core.internals Roundup

[jbrockmendel]

Wishlist:

• It'd be nice if self._data always pointed to a BlockManager object. This would require some changes to the existing accessor classes and Index.
• It'd be nice if the exposed surface of core.internals were reduced.
• DISCUSSION: allow external libraries to define a custom Block #17144 Allow external libraries to define custom Block
• API/ENH: allow Period to be a first-class type via a PeriodBlockinternal type #7964 allow Period to be a first-class type via a PeriodBlock internal type

Refactoring:

• CLN: blocks, not series, should contain logic for arithmetic #9859 blocks, not series, should contain logic for arithmetic

Cleanup:

• 9 "# FIXME" comments
• 6 "# TODO" comments
• identical comments on BoolBlock.convert and ObjectBlock.convert: # TODO: Refactor when convert_objects is removed since there will be 1 path. I think convert_objects has been removed, so this may be ready.
• Several comments suggesting workarounds for numpy<1.7 that may no longer be needed.
• I've been told replace_list is a kludge
• form_blocks vs make_block inconsistency #19179 form_blocks and make_block have small inconsistencies. If these were reconciled, some redundant code could be removed.

Possible Bugs:

• core.internals checking hasattr(item, 'reindex_axis') #19243 Checks for hasattr(item, 'reindex_axis') are used but the reindex_axis methods themselves are never called
• TimedeltaBlock.fillna treats int as second #19233 TimedeltaBlock.fillna assumes integer is a second, not ns (or refusal to assume)

[jbrockmendel]

This is clutter at this point. Closing.