Sync only the declaring module before a run; skip git-ignored paths

`fluksio run` walked the working directory and imported every module in it, which ran a research repository's analysis scripts before the run was submitted. A flow's origin now records the file and module name that declared it, and run/sweep import just that, under the same name so the generated node bodies do not change. A flow with no record still walks, now skipping what .gitignore covers. Every code-defined flow reports `flow` changed once on its next sync, since the origin gained two fields. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QgG6TJw5qBZzaLghQSRVrq
2026-09-16 16:55:02 +02:00
parent 7dfe646e2b
commit fccf89f849
9 changed files with 246 additions and 22 deletions
@@ -550,6 +550,85 @@ def test_a_study_per_directory_imports_under_its_own_name(tmp_path) -> None:
        sys.path[:] = [entry for entry in sys.path if entry != str(tmp_path / "dev")]


+def test_run_imports_the_module_that_declared_the_flow(tmp_path, monkeypatch):
+    """Not every script below the working directory: importing one runs it.
+
+    Imported under the name the last sync recorded, so the node bodies it
+    generates are the ones already stored. With nothing recorded, the walk.
+    """
+    from fluksio.sdk import cli as sdk_cli
+
+    study = tmp_path / "dev" / "s1" / "study.py"
+    study.parent.mkdir(parents=True)
+    study.write_text("")
+    monkeypatch.chdir(tmp_path)
+
+    imported: list[tuple] = []
+    walked: list[list[str]] = []
+    monkeypatch.setattr(sdk_cli, "_import", lambda *args: imported.append(args))
+    monkeypatch.setattr(
+        sdk_cli,
+        "discover",
+        lambda targets, keep_going=False: walked.append(targets) or [],
+    )
+
+    class Engine:
+        origin: dict = {}
+
+        def get_flow(self, name):
+            return {"definition": {"origin": self.origin}}
+
+    engine = Engine()
+    engine.origin = {"declared_in": "dev/s1/study.py", "module": "s1.study"}
+    sdk_cli._sync_first(engine, [], "study")
+    assert imported == [
+        (str(tmp_path / "dev"), "s1.study", tmp_path / "dev" / "s1" / "study.py")
+    ]
+    assert walked == []
+
+    # A name that does not match the file is not trusted, and neither is a
+    # file this checkout does not have: both walk, as an unrecorded flow does.
+    for origin in (
+        {"declared_in": "dev/s1/study.py", "module": "s2.study"},
+        {"declared_in": "dev/s9/study.py", "module": "s9.study"},
+        {},
+    ):
+        engine.origin = origin
+        sdk_cli._sync_first(engine, [], "study")
+    assert walked == [["."], ["."], ["."]]
+
+    # And `--sync` is taken as said.
+    sdk_cli._sync_first(engine, ["dev/s1"], "study")
+    assert walked[-1] == ["dev/s1"]
+
+
+def test_the_walk_skips_what_the_repository_ignores(tmp_path, monkeypatch) -> None:
+    """`mlruns/` is output, not a study, and importing it would run it."""
+    import shutil
+    import subprocess
+    import sys
+
+    import pytest
+
+    from fluksio.sdk.cli import discover
+
+    if shutil.which("git") is None:
+        pytest.skip("needs git")
+    subprocess.run(["git", "init", "-q", str(tmp_path)], check=True)
+    (tmp_path / ".gitignore").write_text("mlruns/\nscratch.py\n")
+    (tmp_path / "mlruns" / "0").mkdir(parents=True)
+    (tmp_path / "mlruns" / "0" / "boom.py").write_text("raise SystemExit('imported')\n")
+    (tmp_path / "scratch.py").write_text("raise SystemExit('imported')\n")
+    (tmp_path / "walked_study.py").write_text("VALUE = 1\n")
+
+    try:
+        discover([str(tmp_path)])
+        assert sys.modules["walked_study"].VALUE == 1
+    finally:
+        sys.modules.pop("walked_study", None)
+        sys.path[:] = [entry for entry in sys.path if entry != str(tmp_path)]
+
+
 def test_two_files_of_one_name_are_refused(tmp_path) -> None:
    """Python keeps one module per name, and a node's body imports by it.