The gaol why we need to determine what ID should be used as stanza-id,
is so we can use the stanza-id in the future for deduplication.
Case we are the sender:
Either we look for a origin-id element, which we will include in the
Future in all messages we send, or until then the ID we set on the
message as attr.
Case we are the receiver:
If our server supports mam:2 we take the ID of the result element,
because mam:2 injects the archive ID live into every message we
receive. If we dont have mam:2 we fall back to the ID of the message
attr.
- Refactor and clean up code around MamMessageReceivedEvent
- Goal is to add a GcMamMessageReceivedEvent later on
- For that cause added a raw-mam-message-received base event
This lets us attach all attributes of the base event to the
new event. Often Events trigger other Events. When that happens
we often want to keep all attr from the previous Event, and just
continue under a new Event. Until now all attr had to be pulled
out of `self.base_event` again.
_continue_connection_request_privacy() should only trigger
once after we received the discoinfo on the hostname.
We do a discoinfo on all items that the server offers,
_continue_connection_request_privacy() was triggered on every received
discoinfo again, hence requesting metacontacts and delimiters again.
also _continue_connection_request_privacy() was called on every
discoinfo error, instead of only on a hostname discoinfo error.
This bug is only triggered if the server doesnt support privacy rules.
The purpose of
`if version and not gajim.contacts.get_contacts_jid_list()`
seemed to be that when cache.db is empty (maybe it was deleted)
`gajim.contacts.get_contacts_jid_list()` should come back empty.
So on an empty roster cache, version was set to None, so that we
request in any case a new roster.
The Problem is that `gajim.contacts.get_contacts_jid_list()` is not
a good indication for an empty cache.db. On start we trigger a
`RosterReceivedEvent` which does a DB query to get the roster. Even
if that DB query comes up empty, the Event is still pushed.
In the event handler `_nec_roster_received` in roster_window.py we
add then previously open controls and our self (if the option is set)
to the roster, making `gajim.contacts.get_contacts_jid_list()` return
these contacts and hence the condition in `request_roster()` always
False.
So the version is set in the roster request, and if there is no new
version on the server, we request no new roster even though we only
have ourself and previously open controls in our roster.
As a solution for this we delete the roster version from the config
in `RosterReceivedEvent` if the DB query comes back empty, which
triggers a new roster request.
- Dont add timers when we autojoin minimized
- Remove timers when the chatcontrol gets minimized
- Send inactive chatstate when the chatcontrol gets minimized
- Add timers when the chatcontrol gets maximized
Set restore timeout inactiv, this should be really an advanced option,
probably only set for specific MUCs not as default for every MUC.
Set restore_lines to 10, so the User has an indication that Gajim can
display history, and that there is probably a setting to adjust.
Set muc_restore_lines to 100, this is a try to get a reasonable amount
of history, especially in small Groupchats. Most server will limit this
to 20-50 anyway. The limit 100 is more of a precaution if we run into a
misconfigured server. This is intended as a workaround until we can
implement MAM for Groupchats.
muc_restore_timeout set to -1 means we want no particular time set, up
to which we request messages. But we should not deliberate request
messages we already have. Hence even if muc_restore_timeout is set
to -1, we should only request up to the last message we received.
Example:
If we receive a message the last message date is now.
If we restart Gajim and join the MUC again, `time.time() - timeout`
will be most likly smaller then when we received the last message
If the timeout is an hour, this would mean that many messages are requested
from the archive which we already received.
So we always want the most current timestamp (MAX).
Because some operations use the _timeout_commit() it happens that a timeout
is active while we shutdown.
This makes sure everything is commited before shutdown.
- remove gui extension point on close, to remove the reference the plugin manager holds to the Class.
- reduce scope of var that holds reference to Gtk.Builder(). We have to be careful with
Gtk.Builder, because it can hold a reference to our Class (methods bound to signals) which creates a reference cycle
which python is not able to detect. Hence neither Gtk.Builder nor our Class are garbage collected.
Why this is not detected as a reference cylce is unclear at the moment.
There are two approaches to circumvent the problem:
1. we lose our reference to Gtk.Builder when closing the window, which lets python garbage
collect the builder, afterwards it can garbage collect our class.
2. we reduce the scope of the var that holds a reference to Gtk.Builder, so that the builder
can be garbage collected at the end of __init__.
I chose to reduce the scope because the builder is not needed class wide.
Yann Leboulanger
Philipp Hörist
SaltyBones
Yann Leboulanger
André Apitzsch