mod_mix_pam: Provide MIX channels as roster entries via hook

[lnjX]

Currently the joined MIX channels of a user are stored in the database of mod_mix_pam, but are not provided to the user in any way. This change subscribes to the hook of mod_roster and provides the joined channels.

TODO:

• [x] Add roster pushes
• [x] Only provide MIX subelements when <mix-annotate/> is set
  • [x] in roster get
  • [x] in roster pushes: Requires to store annotate setting for the session (Need to find out how)
• [x] Change roster hook to use #roster_item{} instead of #roster{} so the roster record can stay how it is (without mix participant ID). This way no empty field for the mix participant id is saved in the roster mnesia db.

[coveralls]

Coverage decreased (-0.01%) to 33.588% when pulling d26f35b8d0910cf6e41e5dc85bb5e53b303e1489 on lnjX:feature/mix-pam-roster into 99d9e315a3c0524d84197e63741790d5893c51f4 on processone:master.

[lnjX]

@badlop Do you've got an idea for how to store the <mix-annotate/> setting per session? Is that easily possible without using the database?

---

Okay, used ejabberd_sm:set_user_info()

[lnjX]

:partying_face:

[badlop]

Umm, there's a problem in commit 76aa98f21e33a106c461671366af5119479e7182, tt breaks mod_shared_roster. Can you take a look?

It's easy to reproduce, for example:

1. Empty database, default configuration, start ejabberd
2. Register an account and login (admin@localhost)
3. That user adds a contact to his roster (user3@localhost)
4. mod_shared_roster immediately crashes:

2022-08-12 12:55:09.951716+02:00 [error] Hook roster_get crashed when running mod_shared_roster:get_user_roster/2:
** exception error: {badrecord,
                      {roster_item,
                          {jid,<<"user3">>,<<"localhost">>,<<>>,<<"user3">>,
                              <<"localhost">>,<<>>},
                          <<>>,[],none,subscribe,undefined}}
   in function  mod_shared_roster:'-get_user_roster/2-fun-2-'/2 (/home/bernar/e/git/ejabberd/src/mod_shared_roster.erl, line 206)
   in call from lists:mapfoldl_1/3 (lists.erl, line 1462)
   in call from mod_shared_roster:get_user_roster/2 (/home/bernar/e/git/ejabberd/src/mod_shared_roster.erl, line 204)
   in call from ejabberd_hooks:safe_apply/4 (/home/bernar/e/git/ejabberd/src/ejabberd_hooks.erl, line 320)
   in call from ejabberd_hooks:run_fold1/4 (/home/bernar/e/git/ejabberd/src/ejabberd_hooks.erl, line 301)
   in call from mod_roster:roster_version/2 (/home/bernar/e/git/ejabberd/src/mod_roster.erl, line 236)
   in call from mod_roster:push_item/3 (/home/bernar/e/git/ejabberd/src/mod_roster.erl, line 499)
   in call from mod_roster:process_subscription/6 (/home/bernar/e/git/ejabberd/src/mod_roster.erl, line 647)
** Arg 1 = [{roster_item,{jid,<<"user3">>,<<"localhost">>,<<>>,<<"user3">>,
                              <<"localhost">>,<<>>},
                         <<>>,[],none,subscribe,undefined}]
** Arg 2 = {<<"admin">>,<<"localhost">>}

[prefiks]

badlop: I think that mod_shared_roster was not recompiled after xmpp changes (i think this record is correct by current definition).

[lnjX]

@badlop I think I've found the issue, https://github.com/processone/ejabberd/pull/3882 should fix it!

[badlop]

make distclean and removing rebar.lock didn't help, as the problem is in that hook that wasn't updated to the new event arguments. #3882 solves it, thanks!