Opened 15 years ago
Last modified 8 weeks ago
#12096 assigned New feature
Model fields are not accessible as attributes of the model class
Reported by: | sejo | Owned by: | Clifford Gama |
---|---|---|---|
Component: | Database layer (models, ORM) | Version: | dev |
Severity: | Normal | Keywords: | model, docstrings, attributes |
Cc: | Triage Stage: | Accepted | |
Has patch: | no | Needs documentation: | no |
Needs tests: | no | Patch needs improvement: | no |
Easy pickings: | no | UI/UX: | no |
Description
When using attribute docstrings in my models they will never be picked up by doctools as the attributes disappear top level from the model.
This makes me to keep separate documentation and code for djagios.
Change History (12)
comment:1 by , 15 years ago
comment:2 by , 15 years ago
Owner: | changed from | to
---|---|
Status: | new → assigned |
comment:3 by , 15 years ago
Triage Stage: | Unreviewed → Accepted |
---|
comment:4 by , 15 years ago
comment:5 by , 14 years ago
Severity: | → Normal |
---|---|
Type: | → Bug |
comment:8 by , 12 years ago
Component: | Core (Other) → Database layer (models, ORM) |
---|
comment:9 by , 9 years ago
Owner: | removed |
---|---|
Status: | assigned → new |
Summary: | model attributes are not set toplevel which causes doctools to miss the attribute docstrings → Model fields are not accessible as attributes of the model class |
Type: | Bug → New feature |
I'm not sure what the implementation should look like. It seems that Jacob's comment about "ModelClass.field
raises AttributeError
is obsolete.
comment:10 by , 9 years ago
To get this to work we need to change some low-level stuff since model fields are stripped out by the metaclass to be put in Model._meta. Since Sphinx's autodoc module allows to manually add members to a class I think it would be better to do in manually than trying to change the way Django models are constructed.
Anyway, I came up with the following code that makes Sphinx's autodoc work for Django models. It is very hackish and unsurprisingly makes some tests fail, so I'm just putting this here for the record in case anyone would want to take that over someday.
class ModelBase(type): def __getattr__(self, attr): if attr == '_meta': raise AttributeError(attr) try: return self._meta.get_field(attr) except FieldDoesNotExist: raise AttributeError(attr) @property def __dict__(self): try: model_dict = { f.name: f for f in self._meta.get_fields() } except AppRegistryNotReady: model_dict = {} model_dict.update(super(ModelBase, self).__dict__) return model_dict # Rest of ModelBase code...
comment:11 by , 2 years ago
Since DeferredAttribute
are systematically added as model class attributes (#26207) this would be a simple matter of adjusting a few __get__(instance=None)
case to return self.field
(DeferredAttribute
and the three classes of related_descriptors
).
The only remaining concern here is backward compatibility. The only documented pattern of accessing a field attribute from the class is for ManyToMany.through
via ManyToManyDescriptor.through
but there might be valid use cases in the wild for expecting descriptors to be returned that this change could break. As long as we document this change and that it's still possible to retrieve descriptors by doing ModelClass.__dict__.get('descriptor')
I think we should be fine though.
Beyond the documentation benefits this could pave the way for interesting features that the verbosity of Model._meta.get_field
such as SQLAlchemy-esque lookups (e.g. Book.objects.filter(Book.title.contains(Book.author.name))
) but most importantly it would restore the most intuitive pattern for field retrieval.
comment:12 by , 8 weeks ago
Owner: | set to |
---|---|
Status: | new → assigned |
Version: | 1.1 → dev |
This is actually because
ModelClass.field
raisesAttributeError
from the descriptor. I've been wanting to stop that, anyway -- it ought to just return theField
object.