QueryBuilder 'order_by' has no effect if the same type is used twice

[greschd]

Describe the bug

When the same type is appended to the QueryBuilder twice (with difference filter clauses etc.), the order_by statement appears to have no effect.

Minimal example

Running the following code, the result does not respect the ctime ordering:

from aiida import orm
from aiida.engine import calcfunction


@calcfunction
def add(x, y):
    return x + y


x = orm.Int(1).store() # explicitly storing so that I know the order of creation
y = orm.Int(3).store()
z = orm.Int(5).store()

c = add(add(x, y), z)

qb = orm.QueryBuilder()
qb.append(orm.Int, filters={'uuid': c.uuid}, tag='c')
qb.append(orm.Int, with_descendants='c')
qb.order_by({orm.Int: {'ctime': 'desc'}})
print(qb.distinct().all())

The printed output is (formatted for readability)

[
  [<Int: uuid: 7c1a439c-9c04-419e-81f9-d4b6cb2de073 (pk: 59793) value: 1>], 
  [<Int: uuid: 3fb15a95-81a9-48e3-9ca7-359738de82d8 (pk: 59794) value: 3>], 
  [<Int: uuid: 8459de2a-eeb2-45aa-9d8a-6cd86d27a5ac (pk: 59795) value: 5>], 
  [<Int: uuid: 41f522fa-a473-46e6-be95-b5774e46b074 (pk: 59797) value: 4>]
]

which is in ascending ctime order. In fact, switching from desc to asc has no effect.

Counterexample: using a different type

Using a different type for the two append statements, the ordering works as expected:

from aiida import orm
from aiida.engine import calcfunction


@calcfunction
def add(x, y):
    return orm.Float((x + y).value)


x = orm.Int(1).store()
y = orm.Int(3).store()
z = orm.Int(5).store()

c = add(add(x, y), z)

qb = orm.QueryBuilder()
qb.append(orm.Float, filters={'uuid': c.uuid}, tag='c')
qb.append(orm.Int, with_descendants='c')
qb.order_by({orm.Int: {'ctime': 'desc'}})
print(qb.distinct().all())

Result:

[
  [<Int: uuid: 59350c15-1421-4ca1-ba8a-7da60815030f (pk: 59802) value: 5>], 
  [<Int: uuid: d782fbec-85cf-4c7a-8b0d-a529d17625f1 (pk: 59801) value: 3>], 
  [<Int: uuid: f04c16eb-91f5-4f76-89b7-a268ef747de1 (pk: 59800) value: 1>]
]

Environment

• Operating system: Ubuntu 18.04, via WSL2
• Python version: 3.7.3
• aiida-core version: Latest develop (477fe30), Django backend

Workaround

A workaround is using the tag and referring to the entity that should be ordered with that, instead of passing the type. If the above is expected behavior, I think the order_by documentation should strongly encourage the use of tags.

[chrisjsewell]

@greschd in #5093, this will now fail, rather than silently converting the object to the first tag (the current behaviour)

    ...: print(qb.distinct().all())
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-18-a52ad4c23029> in <module>
     17 qb.append(orm.Int, filters={'uuid': c.uuid}, tag='c')
     18 qb.append(orm.Int, with_descendants='c')
---> 19 qb.order_by({orm.Int: {'ctime': 'desc'}})
     20 print(qb.distinct().all())

~/Documents/GitHub/aiida_core_develop/aiida/orm/querybuilder.py in order_by(self, order_by)
    599                 if not isinstance(items_to_order_by, (tuple, list)):
    600                     items_to_order_by = [items_to_order_by]
--> 601                 tag = self._tags.get(tagspec)
    602                 _order_spec[tag] = []
    603                 for item_to_order_by in items_to_order_by:

~/Documents/GitHub/aiida_core_develop/aiida/orm/querybuilder.py in get(self, tag_or_cls)
   1403         if self._cls_to_tag_map.get(tag_or_cls, None):
   1404             if len(self._cls_to_tag_map[tag_or_cls]) != 1:
-> 1405                 raise ValueError(
   1406                     f'The object used as a tag ({tag_or_cls}) has multiple values associated with it: '
   1407                     f'{self._cls_to_tag_map[tag_or_cls]}'

ValueError: The object used as a tag (<class 'aiida.orm.nodes.data.int.Int'>) has multiple values associated with it: {'Int_1', 'c'}

Would you consider this to close the issue?

[greschd]

Yeah, I guess that's clear enough.