Repository navigation
"Already defined" + "not defined" errors with SQLAlchemy 1.2 hybrid_property #4430
Description
Activity
Can you include the content of test.py?
Oh sorry, it's just exactly the same as that example code (only the first class def):
from sqlalchemy import Base, Column, String from sqlalchemy.ext.hybrid import hybrid_property class FirstNameOnly(Base): first_name = Column(String) @hybrid_property def name(self): return self.first_name @name.setter def name(self, value): self.first_name = valueI had the wrong line numbers on the error above (edited now), the error is on line 11,
@name.setter.I guess this is because mypy doesn't understand that
@hybrid_propertybecause similar to@property.Also possibly you may have to take this up with https://github.com/JelleZijlstra/sqlalchemy-stubs.
Since there are no stubs for
sqlalchemyby default, the decorator will be seen asAnyvalue by mypy. One option would be to not complain about redefinitions if the first definition is known to have anAnytype.More generally,
hybrid_propertycould be an alias forproperty, and in that case the errors are more clearly false positives.Guido suggested in #220 that someone could try to build hybrid property support on top of mypy's descriptor support. I was able to get this going by defining a typed_hybrid_property generic class (in a pyi file), parameterized by python type and sql type. I had to define my hybrid properties as separate functions with different names (like
_prop_get,_prop_exprand such), and then merge them all into the name I wanted usingprop = hybrid_property(_prop_get, expr=_prop_expr).I have so far been unable to convince mypy to type-check the multipart definition (using
@prop.setter) because that relies on the magic in mypy's semantic analyzer which notices @Property declarations.It would be super-nice if mypy provided a way to "get the @Property behavior" for additional decorators not named in the mypy source.
@lincolnq This is something that probably can be done by a plugin. We currently have a plugin hook for classes decorated with a given decorator. We can probably just add a similar hook for functions.
Btw, we are currently working on SQLAlchemy stubs (and soon mypy plugins) at https://github.com/dropbox/sqlalchemy-stubs
@ilevkivskyi Do you have an idea of how a plugin could be used to silence these errors mentioned above if you use a
.setteror.expression?:test.py:11: error: Name 'name' already defined test.py:11: error: Name 'name' is not definedIt sounds like the support for
@propertyis baked into mypy itself, so I'm not sure how we would use a plugin to avoid these errors.I was running into the same problem with two hybrid fields.
I was able to fix theerror: Name 'XXX' already definedby adding# type: ignoreand interestingly also fix the other issue by just reordering the functions.Original:
class Channel(db.Model): __tablename__ = 'Channel' id = db.Column('id', db.Integer, nullable=False, primary_key=True) names = db.relationship( ChannelName, primaryjoin=ChannelName.channel_id == id, backref='channel', order_by="desc(ChannelName.default)" ) @hybrid_property def name(self) -> Optional[ChannelName]: return self.names[0] if len(self.names) else None @name.expression def name(cls): return select([ChannelName.name]) \ .where(ChannelName.channel_id == cls.id) \ .limit(1) \ .label('name') @hybrid_property def aliases(self) -> List[ChannelName]: return self.names[1:] @aliases.expression def aliases(cls): return select([ChannelName.name]) \ .where(ChannelName.channel_id == cls.id) \ .offset(1) \ .label('aliases') def on_change(channel: Channel): if channel.name: channel.name.default = True for alias in model.aliases: alias.default = False
Error;
error: Name 'name' already defined on line XXX error: Name 'aliases' already defined on line XXX error: overloaded function has no attribute "default" error: overloaded function has no attribute "__iter__" (not iterable)Modified to:
class Channel(db.Model): __tablename__ = 'Channel' id = db.Column('id', db.Integer, nullable=False, primary_key=True) names = db.relationship( ChannelName, primaryjoin=ChannelName.channel_id == id, backref='channel', order_by="desc(ChannelName.default)" ) @hybrid_property # type: ignore def name(self) -> Optional[ChannelName]: return self.names[0] if len(self.names) else None @hybrid_property # type: ignore def aliases(self) -> List[ChannelName]: return self.names[1:] @name.expression # type: ignore def name(cls): return select([ChannelName.name]) \ .where(ChannelName.channel_id == cls.id) \ .limit(1) \ .label('name') @aliases.expression # type: ignore def aliases(cls): return select([ChannelName.name]) \ .where(ChannelName.channel_id == cls.id) \ .offset(1) \ .label('aliases') def on_change(channel: Channel): if channel.name: channel.name.default = True for alias in model.aliases: alias.default = False
Result:
Success: no issues found in 9 source files- added a commit that references this issue
on Oct 20, 2020 Another solution that I found and that doesn't require using
#type: ignoreis to make thehybrid_propertyhave the same typing as a normal property when type checking.from sqlalchemy import Base, Column, String if TYPE_CHECKING: # This makes hybrid_property's have the same typing as normal property until stubs are improved. hybrid_property = property else: from sqlalchemy.ext.hybrid import hybrid_property class FirstNameOnly(Base): first_name = Column(String) @hybrid_property def name(self): return self.first_name @name.setter def name(self, value): self.first_name = value
Reacted by Stephen J. Fuhry, Henrik Christensen, TornaxO7, Thomas Jost, Jon Bringhurst, Zach B, Trevor Gross, Johnny Villegas, Ľubomír Brindza, senseysensor and 1 morefrom typing import TYPE_CHECKINGThank you! :)
It seems that sqlalchemy is deprecating their mypy plugin. Unfortunately, that means it is unlikely for this issue to be solvable in a neat way. That being said, another plugin could eventual implement either #7468 or #6760, to fix this.
That being said, I don't think this issue is actionable beyond the above plugin issues, so I am going to close it.
SQLAlchemy 1.2 was recently released, and changed its behavior of
hybrid_propertya bit to line up with how Python's@propertyworks. Specifically, it's now necessary to use the same method name for both the getter and setter, as seen in the example here (both aredef name(...)): http://docs.sqlalchemy.org/en/latest/changelog/migration_12.html#hybrid-attributes-support-reuse-among-subclasses-redefinition-of-getterThis is causing mypy to error though, it now emits two (contradictory?) errors on the
@name.setterline: