refactor(benchmarks): decouple CI baseline from analytics + unify measurements() (split prep)

benchmarks/__init__.py

@@ -34,8 +34,11 @@
     from benchmarks.snapshot import Metric
 
 # Importing the models / patterns packages triggers each module's
-# ``register(...)`` / ``register_pattern(...)`` call at import time.
-from benchmarks import bench, models, patterns  # noqa: F401, E402
+# ``register(...)`` / ``register_pattern(...)`` call at import time. ``bench``
+# (analytics) is intentionally NOT eagerly imported — that keeps ``import
+# benchmarks`` / the CodSpeed CI baseline free of the local-only layer;
+# ``from benchmarks import bench`` still works (submodule import on demand).
+from benchmarks import models, patterns  # noqa: F401, E402
 
 
 def load_long_df(

benchmarks/cli/_base.py

@@ -44,7 +44,7 @@ class Measure(StrEnum):
 )
 
 PhaseName = Literal[
-    "build", "matrices", "to_lp", "to_netcdf", "from_netcdf", "to_solver"
+    "build", "matrices", "to_lp", "to_netcdf", "from_netcdf", "to_solver", "pipeline"
 ]
 SpecKind = Literal["all", "models", "patterns"]
 
@@ -56,6 +56,7 @@ class Measure(StrEnum):
     "to_netcdf": "benchmarks/test_netcdf.py::test_to_netcdf",
     "from_netcdf": "benchmarks/test_netcdf.py::test_from_netcdf",
     "to_solver": "benchmarks/test_to_solver.py",
+    "pipeline": "benchmarks/test_pipeline.py",
 }
 
 # pytest args that constitute a "smoke" run — quick sizes, no timings.

benchmarks/cli/run.py

@@ -163,6 +163,8 @@ def run(
     def _timing() -> None:
         args: list[str] = []
         args.append(_PHASE_TEST_FILE[phase] if phase is not None else "benchmarks/")
+        if phase == "pipeline":
+            args.append("--pipeline")
         if quick:
             args.append("--quick")
         elif long:

benchmarks/conftest.py

@@ -2,10 +2,17 @@
 
 from __future__ import annotations
 
+from typing import TYPE_CHECKING
+
 import pytest
 
 from benchmarks.registry import BenchSpec, skip_reason
 
+if TYPE_CHECKING:
+    from collections.abc import Callable
+
+    from benchmarks.phases import PhaseCase
+
 # Test modules the CodSpeed instruments measure (edit to change coverage).
 # build + the two export paths: to_lp (LP text) and to_solver (direct handoff,
 # which also exercises matrix-gen). matrices is dropped — a subset of to_solver;
@@ -55,6 +62,16 @@ def pytest_addoption(parser: pytest.Parser) -> None:
             "--quick/--long for patterns, leaving models on the prevailing tier."
         ),
     )
+    parser.addoption(
+        "--pipeline",
+        action="store_true",
+        default=False,
+        help=(
+            "Include the opt-in end-to-end pipeline benchmark (build → matrices "
+            "→ lp in one measured region). Off by default — it re-runs the "
+            "per-phase work and includes the build."
+        ),
+    )
 
 
 def pytest_collection_modifyitems(
@@ -63,13 +80,20 @@ def pytest_collection_modifyitems(
     """
     ``--quick`` drops the PyPSA end-to-end test (~30s; minutes under cachegrind).
     ``--codspeed`` narrows the run to ``CODSPEED_MODULES`` (drops netcdf/matrices).
+    ``test_pipeline`` (end-to-end) is opt-in — deselected unless ``--pipeline``.
     """
     if config.getoption("--quick"):
         skip = pytest.mark.skip(reason="--quick: pypsa end-to-end skipped")
         for item in items:
             if "test_pypsa_carbon_management" in item.nodeid:
                 item.add_marker(skip)
 
+    if not config.getoption("--pipeline"):
+        dropped = [i for i in items if i.path.stem == "test_pipeline"]
+        if dropped:
+            config.hook.pytest_deselected(items=dropped)
+            items[:] = [i for i in items if i.path.stem != "test_pipeline"]
+
     if getattr(config.option, "codspeed", False):
         deselected = [i for i in items if i.path.stem not in CODSPEED_MODULES]
         if deselected:
@@ -105,3 +129,22 @@ def maybe_skip(request: pytest.FixtureRequest, spec: BenchSpec, size: int) -> No
     )
     if reason:
         pytest.skip(reason)
+
+
+def run_case(
+    benchmark: Callable[..., object],
+    case: PhaseCase,
+    request: pytest.FixtureRequest,
+) -> None:
+    """
+    Shared pytest-benchmark driver body for one :class:`PhaseCase`.
+
+    Honours the case's own ``skip`` (e.g. solver not installed) and the size
+    tiers (via :func:`maybe_skip`), then runs the case's measured action under
+    ``benchmark`` inside the case's setup/teardown context.
+    """
+    if case.skip:
+        pytest.skip(case.skip)
+    maybe_skip(request, case.spec, case.value)
+    with case.run() as action:
+        benchmark(action)