builder returns wrong values of inputs defaults when namespaces are used in expose_inputs

[bosonie]

Describe the bug

Several misleading behaviors (fairly weird I would say) have been detected when playing with default of inputs ports that have been exposed inside a namespace. See below to reproduce them.

Steps to reproduce

In verdi shell:

from aiida import orm
from aiida.engine import WorkChain
class ClassA(WorkChain):
    @classmethod
    def define(cls, spec):
        super().define(spec)
        spec.input("www",valid_type=Float,default=lambda: Float(1.0))
        spec.outline(cls.brun)
    def brun(self):
        return None

class ClassB(WorkChain):
    @classmethod
    def define(cls, spec):
        super().define(spec)
        spec.expose_inputs(ClassA,namespace="dddd")
        spec.expose_inputs(ClassA,namespace="aaaa")
        spec.inputs["dddd"]["www"].default = lambda: Float(33.0)
        spec.outline(cls.run)

    def run(self):
        self.report("w")

b=ClassB.get_builder()

First of all, typing b.www() returns something. It shouldn't be since the input is exposed in a namespace.
Second, b.dddd.www() does not return the correct default. Please note that the error is introduced only when the line spec.expose_inputs(ClassA,namespace="aaaa") is present. If removed, the default is correctly returned.

Expected behavior

Should be clear from above

Important node

The bug is only on the value returned by the builder. In fact ClassB.spec().inputs._ports["dddd"]["www"].default() correctly shows the right default. Same for b._process_spec.inputs._ports["dddd"]["www"].default(). And also, when the workchain is launched, the correct inputs are used.
Therefore it only seems a problem of "updating" what is returned by the "getter".

Your environment

• Operating system [e.g. Linux]: Linux
• Python version [e.g. 3.7.1]: 3.7.10
• aiida-core version [e.g. 1.2.1]: 1.6.5

[sphuber]

Thanks @bosonie . Can you have another look at your example and I cannot reproduce the problem.

In [1]: from aiida import orm
   ...: from aiida.engine import WorkChain
   ...: class ClassA(WorkChain):
   ...:     @classmethod
   ...:     def define(cls, spec):
   ...:         super().define(spec)
   ...:         spec.input("www",valid_type=Float,default=lambda: Float(1.0))
   ...:         spec.outline(cls.brun)
   ...:     def brun(self):
   ...:         return None
   ...: 
   ...: class ClassB(WorkChain):
   ...:     @classmethod
   ...:     def define(cls, spec):
   ...:         super().define(spec)
   ...:         spec.expose_inputs(ClassA,namespace="dddd")
   ...:         spec.expose_inputs(ClassA,namespace="aaaa")
   ...:         spec.inputs["dddd"]["www"].default = lambda: Float(33.0)
   ...:         spec.outline(cls.run)
   ...: 
   ...:     def run(self):
   ...:         self.report("w")
   ...: 
   ...: b=ClassB.get_builder()

In [2]: b
Out[2]: {'metadata': {}, 'dddd': {'metadata': {}}, 'aaaa': {'metadata': {}}}

In [3]: b.www()
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-3-087b93e24a03> in <module>
----> 1 b.www()

AttributeError: 'ProcessBuilder-1051c00d-78f5-4878-8f87-6254f37ee16' object has no attribute 'www'

The resulting namespace looks ok to me. And calling b.www doesn't return anything. I am not sure why you are trying to call a member of the builder anyway? Was this a typo? Thought so initially, but then you do it again later in your post with b.dddd.www().

[bosonie]

The fact you call is due to the fact that the default is a lambda, so a function. In any case the same problem can be seen without a lambda, for instance when you have a non_db input port and therefore you define simple python types as defaults.
I confirm my findings. I repeated my example in another machine (fedora) and I get the same problem.

[sphuber]

After looking at it some more, it is clear that this is a bug in 1.6.5 that was already fixed in develop. This is the same symptom of the bug reported in #4420 which was resolved in #4984

[sphuber]

Not sure if we want to release a patch 1.6.6 with a backport or we leave it for now since 2.0 is around the corner. Once v2.0 is released we can close this for sure.

[sphuber]

Closing this since v2.0 has been released