Use [^...] rather than '[...] (neg)' to print NCLASS

[NWilson]

This is a cosmetic change in the tests. It looks much better.

I've also fixed a highly confusing comment in printint.c which suggests that the character map for XCLASS behaves differently to NCLASS (it doesn't now - perhaps it did previously).

[zherczeg]

It looks better, but we loose the "nclass" information from the output. Could we keep it some way? (Note: classes can contain ^ as a letter).

[zherczeg]

  re> /[\^a]/B
------------------------------------------------------------------
        Bra
        [^a]
        Ket
        End
------------------------------------------------------------------

[NWilson]

Well-spotted. This is an existing bug for XCLASS printing - and worse, even backslash isn't escaped currently!

I've fixed those cases too (and added tests).

[zherczeg]

I think [] printing has never been intended to be accurate, just debug data. Since eclasses are introduced, I would prefer ![] more for negation, since we could use it for [^a] (OP_NOTCHAR) as well. This way it would be obvious that "something happens behind" even for a new contributor. This is just my thoughts. What others think?

[NWilson]

You're saying that, because we use "[^a]" to represent OP_NOT, you want the output for an OP_NCLASS to be distinguishable. That makes sense. And we're using the "!" notation in eclasses, so that's a consistent notation to use for OP_NCLASS.

I am half-way through the optimisation work. You might be happy to know that the OP_ECLASS_NOT code will disappear when I'm finished (it will always be constant-folded in to atomic classes within the eclass).

[zherczeg]

Yes. The purpose of /B is debugging, and less constructs have the same text representation is the better.

Removing OP_ECLASS_NOT is good news indeed, nice work! How do you encode [^[\pC]&&[\pL]] without it?

[NWilson]

How do you encode [^[\pC]&&[\pL]] without it?

We know that !(A && B) = !A || !B. So we can distribute the NOT over the AND and OR (and SUB and XOR), and flip the operands and flip the operators.

[NWilson]

Actually... I think I'd rather keep it so that NCLASS prints as [^foo], and change it so that OP_NOT prints as [^x] (not). If we want to make them look separate, then OP_NCLASS should use ordinary class notation, and should be OP_NOT which changes to make it distinguishable.

[zherczeg]

LGTM