[Code scan] Parse extxyz lattice per frame for PBC bond detection

[njzjz]

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

The extxyz detector parses Lattice= only once in _readNfunc() and stores it on self.cell. _readstepfunc() then reuses self.cell for every frame instead of parsing the comment line for the current frame.

Relevant code:

reacnetgenerator/reacnetgenerator/_detect.py

Lines 653 to 656 in 4fa8e2b

	if "Lattice=" in line:
	lattice_str = line.split("Lattice=")[1].split('"')[1]
	lattice_floats = list(map(float, lattice_str.split()))
	self.cell = np.array(lattice_floats).reshape((3, 3))

reacnetgenerator/reacnetgenerator/_detect.py

Lines 671 to 686 in 4fa8e2b

	def _readstepfunc(self, item) -> tuple[list[bytes], tuple[int, int]]:
	step, lines = item
	step_atoms = []
	timestep = step, step # Use step as timestep fallback
	boxsize = self.cell
	if self.pbc and boxsize is None:
	raise RuntimeError("No cell information is given in extxyz.")
	for index, line in enumerate(lines):
	if index > 1:
	s = line.split()
	step_atoms.append(
	(index - 1, Atom(s[0], tuple(float(x) for x in s[1:4])))
	)
	_, step_atoms = zip(*sorted(step_atoms, key=operator.itemgetter(0)))
	step_atoms = Atoms(step_atoms)
	bond, level = self._getbondfromcrd(step_atoms, boxsize)

Impact:

• variable-cell extxyz trajectories, common for NPT/PBC data, use the first frame's cell for all bond detection
• bond perception can become wrong when the box changes over time
• this can alter molecule identity and downstream reaction paths without an explicit error

Suggested fix:

Parse the extxyz metadata/comment line inside _readstepfunc() for each frame and pass that frame-specific lattice to _getbondfromcrd(). Keep the first-frame parsing in _readNfunc() only for atom count/type discovery if needed. Add a regression test with two frames that have different Lattice= values and assert the per-frame cell is used.