I figured it's not too bad because we have the in-memory cache. But I suppose this disables the database cache?

I've tested removing this, but it's required for the detection to work (I tested by modifying the subscription state in this case, not by creating new folders).

I'm not sure why this seems required. When testing with outlook it clearly makes a difference (maybe I'm doing something wrong while testing...), but I failed to reproduce the same behavior in a testcase so far.

I tried testing it like this (couldn't really find a clean way to inject a change in an ongoing ping...): https://git.kolab.org/D5196
but with this the above change is indeed not necessary.

I'll have to re-test with Outlook.

This definitely makes the difference when we just change metadata with Outlook, don't know yet why.

I suppose for some reason

rcube::get_instance()->get_storage()->clear_cache('mailboxes', true);

works for you but doesn't for me. If it works the above should not be necessary after all.

I'm using $config['imap_cache'] = 'db'; here, maybe you use redis/memcache?

I use redis indeed, and it seems cache clearing is broken there (At least the request for the "*" folder always hits).

I think the problem could be that a "*" is a wildcard match in redis.

nope

The place to look at is rcube_cache::remove_record() and rcube_cache::read_record(), it's not that trivial.

This makes it work: https://git.kolab.org/D5211

But I don't follow this "add to remove" approach at all, so I'll leave it to you to translate it into something sensible.

(unless you also don't know the code, then I can look into it).