fill_value is implicitly assumed to be a subscriptable numpy type

[Jonathan Chambers]

fill_value is described as a scalar but in is assumed to be a subscriptable numpy type in_Elemwise._get_fill_value, leads to errors if you supply your own fill value.

e.g.

        zero_args = tuple(
            arg.fill_value[...] if isinstance(arg, COO) else arg
            for arg in self.args
        )

fails with a normal float fill_value

[Deepak Singh]

Can we raise a typeerror exception here . and if it is raised put the varible inside a list like

try: 
   zero_args = tuple(
              arg.fill_value[...] if isinstance(arg, COO) else arg
              for arg in self.args
          )
except TypeError:
     arg.fillvalue = [arg.fillvalue]
   zero_args = tuple(
              arg.fill_value[...] if isinstance(arg, COO) else arg
              for arg in self.args
          )

[Hameer Abbasi]

I would suggest making self.fill_value a read-only property, that can only be set in the ctor.

[Hameer Abbasi]

I feel this is unsupported utilization of an internal API based on the comments. Closing as wontfix.