Consider different non-redundant ROR mutants

[afd]

Dredd implements the optimisations from Table III of this paper:

https://people.cs.umass.edu/~rjust/publ/non_redundant_mutants_jstvr_2014.pdf

@JonathanFoo0523 pointed out here that sometimes the choice of which mutants to keep vs. which to be avoided due to subsumption could be more nuanced.

We should revisit this.

[JonathanFoo0523]

The paper argument still holds in a op b when both a and b are unsigned int.

If one side of operator is 0 and the other side is an unsigned, we can perform a similar (and simpler) analysis to find non redundant ROR mutations.

WLOG, consider a op 0. We have 2 cases for the value of a: (1) a equal to 0; (2) a larger than 0. The outcome of a op 0 for different op is shown below:

+----+------+-----+------+-----+------+------+
|  a | a!=0 | a>0 | a>=0 | a<0 | a<=0 | a==0 |
+----+------+-----+------+-----+------+------+
|  0 |   F  |  F  |   T  |  F  |   T  |   T  |
+----+------+-----+------+-----+------+------+
| >0 |   T  |  T  |   T  |  F  |   F  |   F  |
+----+------+-----+------+-----+------+------+

Note that: (1) The minimal distance mutation for op that produced the same outcome for 2 cases of a is any mutation that produced different outcome for 2 cases of a. Therefore, the sufficient set of non-redundant mutations for these operators is {eq('!=') and eq('==')}, (2) The minimal distance mutation for op that produced different outcomes for 2 cases of a is any mutation that produced the same outcome for 2 cases of a. Therefore, the sufficient set of non-redundant mutations for these operators is {eq('true') and eq('false')},

Here. eq(op) is the set of equivalent mutations under a op 0 (eg: eq('!=') = {!=, >}).

This could allow us to do simpler check for non-redundant ROR operator for unsigned while avoiding equivalent mutations, as noted here.