IPython tab-completion populates the Dict with a few unwanted fields about addict

[sabhiram]

This occurs because some piece of code ends up checking to see if a given attribute is part of the Dict instance.

This will probably be an on-going problem (similar to the need for us to implement a dir method), as long as external code ends up checking to see if the object under question implements a method, or contains an attribute.

[mewwts]

Yup - I'm not entirely convinced there's a solid solution to this.

[sabhiram]

We could alternatively (just for iPython) detect that we are being spawned from an iPython shell and return the appropriate expected values from the getattr code.

not_ipython = False
try:
    __IPYTHON__
except NameError:
    not_ipython = True

I don't have an iPython env around, so ill play around w/ this a bit later. Just food for thought as usual :)

[mewwts]

Thanks @sabhiram, but I hope and think it can be handled in a more elegant way. Perhaps not without a rather large rewrite, but we'll see.

[mewwts]

There's a fix that I have in a local branch which is kind of hacky.
When a nonexistent item is requested, instead of creating an empty Dict inside the original Dict, we return an empty Dict, with some (hidden) properties e.g. 'parent' and 'key'. When an item is set in this returned Dict, it attaches itself to its parent with key='key'. This would eliminate the problems we're experiencing in iPython, and also reduce the importance of prune. This does however have memory-implications.

Thougts @sabhiram, @burk, @Who8MyLunch?

[Who8MyLunch]

The problem described by this issue is that addict.Dict is very forgiving when a key or attribute does not exist. The following code currently raise no exceptions since Dict always creates a new key if one does not already exist:

d = addict.Dict()

val_x = d.X
val_y = d['Y']

bool_z1 = hasattr(d, 'Z1')
bool_z2 = 'Z2' in d

print(d)

print(bool_z1)
print(bool_z2)

The above code produces the following output:

{'X': {}, 'Y': {}, 'Z1': {}}
True
False

Is this really the desired behavior?

Shouldn't these two lines yield similar results?

bool_z1 = hasattr(d, 'Z1')
bool_z2 = 'Z2' in d

[mewwts]

Hi @Who8MyLunch, sorry for the 1Y+ delay in response.

Point here is that in used __contains__ and hasattr uses __getattr__. So either we implement __contains__ to create a new key, or we leave it as it.