Table inheritance do not keep column option

[nimch]

Hi there !

First, thank you for sharing your work.

Second, the issue : when inherit from personal Table class, it seems that Column arguments (like verbose_name, orderable, etc) are not kept, right ?

Example:

class MyTable(tables.Table):
    item1 = tables.Column(verbose_name="Nice column name")
    class Meta:
        model = MyModel
        fields = ('item1',)

class MyTableExt(MyTable):
    item2 = tables.Column()
    class Meta(MyTable.Meta):
        fields = ('item1', 'item2')

In MyTableExt, item1 do not have "Nice column name", always right ?

Thanks.

[jieter]

Hmm, that seems to be the case indeed. Pushed a failing tests to branch issue-337 based on your example. Will look into it, thanks for reporting!

[jieter]

@nimch This is what happens:

The metaclass converts the explicit and implicit columns of MyTable to MyTable.base_columns giving explicit columns higher precedence. When it gets extended into MyTableExt, the metaclass does take into account it's parents base_columns, but the difference in precedence between explicit/implicit columns is lost between those defined in MyTable and MyTableExt. Because there is an implicit definition for item1 in MyTableExt, it overrides the definition in it's parents base_columns.

I'll think about a way to fix this, in the mean time: in this case there is no need to define fields = ('item', 'item2'), if you comment that out it works as expected.

[nimch]

Unfortunately, your proposal don't solve my issue. I still get "default" verbose_names.
Otherwise, every columns defines with accessor argument is empty, same issue probably.
I have to stop investigation for today but I will continue tomorrow.
See you.

[jieter]

@nimch interesting, I did exactly that here, which makes the build pass

[nimch]

Hello, I'm back !

I'm not convinced of your test code... Where is MyTable definition for example ?

I redo tests with minimalist code to be sure (and not my more complex code hacked).
Here the code of my_table.py :

#!/usr/bin/env python3

from django.db import models
from django_tables2 import Table, Column

class MyModel(models.Model):
    item1 = models.CharField()

class MyTable(Table):
    item1 = Column(verbose_name="nice verbose name")

    class Meta:
        per_page = 20
        model = MyModel
        fields = ('item1',)

class MyTableExt(MyTable):
    item2 = Column(verbose_name="nice verbose name too")

    class Meta(MyTable.Meta):
        per_page = 22

Here Python shell :

>>> from tracadpp import my_table
>>> table = my_table.MyTable(my_table.MyModel.objects.all())
>>> print(table.columns['item1'].verbose_name)
nice verbose name
>>> table_ext = my_table.MyTableExt(my_table.MyModel.objects.all())
>>> print(table_ext.columns['item1'].verbose_name)
item1
>>> print(table_ext.columns['item2'].verbose_name)
nice verbose name too

As you can see, verbose name defined in MyTable is lost in MyTableExt.
It could be important : I'm using Python 3.4, Django 1.8.7 and django-tables2 1.1.0.

[jieter]

The diff was collapsed which made the MyTable definition missing see:
https://github.com/bradleyayers/django-tables2/blob/cf63e44bbfae1d08c56578261fd5552c1da33956/tests/columns/test_general.py#L341-L374

The tests run in CI against django 1.8 in python 2.7, 3.3 and 3.4. Your version of django-tables2 shouldn't matter but it would keep things more clear if you upgraded to the latest version.

I just noticed why it fails in your example and works in mine: if you have no class Meta on the child class at all, it works as expected, but as soon as you have an empty class Meta(MyTable.Meta):, the test fails.

[nimch]

Empty Meta class is not allowed so test fails, this is correct.

Same behaviour here : without any definition of Meta in child class, parent class fields' verbose_name are preserved.

Well, inheritance without the ability of modifying Meta option is less useful, no ? Do you consider this is normal behaviour ?

Thanks for your help.

[jieter]

Empty Meta class is not allowed so test fails, this is correct.

It does not fail on it being empty, the test fails on a failed assertion.

I do consider it a bug (issue is still open), but didn't have the time to fix it yet. If you have time and skill to fix it, feel free to open a pull request!

[nimch]

Hi !
I've just tested your correction and it's ok for me : verbose_name, accessor, etc are preserved in child table.
Good job !

[jieter]

@nimch thanks for verifying!