Opened 6 years ago
Closed 6 years ago
#29505 closed Bug (fixed)
SchemaEditor prohibits setting a field's default value to a callable that'll be used as the default
Reported by: | Eugene Pakhomov | Owned by: | nobody |
---|---|---|---|
Component: | Migrations | Version: | 2.0 |
Severity: | Normal | Keywords: | default |
Cc: | Triage Stage: | Accepted | |
Has patch: | yes | Needs documentation: | no |
Needs tests: | no | Patch needs improvement: | no |
Easy pickings: | no | UI/UX: | no |
Description
I'm trying to use https://github.com/cognitect/transit-python and I created a custom KeywordField
for this.
The problem is that transit.transit_types.Keyword
is a callable. Now, even if I override _get_default
in KeywordField
, it still won't work because django.db.backends.base.schema.BaseDatabaseSchemaEditor#effective_default
checks the value again.
I think a simple fix would be to remove this check from effective_default
and immediately make a call where it's necessary (datetime.date
etc).
Attachments (1)
Change History (9)
comment:1 by , 6 years ago
Component: | Database layer (models, ORM) → Migrations |
---|---|
Easy pickings: | unset |
comment:2 by , 6 years ago
Suppose I have some custom type like this one
class MyStr(str): def __call__(self, other): return self + other
and a corresponding custom field type that inherits from Django's TextField
and implements all necessary methods so that MyStr
is used everywhere instead of the regular str
.
Currently, it's not possible without manually creating a DatabaseWrapper
with a custom DatabaseSchemaEditor
because even if you handle the fact that instances of MyStr
are callable in the custom field type, they will still be called in django.db.backends.base.schema.BaseDatabaseSchemaEditor#effective_default
.
Attached a patch that fixes this issue. Tests appear to be fine.
by , 6 years ago
Attachment: | 29505.diff added |
---|
comment:3 by , 6 years ago
Resolution: | → wontfix |
---|---|
Status: | new → closed |
Summary: | Impossible to set a Field's default value to a callable → Impossible to set a Field's default value to a callable that takes an argument |
I believe the proposed patch would break callable defaults. In effective_defautl()
, field.get_default()
may return a callable and Field.get_prep_value()
won't work properly if it receives a callable instead of a value. I guess test coverage is lacking.
I think you should change your __call__()
method to make other
an optional argument.
comment:4 by , 6 years ago
As I said in the issue description - the types under question are not under my control, so I cannot change __call__()
. And my concern doesn't have to do anything with my particular situation - I have already solved it by setting a custom DatabaseSchemaEditor
. My concern is that this issue is still there, and other people will have to spend some time debugging it.
field.get_default() may return a callable
You're incorrect. Django itself doesn't do that. And Model#__init__()
implementation calls get_default()
without checking whether it's result is callable or not because the default implementation of get_default()
always does this check. If get_default()
could return a callable, Model#__init__()
would not work correctly - it would set the corresponding field in all instances to the callable, without calling it to get the default value.
comment:5 by , 6 years ago
Also, the change in title is incorrect.
Technically, it's possible to set the default value to something that is a callable without arguments, yes. But it's impossible to also have this value as the default, instead of just have it be called for a default.
Consider my example above but without the other
argument. You still won't be able to set an instance of MyStr
as the default value.
comment:6 by , 6 years ago
Has patch: | set |
---|---|
Needs tests: | set |
Resolution: | wontfix |
Status: | closed → new |
Summary: | Impossible to set a Field's default value to a callable that takes an argument → SchemaEditor prohibits setting a field's default value to a callable that'll be used as the default |
Triage Stage: | Unreviewed → Accepted |
Type: | Uncategorized → Bug |
After a second look, I agree with your analysis. The patch looks okay, and the test will probably go in tests/schema
.
I don't understand the issue. Please explain in more detail and perhaps offer a patch to explain.