copy.deepcopy() copies a structure by sending every element back through deepcopy(). For elements that need no copying at all — strings, ints, None, booleans, floats and the other immutable atomic types — that round trip still costs a function call each, even though the value handed back is the same object.
Real data is dominated by these atomic leaves. A parsed JSON document, a settings dict cloned before mutation, a record copied inside a framework: the keys are strings and most values are strings and numbers. Copying such a structure spends most of its time calling deepcopy() only to receive the same object straight back.
Deep-copying 105 JSON documents drawn from the top-1000 PyPI projects takes 1.20 ms today. Folding the atomic-type check into the dict, list and tuple copiers brings that to 970 µs, 23% faster, with identical handling of shared references, recursive structures and int/tuple subclasses.
Linked PRs
copy.deepcopy()copies a structure by sending every element back throughdeepcopy(). For elements that need no copying at all — strings, ints,None, booleans, floats and the other immutable atomic types — that round trip still costs a function call each, even though the value handed back is the same object.Real data is dominated by these atomic leaves. A parsed JSON document, a settings dict cloned before mutation, a record copied inside a framework: the keys are strings and most values are strings and numbers. Copying such a structure spends most of its time calling
deepcopy()only to receive the same object straight back.Deep-copying 105 JSON documents drawn from the top-1000 PyPI projects takes 1.20 ms today. Folding the atomic-type check into the dict, list and tuple copiers brings that to 970 µs, 23% faster, with identical handling of shared references, recursive structures and int/tuple subclasses.
Linked PRs