Support rules for negated expressions not generated correctly

[tsandall]

In some cases, PE has to generate support rules for negated expressions. When it does this, the support rules must be given names. Currently PE uses the outer query ID and the expression index to produce unique names, however, this is not correct when there are multiple solutions. For example:

p { r[x]; not input[x] }
r[1]
r[2]

Assuming the query is p = true then when p is evaluated, the query ID will be 1 (the identifiers are zero-based) and the expression index of the negated expression will be 1 (the indices are also zero-based.) In this case, the current output from PE when support is generated will be something like:

p { not __not1_1__ }
p { not __not1_1__ }
__not1_1__ { input[1] }
__not1_1__ { input[2] }

The problem of course is that this output is incorrect. p should be true if not input[1] OR not input[2].

To resolve this issue we can encode the negated query ID into the support rule name, e.g:

p { not __not_1_1_3__ }
p { not __not_1_1_5__ }
__not1_1_3__ { input[1] }
__not1_1_5__ { input[2] }