[Code scan] Make ASE custom cutoff tests fail when pair cutoffs are ignored

[njzjz]

This issue is a Codex global repository scan finding for deepmodeling/reacnetgenerator at commit 4fa8e2b.

The ASE custom-cutoff tests currently do not fail if pair-specific custom cutoffs are ignored.

Relevant code:

reacnetgenerator/reacnetgenerator/_detect.py

Lines 276 to 288 in 4fa8e2b

	"inputfilename",
	"atomname",
	"stepinterval",
	"nproc",
	"pbc",
	"cell",
	"use_ase",
	"ase_cutoff_mult",
	"custom_cutoffs",
	],
	["N", "atomtype", "step", "timestep", "temp1it", "moleculetempfilename"],
	)
	self._parsed_custom_cutoffs = self._parse_custom_cutoffs(self.custom_cutoffs)

reacnetgenerator/tests/test_detect.py

Lines 198 to 213 in 4fa8e2b

	def test_custom_cutoffs_override(self, detect_instance):
	"""Test that custom cutoffs override global settings."""
	# Modify the detect instance to have custom cutoffs
	detect_instance.rng.custom_cutoffs = "H-O:1.0" # Very short cutoff
	# Create a simple water molecule where H-O distance is ~1.0 Angstrom
	atoms = Atoms(
	"H2O", positions=[[0.5, 0.0, 0.0], [-0.5, 0.0, 0.0], [0.0, 0.0, 0.0]]
	) # H-O distance is 0.5
	cell = np.eye(3) * 10
	bond, _bondlevel = detect_instance._getbondfromase(atoms, cell)
	# With a 1.0 cutoff, H-O should still be bonded
	# Check that oxygen is bonded to at least one hydrogen
	assert 0 in bond[2] or 1 in bond[2] # O (index 2) bonded to H (index 0 or 1)

reacnetgenerator/tests/test_ase.py

Lines 182 to 216 in 4fa8e2b

	def test_ase_with_custom_cutoffs_integration(self):
	"""Integration test for ASE mode with custom cutoffs."""
	# Create a temporary file with a simple water molecule
	water_file = "test_water.dump"
	with open(water_file, "w") as f:
	f.write("""ITEM: TIMESTEP
	0
	ITEM: NUMBER OF ATOMS
	3
	ITEM: BOX BOUNDS pp pp pp
	0.0 10.0
	0.0 10.0
	0.0 10.0
	ITEM: ATOMS id type x y z
	1 1 0.757 0.586 0.0
	2 1 -0.757 0.586 0.0
	3 2 0.0 0.0 0.0
	""")
	rng = ReacNetGenerator(
	inputfiletype="lammpsdumpfile",
	inputfilename=water_file,
	atomname=["H", "O"],
	pbc=True,
	use_ase=True,
	custom_cutoffs="H-O:1.5", # Custom cutoff for H-O bonds
	)
	# Run the detection process
	detect_class = _DetectLAMMPSdump.gettype(rng)
	detect_class.detect()
	# Should have processed the file
	assert hasattr(detect_class, "temp1it")
	assert detect_class.temp1it > 0

Why the current tests are weak:

• _DetectCrd.__init__() parses self.custom_cutoffs into self._parsed_custom_cutoffs once at construction time.
• test_custom_cutoffs_override mutates detect_instance.rng.custom_cutoffs after the detector was already constructed, so it does not update self._parsed_custom_cutoffs.
• The geometry in that test has 0.5 A H-O distances, which are expected to bond under the default natural cutoff anyway.
• The integration test only checks that the detection ran and produced temp1it > 0, not that the custom cutoff changed the resulting bonds/molecules.

Impact:

A regression that completely ignores --ase-pair-cutoffs can still pass the current tests.

Suggested fix:

Build the detector with custom cutoffs before initialization, and use a geometry where default cutoffs and custom cutoffs produce different bond graphs. Assert the actual bond lists or molecule count, not just that detection completed.