Index is not hashable

[ghost]

hash(df.index)
> /home/user1/src/pandas/pandas/core/index.py(346)__hash__()
    345     def __hash__(self):
--> 346         return hash(self.view(np.ndarray))
    347 

/home/user1/src/pandas/pandas/core/index.pyc in __hash__(self)
    344 
    345     def __hash__(self):
--> 346         return hash(self.view(np.ndarray))
    347 
    348     def __setitem__(self, key, value):

TypeError: unhashable type: 'numpy.ndarray'

[wesm]

This could be made to work but won't make Index a valid dict key (since == is vector compare)

[ghost]

since index keys are immutable, having a hash based on them would have enabled a fast check for
index inequality without a full vector compare. not sure it's worth it.

[wesm]

I'd thought at one point about adding an MD5 hash. That would work for indexes that don't contain Python objects

[wesm]

Well, I guess you could hash the python objects then md5 the array of hash values. Some probability of collisions / equal md5 hash but unequal indexes. harrumph

[ghost]

it would need to be at construction time o'course, otherwise no gain over vector compare,
any(idx1==idx2) does not fail fast, so the best case and worst case are the same.
But It turned out that for my case it was enough to check for identity rather then equality.

[wesm]

We should just write a custom set of vector compares-- the problem with np.array_equal is that it compares all elements rather than failing at the first unequal pair

[wesm]

Marked as 0.11. Should think about this eventually

[ghost]

marked 'someday'.

[jtratner]

is_ method covers some potential uses for this (i.e., comparing views), definitely would be nice to at least have a short-circuiting equality function that we could use in equals.

[ghost]

In the year since I haven't needed this again.
Intuitevely, Index is immutable -> should be hashable. But indices have name(s) and those are mutable.
And this raises tricky issues of identity, equality and context-specific "right thing to do".

Would need much stronger use case to commit to semantics on implementaition,
and I can't think of one now. Leave it until that mystical, musical, majestical day.