user>rmr>Keyval types and combinations

First a table to categorize keyval pairs. Keys and values fall in three categories: NULL, not-NULL but 0-length ("length" in the rmr sense, that is nrow for matrices and data frames, length otherwise) and the "normal case" (not-NULL, length greater than 0).

key/val	NULL	0-length such as integer(0)	length > 0
NULL	keyval(NULL, NULL)	keyval(NULL, NULL)	NULL keys
0-length such as integer(0)	keyval(NULL, NULL)	keyval(NULL, NULL)	NO
length > 0	NO	NO	normal case

The idea is that keys and values need to be the same length or rmr2 will try to use recycling to achieve that, and recycling fails on zero-length arguments. The only exception is NULL keys with non-zero length values, which means "keys omitted". Typical use is map only jobs, or input to the map function. When paired with NULL or 0-length values, that creates an empty keyval pair, otherwise a keyval pair with values but NULL keys. When the keys are not-NULL 0-length, values must be too and the two form an empty keyval pair. When the keys are non-zero length, the values should as well. So we have three categories of keyval pairs, 0-length (or empty), NULL keys and non-zero-length values (NULL keys hereafter), and non-zero-length everything (normal case hereafter). Can we mix and match them as argument to c.keyval and as return values from map and reduce? This is answered in the following table.

c.keyval	0-length	NULL keys	normal case
0-length	0-length	NULL keys	normal case
NULL keys		NULL keys	NO
normal case			normal case

It is symmetric and can be summarized as follow:

1. 0-length is the neutral element and can be combined with every other type
2. each type can be combined with the same type (diagonal)
3. NULL keys can not be combined with the normal case.