axis[key] does not work when key contains special characters

[gdementen]

Those messy string evaluations are getting in the way once more

>>> arr = ndtest(3)
>>> agg = arr.sum('a1:a2;a0:a1;a0,a2')
>>> agg
a  a1:a2  a0:a1  a0,a2
       3      1      2
>>> agg[agg.a['a1:a2']]
ValueError                                Traceback (most recent call last)
...
ValueError: a['a1':'a2'] is not a valid label for any axis
>>> agg['a1:a2']
3
>>> agg[X.a['a1:a2']]
3

[gdementen]

one way to resolve this would be to implement #34, though I am reluctant to do this.

[gdementen]

Another way would be to first try the key as is in Axis.__getitem__:

if key in self:
    return LGroup(key, name, self)
else:
    key = _to_key(key)
    return LGroup(key, name, self)

and AVOID using _to_key in LGroup.__init__

We could even return a PGroup directly. This would be more efficient (only translates the group once) in the usual case: use the group on an array containing that axis and slightly less efficient on the unusual case (use the group on an array without the axis).

if key in self:
    return PGroup(self.translate(key), name, self)
else:
    key = _to_key(key)
    return PGroup(self.translate(key), name, self)

[gdementen]

not using _to_key in LGroup.__init__ would prevent users from using LGroup directly with an "interpretable" string, which is currently the only way to make a group without axis using an interpretable string, eg:

LGroup('10..19', 'teens') 

But this kind of group should be created using another syntax anyway G['10..19'] >> 'teens' or something like that. See #23.

[gdementen]

implementing #34 seems unavoidable.
We could also choose to avoid using/adding special characters in any label we produce (e.g. aggregate labels), but then we would either loose some information, or produce verbose labels. For example, we would get:

>>> agg = arr.sum('a1:a2;a0:a1;a0,a2')
>>> agg # option 1 (loose info)
a  a1_a2  a0_a1  a0_a2
       3      1      2
>>> agg # option 2 (verbose)
a  a1_to_a2  a0_to_a1  a0_a2
          3         1      2

PS: expanding the slices labels to its labels (e.g. a0:a2 to a0_a1_a2) does not seem like a realistic option