invert sense of ? with compound assignment

[rhendric]

The code a ?= b is basically a? || a = b, which is all well and good; it makes sense to want to assign to a if it's nullish.

However, prior to this commit, the code a ?+= b was basically a? || a += b, which is useless. If a is nullish, I certainly don't want to add anything to it.

Contrast with accessignment, where a?=b is basically a? && a.=b. Obviously better this way.

This commit makes a ?_= b mean a? && a _= b, as with accessignment, for any compound assignment operator. (a ?= b and a ?:= b are unchanged.)

Technically a breaking change, but I don't see how the feature would have been put to any good use before, so I'm treating this as a minor bug fix and waiting one week before merging on June 26, unless, as always, there are objections.

[vendethiel]

Which case does this serve that &&_= doesn't? Maybe we should just forbid it.

[rhendric]

? and && differ on falsy-but-not-nullish values (0 and the empty string). So

actor.health ?+= 10

increments actor’s health if it exists (even if it's 0), whereas

actor.health &&+= 10

only does so if actor has a nonzero health.

I feel like there are at least as many, if not more, use cases for the former as for the latter.

Again, I'm shooting for consistency with what accessignment does: compare actor.health?=add 10, if health for some reason is an object with an add method instead of a primitive number. Forbidding this with operators but permitting it with accessignment doesn't make a lot of sense to me, though I'm open to arguments that, e.g., the operator version is more confusing for some reason.

[vendethiel]

I'm not too happy on a ?= b being different from a?=b either, tbh. I know why it's different, but I think requiring an explicit a?.=b isn't too bad.

[rhendric]

That seems like a completely independent question—or am I missing your point? Even with the explicit dot, a ?.= b does nothing if a is nullish, and I'm making the case that a ?+= b should similarly do nothing if a is nullish—basically, compound assignment operators are more like accessignment then they are like simple assignment, in that they both read and write the LHS. Any further concerns about that?

[vendethiel]

No, that's a completely independent point indeed.
I'm fine with this PR as-is, I only have doubts that it won't actually be used.