Um, no, this was wrong (or, at least, a change):

From our code:

		/* BOLT #7:
		 *   - if a channel's latest `channel_update`s `timestamp` is
		 *     older than two weeks (1209600 seconds):
		 *    - MAY prune the channel.
		 */
		/* This is a fancy way of saying "both ends must refresh!" */
		if (!is_halfchan_defined(&chan->half[0])
		    || chan->half[0].bcast.timestamp < highwater
		    || !is_halfchan_defined(&chan->half[1])
		    || chan->half[1].bcast.timestamp < highwater) {
			status_debug(
			    "Pruning channel %s from network view (ages %"PRIu64" and %"PRIu64"s)",
			    type_to_string(tmpctx, struct short_channel_id,

i.e. one side can't keep a channel alive unilaterally, both sides must send every 2 weeks.

I see what you mean, and our code does this as well, but I thought the confusion fixed here was a different one, let me try to explain. Let's say we have a channel Alice <-> Bob.

Alice creates a first channel update, channel_update_ab_1 then another one, channel_update_ab_2. The "oldest" channel update could be understood as being channel_update_ab_1, whereas we only want to consider the latest channel_update_ab_2. That's the confusion I thought we were fixing here.

I think we should clarify it further by saying that we consider the latest channel update in each direction, and if the oldest of those two is older than two weeks, ...

I'll open a PR.