Feature description
It would be great if ownership semantics can be communicated in the C++ source code via annotations, so we don't have to write Pythonizations by hand just for this, and carefully keep them in sync with the behavior of the C++ code.
This is relevant for both arguments and return values.
For arguments, we need something to indicate ownership adoption (_transfer_ownership is just a placeholder name to illustrate the idea):
void foo(TObject * _transfer_ownership obj) {
std::unique_ptr<TObject>{obj};
}
For return values, we need something to indicate ownership on the caller side (_caller_owns is again just a placeholder):
TObject *_caller_owns bar() {
return new TObject{};
}
Object ownership problems come up often on the forum (example post), so having a robust way to convey the correct ownership semantics would be a game changer for the robustness of the ROOT Python interfaces.
Alternatives considered
- Manual Pythonizations like in ee79dfd, but this doesn't scale well.
- Migrating C++ interfaces to smart pointers, but this would break backwards compatibility.
Feature description
It would be great if ownership semantics can be communicated in the C++ source code via annotations, so we don't have to write Pythonizations by hand just for this, and carefully keep them in sync with the behavior of the C++ code.
This is relevant for both arguments and return values.
For arguments, we need something to indicate ownership adoption (
_transfer_ownershipis just a placeholder name to illustrate the idea):For return values, we need something to indicate ownership on the caller side (
_caller_ownsis again just a placeholder):Object ownership problems come up often on the forum (example post), so having a robust way to convey the correct ownership semantics would be a game changer for the robustness of the ROOT Python interfaces.
Alternatives considered