See discussion in bug 1615981.

I think we should add a migration to remove all the chat accounts that are dead since a long time.

Any reason not to do this?

I'm not 100% sure what this means, to be honest. I think by it you mean fully remove the code for account types we've already "deactivated" (e.g. Facebook)?

As I mentioned in the other bug, it would be interesting to know if people still have these accounts as active at all.

One benefit to keep this code is that people still know which account was which easily, which could be useful for e.g. looking at logs. We should ensure this doesn't break if we remove any of this code.

As far as I remember, we kept them so that it would be possible to see logs. I don't remember the details though. Are we removing the logs from disk when deleting a chat account?

It's not clear to me what value it would have to see that a user had a Facebook account setup, "active" or not. It's pretty clear it's dead since many years with no chance of coming back. It could only show many people have facebook, some of them may have set up the facebook chat account at the time. What decisions could be made from data like this?

But yes, the intention would be to fully remove deactivated account types.
I agree logs should be kept.

(In reply to Magnus Melin [:mkmelin] from comment #4)

It's not clear to me what value it would have to see that a user had a Facebook account setup, "active" or not. It's pretty clear it's dead since many years with no chance of coming back. It could only show many people have facebook, some of them may have set up the facebook chat account at the time. What decisions could be made from data like this?

Then I would argue -- what value is that telemetry at all? For any type of account.

It paints a vague picture of our users and lists how many accounts users have usually. I'm not sure it's necessarily basis for any big decisions, though who knows.
The difference to the dead im account types is that even if, say, unlikely, 90% of our users had connected Facebook, there's really nothing we could do about that. If we find, say, nntp is unused, that could have implications for the future.

Khushil, can you investigate, and see how this could happen but still give access to logs? Do we need a special "defunct" account type?

(In reply to Magnus Melin [:mkmelin] from comment #7)

Khushil, can you investigate, and see how this could happen but still give access to logs? Do we need a special "defunct" account type?

A special "defunct" account type is pretty much what we have today.

Right now, we have Yahoo, Twitter, and Facebook account types that are not working and showing only old logs. We have Gtalk, IRC, XMPP, Odnoklassniki, and Matrix in working condition.
So now, how do we want to tackle the accounts which are not working? It's already in the condition where they just show logs.

I'm not sure, that's what we need to figure out in this bug.
The goal should be that no future code changes would need to care in any way for these old accounts. As time goes by it will get more and more difficult to test them in any way.

The suggestion to use some kind of defunct account type was that they would perhaps be removed as real accounts but kept under this "generic defunct" account.

(In reply to Magnus Melin [:mkmelin] from comment #10)

The goal should be that no future code changes would need to care in any way for these old accounts. As time goes by it will get more and more difficult to test them in any way.

This is pretty much the current state.

The suggestion to use some kind of defunct account type was that they would perhaps be removed as real accounts but kept under this "generic defunct" account.

There's already an "Unknown account" type as well, although I don't remember if you can access logs from it.

I really don't think there's work to do here.

(In reply to Patrick Cloke [:clokep] from comment #11)

I really don't think there's work to do here.

I agree with Patrick.

Closing then.
If we ever go touching the icons, we may want to have them all use a generic icon instead.