the tests; an
+unavailable solver is an error, not a silent switch.
 
 ## Layout
 
@@ -63,7 +64,7 @@ protein content and dilution rate on top.
 
 ```bash
 python -m overflow.build                 # every condition
-python -m overflow.build CN4 --solver glpk
+python -m overflow.build CN4 --solver glpk   # instead of the default, Gurobi
 ```
 
 Each condition gets `models/ecModel_P_<cond>.yml` and, in `results/`, the
@@ -166,5 +167,5 @@ pytest -m "not slow"         # skip the tests that load a genome-scale model
 pytest -m integration        # build a condition end to end; minutes
 ```
 
-The suite runs on GLPK by default, so it needs no commercial solver licence;
-set `OVERFLOW_TEST_SOLVER` to use another.
+The suite runs on the same solver as the analysis, Gurobi by default. Set
+`OVERFLOW_SOLVER=glpk` to run it without a licence, as CI does.

src/overflow/__init__.py

@@ -6,6 +6,9 @@
     Condition,
     load_conditions,
 )
+from overflow.solver import use_solver
+
+use_solver()
 
 __all__ = [
     "CONDITION_ORDER",
@@ -15,4 +18,5 @@
     "load_conditions",
     "load_gem",
     "load_model",
+    "use_solver",
 ]

src/overflow/analyze_usage.py

@@ -12,6 +12,7 @@
 import pandas as pd
 from geckopy import load_ec_model
 
+from overflow.solver import HELP, use_solver
 from overflow.adapter import build_adapter
 from overflow.config import (
     BIO_RXN,
@@ -71,13 +72,10 @@ def main(argv: Optional[list[str]] = None) -> int:
     parser.add_argument("--results-dir", type=Path, default=RESULTS_DIR)
     parser.add_argument("--suffix", default="_ribosome",
                         help="model file suffix, '' for the models without a ribosome")
-    parser.add_argument("--solver")
+    parser.add_argument("--solver", help=HELP)
     args = parser.parse_args(argv)
 
-    if args.solver:
-        import cobra
-
-        cobra.Configuration().solver = args.solver
+    use_solver(args.solver)
 
     tables = analyse(args.conditions, args.models_dir, args.suffix)
     usage = combine_usage(tables)

src/overflow/build.py

@@ -19,6 +19,7 @@
 import pandas as pd
 from geckopy import fill_enz_concs, save_ec_model, set_prot_pool_size
 
+from overflow.solver import HELP, use_solver
 from overflow.adapter import build_adapter, load_model
 from overflow.biomass import apply_protein_content
 from overflow.config import (
@@ -347,7 +348,7 @@ def main(argv: Optional[list[str]] = None) -> int:
     parser.add_argument("--ngam-steps", type=int, default=100)
     parser.add_argument(
         "--solver",
-        help="LP solver to use, e.g. gurobi or glpk (default: cobrapy's)",
+        help=HELP,
     )
     parser.add_argument("--no-complex-fix", action="store_true")
     parser.add_argument("--gam", choices=("model", "polymerization"), default="model")
@@ -383,10 +384,7 @@ def main(argv: Optional[list[str]] = None) -> int:
     )
     args = parser.parse_args(argv)
 
-    if args.solver:
-        import cobra
-
-        cobra.Configuration().solver = args.solver
+    use_solver(args.solver)
 
     conditions = load_conditions()
     table = read_proteomics()

src/overflow/build_ribosome.py

@@ -14,6 +14,7 @@
 import pandas as pd
 from geckopy import load_ec_model, save_ec_model
 
+from overflow.solver import HELP, use_solver
 from overflow.adapter import build_adapter
 from overflow.config import (
     BIO_RXN,
@@ -160,14 +161,11 @@ def main(argv: Optional[list[str]] = None) -> int:
     parser.add_argument("--results-dir", type=Path, default=RESULTS_DIR)
     parser.add_argument(
         "--solver",
-        help="LP solver to use, e.g. gurobi or glpk (default: cobrapy's)",
+        help=HELP,
     )
     args = parser.parse_args(argv)
 
-    if args.solver:
-        import cobra
-
-        cobra.Configuration().solver = args.solver
+    use_solver(args.solver)
 
     conditions = load_conditions()
     table = read_proteomics()

src/overflow/run_sampling.py

@@ -16,6 +16,7 @@
 
 import pandas as pd
 
+from overflow.solver import HELP, use_solver
 from overflow.adapter import load_gem
 from overflow.atp import selected_fluxes
 from overflow.biomass import GAM_NO_POLYMERIZATION, current_gam
@@ -172,23 +173,21 @@ def main(argv: Optional[list[str]] = None) -> int:
         "--keep-loops", action="store_true",
         help="do not tighten reactions to their loop-free range first",
     )
-    parser.add_argument("--solver")
+    parser.add_argument("--solver", help=HELP)
     args = parser.parse_args(argv)
 
     if args.procs:
         use_fork_start_method()
 
-    if args.solver or args.procs:
+    use_solver(args.solver)
+    if args.procs:
         import cobra
 
-        if args.solver:
-            cobra.Configuration().solver = args.solver
-        if args.procs:
-            # The loop-free screening is a flux variability analysis, which
-            # reads its process count from here rather than from the
-            # sampler's argument. Left unset it runs on one core and
-            # dominates the run.
-            cobra.Configuration().processes = args.procs
+        # The loop-free screening is a flux variability analysis, which
+        # reads its process count from here rather than from the
+        # sampler's argument. Left unset it runs on one core and
+        # dominates the run.
+        cobra.Configuration().processes = args.procs
 
     results = run(
         args.conditions, n_samples=args.samples, seed=args.seed,

src/overflow/solver.py

@@ -0,0 +1,26 @@
+"""The LP solver the analysis runs on: Gurobi unless told otherwise."""
+import os
+from typing import Optional
+
+import cobra
+
+DEFAULT_SOLVER = "gurobi"
+HELP = "LP solver to use, e.g. gurobi or glpk (default: gurobi, or $OVERFLOW_SOLVER)"
+
+
+def use_solver(name: Optional[str] = None) -> str:
+    """Make ``name`` cobra's solver, and return it.
+
+    Without a name it is ``$OVERFLOW_SOLVER``, else Gurobi. Asking for a solver that
+    is not available is an error rather than a quiet switch to another one.
+    """
+    name = name or os.environ.get("OVERFLOW_SOLVER") or DEFAULT_SOLVER
+    try:
+        cobra.Configuration().solver = name
+    except Exception as error:
+        raise RuntimeError(
+            f"cannot use the {name} solver ({error}). Gurobi is the default: install "
+            "gurobipy with a licence, or choose another solver with --solver or "
+            "OVERFLOW_SOLVER."
+        ) from error
+    return name

tests/conftest.py

@@ -1,13 +1,7 @@
-import os
-
-import cobra
 import pytest
 
-# Tests run on the open solver by default: they are small enough not to
-# need a commercial one, and a licence that is busy or absent should not
-# turn into a wall of failures that look like code.
-cobra.Configuration().solver = os.environ.get("OVERFLOW_TEST_SOLVER", "glpk")
-
+# Importing overflow selects the solver: Gurobi unless OVERFLOW_SOLVER names
+# another (CI has no licence and sets it to glpk).
 from overflow import build_adapter, load_conditions, load_gem, load_model
 
 

tests/test_solver.py

@@ -0,0 +1,47 @@
+import pytest
+
+import overflow.solver as solver
+
+
+class FakeConfiguration:
+    requested = []
+
+    @property
+    def solver(self):
+        return self.requested[-1] if self.requested else None
+
+    @solver.setter
+    def solver(self, name):
+        if name == "missing":
+            raise ValueError("Solver not available")
+        self.requested.append(name)
+
+
+@pytest.fixture
+def fake(monkeypatch):
+    FakeConfiguration.requested = []
+    monkeypatch.setattr(solver.cobra, "Configuration", FakeConfiguration)
+    monkeypatch.delenv("OVERFLOW_SOLVER", raising=False)
+    return FakeConfiguration
+
+
+def test_gurobi_is_the_default(fake):
+    assert solver.DEFAULT_SOLVER == "gurobi"
+    assert solver.use_solver() == "gurobi"
+    assert fake.requested == ["gurobi"]
+
+
+def test_the_environment_overrides_the_default(fake, monkeypatch):
+    monkeypatch.setenv("OVERFLOW_SOLVER", "glpk")
+    assert solver.use_solver() == "glpk"
+
+
+def test_a_named_solver_beats_the_environment(fake, monkeypatch):
+    monkeypatch.setenv("OVERFLOW_SOLVER", "glpk")
+    assert solver.use_solver("gurobi") == "gurobi"
+
+
+def test_an_unavailable_solver_is_an_error_not_a_fallback(fake):
+    with pytest.raises(RuntimeError, match="cannot use the missing solver"):
+        solver.use_solver("missing")
+    assert fake.requested == []