Wrap cclib parser crashes into a clean TSValueError

.gitignore

@@ -57,6 +57,8 @@ cover/
 
 # Django stuff:
 *.log
+# ... except real QM-program logfiles kept as test fixtures.
+!tests/data/**/*.log
 local_settings.py
 db.sqlite3
 db.sqlite3-journal

ThermoScreening/calculator/qm.py

@@ -104,12 +104,23 @@ def read_cclib(path):
     Raises
     ------
     TSValueError
-        If cclib is not installed, the file(s) cannot be parsed, or there are no
-        vibrational frequencies.
+        If cclib is not installed, the file(s) cannot be parsed (including a
+        parser crash on an unusual output format, wrapped from cclib), or there
+        are no vibrational frequencies.
     """
     cclib = _import_cclib()
 
-    data = cclib.io.ccread(_normalize_source(path))
+    try:
+        data = cclib.io.ccread(_normalize_source(path))
+    except Exception as exc:
+        # cclib's parsers are format/version-sensitive and can raise on an
+        # output format they don't fully handle (seen on real Gaussian 16 logs
+        # with cclib 1.8.1); surface that as our own exception type instead of
+        # letting an arbitrary cclib-internal exception escape uncontrolled.
+        raise TSValueError(
+            f"cclib raised {type(exc).__name__} while parsing "
+            f"'{_describe_source(path)}': {exc}"
+        ) from exc
     if data is None:
         raise TSValueError(f"cclib could not parse '{_describe_source(path)}' as a QM output.")
     if getattr(data, "vibfreqs", None) is None or not len(data.vibfreqs):

tests/calculator/test_qm.py

@@ -25,6 +25,10 @@
     if p.suffix != ".md"
 )
 
+_GAUSSIAN_DATA_DIR = Path(__file__).resolve().parents[1] / "data" / "calculator" / "gaussian"
+_REAL_GAUSSIAN_WATER = str(_GAUSSIAN_DATA_DIR / "water_neutral_opt_freq.out")
+_REAL_GAUSSIAN_UNPARSEABLE = str(_GAUSSIAN_DATA_DIR / "mp2_avdz_freq_tight.log")
+
 # water: geometry (Angstrom), three real modes, SCF energy in eV (~ -76.4 Ha)
 _WATER = dict(
     atomnos=np.array([8, 1, 1]),
@@ -187,3 +191,46 @@ def test_cclib_thermo_real_turbomole_output():
     assert thermo.electronic_energy() == pytest.approx(-951.0820621320888)
     assert math.isfinite(thermo.total_EeGtot())
     assert thermo.total_entropy("cal/(mol*K)") > 0
+
+
+# --- real Gaussian 16 output --- #
+#
+# See tests/data/calculator/gaussian/README.md for provenance.
+
+
+def test_read_cclib_real_gaussian_output():
+    atoms, freqs, energy = read_cclib(_REAL_GAUSSIAN_WATER)
+
+    assert list(atoms.get_chemical_symbols()) == ["O", "H", "H"]
+    assert list(freqs) == pytest.approx([2169.7613, 4141.3837, 4392.5759])
+    assert energy == pytest.approx(-74.96589788571758)
+
+
+def test_cclib_thermo_real_gaussian_output():
+    thermo = cclib_thermo(_REAL_GAUSSIAN_WATER)
+
+    assert thermo.electronic_energy() == pytest.approx(-74.96589788571758)
+    # gas-phase water's experimental standard entropy is ~45 cal/(mol K)
+    assert thermo.total_entropy("cal/(mol*K)") == pytest.approx(45.0, abs=2.0)
+    assert math.isfinite(thermo.total_EeGtot())
+
+
+def test_read_cclib_wraps_a_real_cclib_parser_crash():
+    # this real Gaussian 16 file's "Leave Link" timing line crashes cclib
+    # 1.8.1's own Gaussian parser with an unhandled ValueError (confirmed:
+    # still the latest cclib release at the time this test was written) --
+    # that must not leak out of read_cclib as a raw, arbitrary exception
+    with pytest.raises(TSValueError, match="cclib raised"):
+        read_cclib(_REAL_GAUSSIAN_UNPARSEABLE)
+
+
+def test_read_cclib_wraps_any_ccread_exception(monkeypatch):
+    # a cclib-version-independent guard for the same behaviour: whatever
+    # cclib.io.ccread raises internally must come out as a clean TSValueError,
+    # not the original exception type
+    def fake_ccread(source):
+        raise ValueError("some internal cclib parsing failure")
+
+    monkeypatch.setattr(cclib.io, "ccread", fake_ccread)
+    with pytest.raises(TSValueError, match="cclib raised ValueError.*some internal"):
+        read_cclib("whatever.log")

tests/data/calculator/gaussian/README.md

@@ -0,0 +1,18 @@
+Real Gaussian 16 log files, used to test `read_cclib`/`cclib_thermo` against
+genuine (non-mocked) Gaussian output.
+
+- `water_neutral_opt_freq.out` — a clean neutral water optimization + frequency
+  job. Parses successfully; used as the real-data success-path test.
+- `mp2_avdz_freq_tight.log` — an MP2/aug-cc-pVDZ frequency job whose "Leave
+  Link" timing line (`MaxMem=... cpu: ... elap: ...`) crashes cclib 1.8.1's
+  Gaussian parser (confirmed: still the latest cclib release at the time).
+  Root cause is an unhandled `ValueError` inside cclib's own parser; what
+  actually reaches `read_cclib` depends on the surrounding logging setup (a
+  bare `cclib.io.ccread` call raises that `ValueError` directly, but cclib's
+  own `logger.error()` call right before it can itself raise first under some
+  logging configurations -- either way, `read_cclib` wraps whatever comes out
+  into a clean `TSValueError`). Used as the regression test for that
+  wrapping, so no cclib parsing exception escapes uncontrolled.
+
+Source: [cclib/cclib-data](https://github.com/cclib/cclib-data), the cclib
+project's own public regression-test data, `Gaussian/Gaussian16/`.