BUG: fix construction of NonConsolidatableBlock with inconsistent ndim

[jorisvandenbossche]

• closes BUG: NonConsolidatebleBlock with ndim 1 'take_nd' with len 1 gives wrong ndim #27785
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

[jorisvandenbossche]

cc @jbrockmendel

[jorisvandenbossche]

Alternative option is to pass self.ndim to the make_block_same_class call in Block.take_nd, but this might be more general for other locations where make_block_same_class is called without specifying ndim. I don't directly can think of a case where you don't want to preserve the ndim unless it was specified.

And eg make_block just above always uses self.ndim

[jbrockmendel]

This seems reasonable to me

[jbrockmendel]

I tried once to require that ndim be passed and found that some of the downstream tests failed because e.g. pyarrow uses make_block directly

[jorisvandenbossche]

I tried once to require that ndim be passed

That's indeed another option to pass self.ndim to all calls to self.make_block_same_class (there are not too many of them I think), if you prefer.

e.g. pyarrow uses make_block directly

But this is self.make_block_same_class which is probably OK to do?

[jbrockmendel]

But this is self.make_block_same_class which is probably OK to do?

Yah, looks like pyarrow is only calling the make_block function, not the methods.

[jreback]

thanks @jorisvandenbossche