RangeIndex copy behaviour differs from other indices

[toobaz]

It is my understanding that the copy parameter of RangeIndex is present virtually only for API compatibility with other indices: however the behaviour for copy=False differs from other indices in what follows:

In [2]: i = pd.Int64Index(range(10))

In [3]: i.name = 'original'

In [4]: j = pd.Int64Index(i)

In [5]: j.name

In [6]: j.name = 'copy'

In [7]: i.name
Out[7]: 'original'

In [8]: i = pd.RangeIndex(10)

In [9]: i.name = 'original'

In [10]: j = pd.RangeIndex(i)

In [11]: j.name
Out[11]: 'original'

In [12]: j.name = 'copy'

In [13]: i.name
Out[13]: 'copy'

In other words, since RangeIndex doesn't have any data to copy, the value of the copy parameter should be irrelevant. For homogeneity with other indices, we should always act as if copy=True. The docs should also be changed to state that copy parameter has no effect and is present only for API compatibility. I'll push a PR if you agree.

[jreback]

this is a bug in the constructor I think. It should match what Int64Index does.
It doesn't matter that RangeIndex doesn't have data to copy, it still should make a new index.

[jreback]

So Int64Index, Float64Index all have the same issue.
They are not propogating the name of a passed index on construction. That is the issue here.
RangeIndex actually is ok.

These look ok

In [6]: a = Index(list('abc'),name='foo')

In [7]: Index(a)
Out[7]: Index([u'a', u'b', u'c'], dtype='object', name=u'foo')

In [20]: i = Index(xrange(3),name='foo')

In [21]: i
Out[21]: RangeIndex(start=0, stop=3, step=1, name=u'foo')

In [22]: pd.RangeIndex(i)
Out[22]: RangeIndex(start=0, stop=3, step=1, name=u'foo')

While Int64Index does not

In [8]: i = Index(range(3),name='foo')

In [9]: Index(i)
Out[9]: Int64Index([0, 1, 2], dtype='int64', name=u'foo')

In [10]: pd.Int64Index(i)
Out[10]: Int64Index([0, 1, 2], dtype='int64')

[toobaz]

I think we have 2 problems here: the one I reported and the one you now point at. If we fix the one you're pointing at, we will still have a difference that Int64Index(Int64Index(), copy=*) will create two indices no matter the value of copy, while RangeIndex(RangeIndex(), copy=False) will not, and this reflects in the impossibility of giving different names. So you're right in Int64Index & co. being buggy, but this doesn't have much to say on my PR.

[toobaz]

Or to put it shortly: this test fails in master. I think it shouldn't. Do you agree? [Y/n]

[jreback]

ok, going to merge your fix, pls open a new issue for the Int64Index/Float64Index name issues I desrcribe above.