BLD/TST: fix bool block failures when strings are passed to replace list

[cpcloud]

closes #6353

[cpcloud]

I think if this is True then res will always be False in which case I can just return an array of Falses directly and not construct and then fill

[cpcloud]

Oh darn this will not pass. This allows different size lists thru

[cpcloud]

gotta run will get to this tonight after work

[cpcloud]

@jreback This line (from TestSeries.test_replace) is masking the same failure, which is a TypeError and it in fact should be a TypeError, but not because it's trying to index a bool object.

ser = Series(tm.makeTimeSeries().index)
self.assertRaises(TypeError, ser.replace, range(1, 3), [np.nan, 0])

[cpcloud]

I think we should raise a TypeError on these kinds of replacements. It makes no sense to do something like

s.replace({'asdf': 'fda'})

on a bool Series. I would argue that even something like

s = Series([True, False])
s.replace({'asdf': 'fda', True: 'abc'})

should raise as well, because only part of the replacement is valid. @jreback What do you think?

[cpcloud]

Darn, that's not really symmetric with other types though.

[cpcloud]

Problem is that the time series behavior is throwing typeerrors for the wrong reasons, but i don't see a general purpose way to fix both this issue and that one without special case one of them

[cpcloud]

But since we have to maintain the time series behavior for back compat and the bool behavior is now being defined, i think we should raise a typeerror on any of these kinds of conversions...doing that doesn't break any existing code, but makes it a bit more confusing for users to figure out what works .... e.g.,

s = Series([True, False])
s.replace({'a': 'b'})

will raise, but

s.replace('a', 'b')

will not

[jreback]

can you make an entire pass thru on boolean? I think it should just return directly (and if they actually try something like:

Series([True,False]).replace(True : 'b')

maybe just raise

but all other's will simply pass thru...
?

[jreback]

should the possibly compare stuff (and raising) just be in mask_missing?

[cpcloud]

Tried that already, it breaks things

[jreback]

ok then

merge when ready

[cpcloud]

I'm going to write a bit of docs and release notes first ... just to clarify these somewhat dark corners of the replace API

[jreback]

github looks back to form now......pls merge when you can....(of course docs good too!)

[jreback]

looks cool...thanks!

[cpcloud]

got that other replace fix coming soon to a pandas repo near you