REF: simplify operating-columnwise dispatch

[jbrockmendel]

After this we have one usage of split_and_operate left and 2 places where we can almost use @maybe_split. In both cases I'm still trying to untangle things.

[jreback]

rebase after we merged the other one, but have some questions here

[jbrockmendel]

rebase after we merged the other one, but have some questions here

rebasd + green. questions?

[jreback]

I find using a decorator & having _split very confusing.

[jbrockmendel]

hah i find split_and_operate (the non-decorator alternative) confusing. maybe it can be simplified now that putmask no longer uses it (so one or more of its args can be removed)

[jbrockmendel]

so i can re-write split_and_operate to be basically just the if self.ndim == 2 and self.shape[0] != 1 part of maybe_split, but then every method that is decorated in this PR looks like:

def foo(self, *args, **kwargs):
    if self.ndim == 2 and self.shape[0] != 1:
       return self._split_and_operate(Block.foo, *args, **kwargs)
    [...]

and avoiding that repetition seems like exactly the use case for a decorator

[jbrockmendel]

tried this out and it gets more verbose bc we have to write out *args/**kwargs each time

[jreback]

mixing these approaches just makes this more confusing. pick.

[jbrockmendel]

ok, will update shortly.

[jreback]

if you can remove _split entirely and just use the decorator, I think that might be better.

[jbrockmendel]

trouble is we also use it in BlockManager.get_bool_data

[jbrockmendel]

ohh, i think i may now understand the confusion.

is the confusing thing "why are we iterating over self._split() here instead of using the decorator? i.e. why not use the decorator everywhere?"?

if so, the reason is bc in this method, the conditions under which we want to split are slightly different from the conditions checked in the decorator. im hopeful we can get some further simplification, but dont want to get it by jamming too much complexity into maybe_split (which is what got split_and_operate confusing in the first place)

[jbrockmendel]

updated to re-write and re-use split_and_operate

[jreback]

rebase, but need to relook

[jbrockmendel]

rebased, green ex two issues on master being addressed elsewhere

[jreback]

ok improves on existing (tangled code), splitting block ops not very clear where its happening, but I am sure future PRs can help address.