[Code scan] HybridDriver mutates caller-provided driver config dictionaries

[njzjz]

This issue is part of a Codex global repository code scan.

HybridDriver.__init__() deletes the "type" key from each caller-provided config dictionary. This mutates user input and makes the same config object unusable for a later construction.

Affected code:

dpdata/dpdata/driver.py

Lines 137 to 146 in a7a50bf

	def __init__(self, drivers: list[dict | Driver]) -> None:
	self.drivers = []
	for driver in drivers:
	if isinstance(driver, Driver):
	self.drivers.append(driver)
	elif isinstance(driver, dict):
	type = driver["type"]
	del driver["type"]
	self.drivers.append(Driver.get_driver(type)(**driver))
	else:

Minimal reproducer:

from dpdata.driver import Driver, HybridDriver

class DummyDriver(Driver):
    def label(self, data):
        return data

Driver.register("scan_dummy")(DummyDriver)

config = [{"type": "scan_dummy"}]
print("before", config)
HybridDriver(config)
print("after", config)
HybridDriver(config)

Current behavior:

before [{'type': 'scan_dummy'}]
after [{}]
KeyError: 'type'

HybridDriver should copy each config dict before popping or deleting keys.

[dosubot]

Confirmed — lines 141–142 do a bare del driver["type"] on the caller's dict [1]. A shallow copy before popping is sufficient here since the remaining kwargs are passed as keyword arguments:

elif isinstance(driver, dict):
    driver = driver.copy()
    type = driver.pop("type")
    self.drivers.append(Driver.get_driver(type)(**driver))

If any driver accepts nested mutable config values that it could also mutate, copy.deepcopy would be safer, but .copy() + .pop() matches the pattern used in similar fixes elsewhere in dpdata (e.g. PR #671 for mixed-type DeepMD data [2] and PR #471 for atom-type mutation during labeling [3]).

To reply, just mention @dosu.

---

Share context across your team and agents. Try Dosu.