Review Request 112987: Use nco:contactUID instead of nco:imID to identify a contact [part 2/2]

[Daniele E. Domenichelli]

> On Sept. 29, 2013, 3:24 a.m., David Edmundson wrote:
> > KTp/im-persons-data-source.cpp, line 84
> > <http://git.reviewboard.kde.org/r/112987/diff/1/?file=193067#file193067line84>
> >
> >     this can crash.
> >     
> >     if the account is offline, accountForContact won't work, at best it will return a null accountptr, which will then crash.
> >     
> >     This is important as it is called from onContactInvalidated.
> 
> Daniele E. Domenichelli wrote:
>     I'm not sure what to return here when the account is null... something like ktp://invalid_account#<contact_id> would work?
>     Can the contact be invalid as well?
> 
> David Edmundson wrote:
>     no, you _need_ it to return the correct account ID otherwise you won't delete the contact from the hash map correctly.
>     
>     It will return the wrong account because the code for matching contacts --> account is 
>     
>     foreach(account, allAccounts) {
>       if (account->connection == contact->contactManager->connection) {
>         return account;
>       }
>     }
>     
>     This works whilst you're online, but as soon as you're offline both of those are null. 
>     (EDIT: from the above snippit it makes it sound like that will work, but we have a check for nulls and just return a null AccountPtr)
>     
>     It's pretty terrible.
>     I guess we need to rethink how we do that.
>     
>     Long term a connection->account() method in TpQt will be ideal. 
>     
>     Short term, we either need to not do this call here, or we need a better hack for doing accountForContact. Maybe a hashmap inside GlobalContactManager?
> 
> Daniele E. Domenichelli wrote:
>     Can we save the account unique identifier inside KTp::Contact? Or is the contact going to be null as well?
>     If that's the case, how was that not crashing before?
> 
> David Edmundson wrote:
>     It didn't crash because it only used the contactId before; 
>     accountForContact was only used for channel starting which has guards around null accounts.
>     
>     We could store the account ID in the contact, the question is finding a neat place to set it when the contact is created. 
>     I don't want to have a public member in KTp::Contact that is only valid half the time, depending on whether the contact happens to have gone through GlobalcontactManager or not.
> 
> Daniele E. Domenichelli wrote:
>     Something like   KTp::Contact::Contact() -> manager->connection()->objectPath().remove(TP_QT_CONNECTION_OBJECT_PATH_BASE) could work?
> 
> David Edmundson wrote:
>     \o/ YES!
>     
>     I like it.
>     It still needs a minor tweak, but we can use it. 
>     
>     IIRC contact only has a weakptr to contactManager and vice versa (or maybe it's contactManager->connection) so it's _possible_ to get null here. 
>     You'll have to check, I seem to remember it from somewhere.
>     
>     So either: 
>      we keep a reference to the connection somewhere 
>     or
>      we set the accountID (or even a weakptr to the account) in the Contact constructor.

Hmm /o\ NO

I just realized that's a different string for gabble... I think the xmpp resource name is appended after a "_2f" :(
We could check for that and cut the string, but I'm not sure if it is safe...

An alternative is to use everywhere the _connection_ ID instead of the _account_ ID


- Daniele E.


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
http://git.reviewboard.kde.org/r/112987/#review40977
-----------------------------------------------------------


On Sept. 29, 2013, 12:35 a.m., Daniele E. Domenichelli wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> http://git.reviewboard.kde.org/r/112987/
> -----------------------------------------------------------
> 
> (Updated Sept. 29, 2013, 12:35 a.m.)
> 
> 
> Review request for Telepathy, David Edmundson and Martin Klapetek.
> 
> 
> Repository: ktp-common-internals
> 
> 
> Description
> -------
> 
> Use nco:contactUID instead of nco:imID to identify a contact
> 
> NOTE: Changes in this review request https://git.reviewboard.kde.org/r/112970/ are required for this to work
> 
> 
> Diffs
> -----
> 
>   KTp/im-persons-data-source.cpp ebd01bae2f029ba5335c99babedec31b88cd7620 
>   KTp/Models/kpeopletranslationproxy.cpp a9bce378271006a9003c7757e49ba5b76ea97223 
> 
> Diff: http://git.reviewboard.kde.org/r/112987/diff/
> 
> 
> Testing
> -------
> 
> Tested with contact list, changed presence a few times.
> No more duplicates, presence works as expected
> 
> 
> Thanks,
> 
> Daniele E. Domenichelli
> 
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kde-telepathy/attachments/20130930/1fc994e2/attachment-0001.html>