Comment 16 for bug 337394

Ted Gould wrote:
> On Tue, 2009-04-07 at 15:14 +0000, Mark Shuttleworth wrote:
>
>> Do we want to support append-on-replace at all? I find the "Joe is
>> offline/Joe is online" append really ugly. That should be a replace -
>> period. The append only makes sense when it WOULD have been two
>> consecutive messages.
>>
>> Ted, I think this sort of confusion is a result of us not being willing
>> to define appending more explicitly in the API. Because we wanted it to
>> be implicit and "easier for the developer", we've ended up with a
>> mishmash, both on our side (we aren't even implementing our own spec
>> correctly!) and in apps.
>>
>
> I'm sorry, perhaps I missed something in the conversation. We should
> absolutely not support append on replace. Because if we do that we end
> up in the case where we can't replace notifications that are appended.
> I think that's pretty much what the original bug report stated.
>
> If you send:
>
> "message A" -> append
> "message B" -> append
>
> That might be what you want to appear as:
>
> +----------------+
> | message A |
> | message B |
> +----------------+
>
> If for some reason "message A" changes then there is no way to refer to
> it if the replace is only "message B".
>
Agreed.

So how did we end up here? I think it's because we were too lazy fairy
(;-)) when we introduced append, in the misguided idea that "making it
just happen easily" would be good for developers. But in the end,
because we did not make it a concrete explicit process, where the
developer says "append this to THAT message", we end up in the vague
world where even our own notification system is confused, and developers
more so.

Mark