Re: [Jmap] message vs. annotation

[Ted Lemon <[e-mail]>]

On Apr 4, 2017, at 10:43 PM, Neil Jenkins <[e-mail]> wrote:
> This is the key here. While mailboxes, keywords and annotations could be viewed as conceptually the same thing and so combined into one perfectly abstract construct, I strongly believe that to do so would be a mistake. It would make it harder for client authors to understand and use, harder to implement efficiently on the server, and harder to map backwards compatibility to IMAP.

I think that the reason to have keywords as a special type of thing is so that you can define standard keywords.   Otherwise they don't seem any different than labels/mailboxes.   I think both types of object have essentially the same desired set of semantics: "give me all the things that match this label, quickly."  "tell me what things with this label have changed/appeared/disappeared."

The reason I pushed back on annotations in the meeting is that I think that annotations are actually likely to be more structured objects than simply "things I need to get done by Friday."   "Things I need to get done by friday" is not a static object: it's a verb with modifiers.   If you treat it as if it's just an annotation, I think that you have an overly limited solution to that problem.   The reason I reacted to that specific example at the microphone the way I did is that this sort of ad hoc label that mail programs currently offer is actually a great example of a missed opportunity: something that could really be a kick-ass feature instead winds up being just another way for me to do bookkeeping, and just another thing for me to remember.

If a thing is a "thing I should get done by Friday," there should be some code running that's helping me to arrange to actually get it done by Friday, not just a notice for me to read on Saturday and feel guilty about.

And an annotation at least as described during the conversation in the meeting last week would not be sufficient for this.   Sure, you could hang more stuff off it to make it work, or have something that cleverly parses these annotations and does something smart with them, but it's all way too sloppy and ad hoc to be useful.

So while I agree that everything isn't the same object, and there may be some value to having mailbox tags be different than annotation tags, I think that an annotation really ought to be a first-class object that can be linked to a message.   The link probably has to be two-way: give me all the messages with this annotation, and give me all the annotations on this message.

Of course, the response at the mic was that a message could have multiple annotations, each one with its own name, and not all downloaded to the client all the time.   I don't disagree with this.   And if indeed each annotation has a name, maybe that's enough structure.   But it doesn't feel right to me.   If there are two objects marked "get done by Friday," I think the question "list objects marked get done by Friday" is a really obvious one.