Is the plan to push on with the [array]-removal? I'm guessing it's a reasonably long term goal - there are still quite a few occurrences, representing a fair chunk of work.
From the Thunderbird point-of-view (Bug 1562158), I'm particularly interested in any upcoming plans there might be for converting nsICollation. We've got a few interfaces where collation keys are exposed, and converting them to Array[T] would mean a bunch of unnecessary extra malloc/copy/free conversions when talking to nsICollation. So I'm trying to decide if I should hold off on converting those interfaces, or just bite the bullet and add conversion code.

The plan is to generally push on with it, but I had run into some complicated cases that I wasn't sure how to best handle where data is being handed around and in Places, and was not worrying too much about it until Thunderbird got further along in its removals, since we can't remove array until that happens.

As far as nsICollation goes, I just looked and it looks like converting it on the Gecko side would not be too hard; there's really only one consumer. I assume such a conversion would break Thunderbird, but maybe it makes sense for me to write the Gecko-side bit, then you write the Thunderbird bits on top of it, then once it's all ready I land the Gecko part and you simul-land the Thunderbird part?

That'd be great! From an OCD point of view it'd be very satisfying to clean up all the occurrences on the TB side :-)

I filed bug 1588260 on nsICollation.