Download data/tests/test_libgrid.py from introvoyz042/haddock3: direct link, hf CLI and curl.
- Browser
- Download file 4.5 kB
-
https://huggingface.co/datasets/introvoyz042/haddock3/resolve/main/data/tests/test_libgrid.py
- Command line
-
hf download hf://datasets/introvoyz042/haddock3/data/tests/test_libgrid.py
-
curl -L -o test_libgrid.py https://huggingface.co/datasets/introvoyz042/haddock3/resolve/main/data/tests/test_libgrid.py
4.5 kB
| import pytest | |
| from haddock.libs.libgrid import ( | |
| GridJob, | |
| CompositeGridJob, | |
| JobStatus, | |
| Tag, | |
| ping_dirac, | |
| validate_dirac, | |
| ) | |
| def dummy_paths(tmp_path): | |
| toppar = tmp_path / "toppar" | |
| toppar.mkdir() | |
| module = tmp_path / "module" | |
| module.mkdir() | |
| return toppar, module | |
| def test_gridjob_create_jdl_and_job_script(dummy_paths): | |
| toppar, module = dummy_paths | |
| job = GridJob(input="input", toppar_path=toppar, module_path=module) | |
| job.name = "testjob" | |
| job.expected_outputs = ["foo.out"] | |
| job.jdl = job.loc / "job.jdl" | |
| job.job_script = job.loc / "job.sh" | |
| job.create_jdl() | |
| job.create_job_script() | |
| assert job.jdl.exists() | |
| assert job.job_script.exists() | |
| content = job.jdl.read_text() | |
| assert f'JobName = "{job.name}";' in content | |
| assert '"foo.out"' in content | |
| def test_gridjob_process_input_f(tmp_path, dummy_paths): | |
| toppar, module = dummy_paths | |
| input_str = '$output_test = "result.txt";\n' | |
| job = GridJob(input=input_str, toppar_path=toppar, module_path=module) | |
| job.name = "testjob" | |
| job.loc = tmp_path | |
| job.input_str = input_str | |
| job.process_input_f() | |
| assert "result.txt" in job.expected_outputs | |
| inp_file = tmp_path / f"{job.name}.inp" | |
| assert inp_file.exists() | |
| def test_compositegridjob_parse_input_and_process(dummy_paths): | |
| toppar, module = dummy_paths | |
| input_list = ["line1\nstop\n", "line2\nstop\n"] | |
| job = CompositeGridJob(input=input_list, toppar_path=toppar, module_path=module) | |
| assert len(job.input_str_list) == 2 | |
| job.process_input_f() | |
| for idx in range(2): | |
| inp_file = job.loc / f"{idx}_{job.name}.inp" | |
| assert inp_file.exists() | |
| def test_gridjob_repr(dummy_paths): | |
| toppar, module = dummy_paths | |
| job = GridJob(input="input", toppar_path=toppar, module_path=module) | |
| s = repr(job) | |
| assert "ID:" in s and "Name:" in s and "Status:" in s | |
| def test_jobstatus_from_string(): | |
| assert JobStatus.from_string("Running") == JobStatus.RUNNING | |
| assert JobStatus.from_string("unknown") == JobStatus.UNKNOWN | |
| def test_tag_enum(): | |
| assert Tag.PROBING.value == "Probing" | |
| assert Tag.DEFAULT.value == "Default" | |
| def test_parse_output(): | |
| s = "JobID=123 Status=Running Site=TestSite;" | |
| d = GridJob.parse_output(s) | |
| assert d["JobID"] == "123" | |
| assert d["Status"] == "Running" | |
| assert d["Site"] == "TestSite" | |
| def test_process_line_and_find_output(): | |
| line = '$output_test = "foo.txt";' | |
| new_line, found = GridJob._process_line(line) | |
| assert new_line == line | |
| output = GridJob._find_output(line) | |
| assert output == "foo.txt" | |
| def test_clean_timings(dummy_paths): | |
| toppar, module = dummy_paths | |
| job = GridJob(input="input", toppar_path=toppar, module_path=module) | |
| job.timings = {"foo": 123} | |
| job.clean_timings() | |
| assert job.timings == {} | |
| def test_clean_removes_tmpdir(dummy_paths): | |
| toppar, module = dummy_paths | |
| job = GridJob(input="input", toppar_path=toppar, module_path=module) | |
| loc = job.loc | |
| assert loc.exists() | |
| job.clean() | |
| assert not loc.exists() | |
| def test_ping_and_validate_dirac(monkeypatch): | |
| monkeypatch.setattr("shutil.which", lambda cmd: "/usr/bin/" + cmd) | |
| monkeypatch.setattr( | |
| "subprocess.run", | |
| lambda *a, **k: type( | |
| "R", (), {"returncode": 0, "stderr": b"", "stdout": b""} | |
| )(), | |
| ) | |
| assert validate_dirac() | |
| assert ping_dirac() | |
| def test_process_line(tmp_path, monkeypatch): | |
| # No match | |
| line = "no pattern here" | |
| new_line, found = GridJob._process_line(line) | |
| assert new_line == line | |
| assert found is None | |
| # AT_PATTERN match and file exists | |
| test_file = tmp_path / "foo.txt" | |
| test_file.write_text("data") | |
| line = f"@@{test_file}" | |
| monkeypatch.chdir(tmp_path) | |
| new_line, found = GridJob._process_line(line) | |
| assert new_line == line.replace(str(test_file), "foo.txt") | |
| assert found == str(test_file) | |
| # VAR_PATTERN match and file does not exist | |
| line = '($ test="bar.txt")' | |
| new_line, found = GridJob._process_line(line) | |
| assert new_line == line | |
| assert found is None | |
| def test_find_output(): | |
| line = '$output_test = "foo.txt";' | |
| output = GridJob._find_output(line) | |
| assert output == "foo.txt" | |
| line2 = "$output_test = bar.txt;" | |
| output2 = GridJob._find_output(line2) | |
| assert output2 == "bar.txt" | |
| line3 = "no output here" | |
| output3 = GridJob._find_output(line3) | |
| assert output3 is None | |