[RFC2 Patch 5/5] lib: iterator API for message properties

[Daniel Kahn Gillmor]

On Tue 2016-05-31 21:12:21 -0400, David Bremner <david at tethera.net> wrote:
> I was thinking a bit about how to dump/restore these.
>
> The most upwardly compatible way that i thought of is something like
>
> #= msg-id key=val key=val
>
> i.e. duplicate the msg-id for messages with properties
>
> This would be ignored by old notmuch-restore.
>
> Otherwise, maybe something like
>
> msg-id -- +tag +tag # key=val key=val
>
> I'm not sure. this might crash old notmuch-restore.
>
> How important is backward compatibility, and how important is minimizing
> dump size? It's a bit hard to predict the things people might use
> message properties for, but for thread surgery, I would expect a small
> number of messages with properties.

The other concern is our conception of how properties are unset/removed,
right?

With tags, it's possible to include -blah to remove the tag "blah".  how
do we remove/clear/overwrite these tags?  what about using +key=val or
-key=val to set/unset certain key/value combinations, and a value-less
key= to remove all values matching a given key?

alternately:

 key=val (clears all values for "key", and sets a new value "val")
 key+=val (appends a value "val" for "key")
 key-=val (removes any "key" set to "val")
 key= (clears all values for "key"

---------

However we resolve this particular decision, it'd be nice to have a
stable, sane story about backward compatibility going forward, so that
we don't have to worry about it in the future.

For example, each dump file could start with a line like:

  #version 1

and notmuch restore would assume that without "#version n" as the first
line, it's version 0.  then notmuch restore could decline to parse dump
files of a version that it doesn't know about.

Alternately, we could have the first line be something like:

   #features config properties

and if the first line is not #features, then we assume that no features
are in place -- but if restore sees features it doesn't know about, it
can offer to proceed while warning the user that we might miss something
(or that something might break).


Thanks for working on this, David!  I think this is going to be really
useful!

    --dkg