Re: [Jmap] Address Validation (was Re: Submission)

[Paul Smith <[e-mail]>]

On 25/04/2017 10:26, Adrien de Croy wrote:
>
> I think the problem with lack of support for MDN / DSN in clients may 
> be due to lack of implementation in the network of support for it.
>
> If there's still a significant number of deployed systems that don't 
> support it, then clients have increased doubts about its worth. 
>  Chicken and egg issue again.
>
> I wouldn't presume that the reason that it doesn't appear so much in 
> the wild in clients is due to lack of interest.  Maybe some MUA 
> authors could be approached for comment.
>
> We looked several times at supporting DSN etc in our mail system, and 
> balked at some of the more onerous requirements.

DSN wasn't that hard when we implemented it (many years ago so I've 
forgotten details). It wasn't trivial either (we had to change the way 
envelope data was stored), but there are some things which have been far 
more onerous (it's a 'separate' thing, so potential breaking effects in 
other code are limited - unlike full 8 bit support which affects pretty 
much everything).

But - if DSN is hard to implement, then any 'delivery status' type thing 
could be similarly hard so may affect JMAP implementation if delivery 
status is a mandatory part of that standard.


> A lot of the extension RFCs have onerous requirements for interfacing 
> with non-supporting systems.  8BITMIME springs to mind here.  So we 
> are binary safe (even embedded NULLs), but don't advertise 8BITMIME 
> due to not wanting to take on the responsibility of what to do when 
> forwarding to another server that doesn't advertise it.  Maybe that 
> just makes us lame.  Probably should revisit both these.

We support 8 bit, but not embedded NULLs. Changing that would be a big 
thing (probably not to implement in the message store, but to just to 
test (and write new tests for) EVERYTHING that looks at the data). We've 
thought about it, but since we probably wouldn't advertise 8BITMIME 
anyway, for the reasons you say, it's easier just not to potentially 
break things by supporting NULL characters ;-)

To me, having to rewrite messages is a really bad idea (DKIM etc), so 
8BITMIME just isn't worth doing now. If we were designing SMTP nowadays, 
obviously full 8 bit support would be mandatory, but we're not.