CLN: CacheableOffset not Properly Used

[cancan101]

I believe that many of the DateOffsets that extend CacheableOffset do so incorrectly.

Looking at pandas.tseries.offsets, there are some subclasses of DateOffset are declared as class Foo(DateOffset, CacheableOffset) and some as class Foo(CacheableOffset, DateOffset).

All that CacheableOffset does is to set the class field _cacheable = True. DateOffset sets _cacheable = False. Therefore, depending on the order of the parent classes, in some cases _cacheable will be False and some cases True

[jreback]

ok...can you find an instance where this is 'wrong'? and a PR to clean up? (could just be some old code)

[cancan101]

That code does not look old. I am not exactly sure how _cacheable is being used; I will have to look more closely. I can certainly provide a python example of how it is broken:

class Cache(object):
   _foo = True

class DT(object):
   _foo = False

class A(Cache, DT):
   pass

class Bob(DT, Cache):
   pass

B()._foo
False

A()._foo
True

[cancan101]

On closer inspection, I do not see where either _cacheable or _should_cache is used.

[cancan101]

I take that back. I found the following in tseries.index:

            if (offset._should_cache() and
                not (offset._normalize_cache and not _normalized) and
                    _naive_in_cache_range(start, end)):
                index = cls._cached_range(start, end, periods=periods,
                                          offset=offset, name=name)
            else:
                index = _generate_regular_range(start, end, periods, offset)