not sure how useful this is....but say we provide dropna=True (which is the current behavior), but if dropna=False then can reindex by all the combinations of each index level as a reindexer...
This would make for a much better solution for #3820.
I think its the same op with 2 diff questions....but yes
This one however: http://stackoverflow.com/questions/16994485/ensuring-the-cartesian-product-of-keys-appears-in-a-pandas-table
vaguely related #3999
user can just use multiindex.from_product and reindex