Is overriding node labels with __label__ considered safe?

[nquinn-ek]

Hi,

I wanted to name my Neomodel model classes with a trailing "Node", for instance PersonNode, in order to distinguish them from other types of models (specifically, Pydantic models). However, I don't want all my nodes in the graph to have "Node" in the name, so I looked in the docs for a way to override the label instead of inheriting it from the class, and found nothing.

But, from looking at the __new__ method for the NodeMeta class, it appears that the label of the node is stored in the __label__ property, and that label is taken from the name of the class unless already set:

cls.__label__ = namespace.get("__label__", name)

So it appears we can override the label of the node by setting this as a attribute on the class:

class PersonNode(neomodel.StructuredNode):
    __label__ = "Person"

I tested it in the CLI and it seems to work. It seems similar to what you can do in SQLAlchemy to point to the name of the table:

class User(Base):
    __tablename__ = "user"

So I am wondering, does anyone know if this considered part of the public API and safe to hook into, or a private implementation detail that it's risky to rely on?

[aanastasiou]

@nquinn-ek I would not call it "best practice" (__label__ is "private" anyway), as it would differentiate what you perceive as an entity name in your program VS the backend. The label of a given entity is very important for the proper serialisation / deserialisation of the entity with the backend as well as setting up any indices. Most of the times, you don't have to deal with setting it manually.

Having said that, if you are extending a data model and you have such clashes, you could resolve them by modifying the __label__ as a last resort. It would be risky if you run across situations where you have to re-set that label a few times, in which case you would have to cleanup any remnants manually.

Hope this helps.

[nquinn-ek]

thanks!

[yachal-upson]

Here for the same reason. Frustrated that label is inferred from class name.
In my case I'm developing a FastAPI backend so have Pydantic models sitting alongside Neomodel models, with a need to differentiate for sanity. Like the OP said, I too wanted to append "NodeSchema" to my neomodels, and "NodeModel" to my pydantics.

Had to resort to setting label at declaration (will not be touched after). Would have almost been nice to have a public class attribute.