Re: [PATCH v1 0/3] Address completion entirely in elisp.

[David Edmondson <dme@dme.org>]

On Fri, Sep 05 2014, Mark Walters wrote:
> On Fri, 05 Sep 2014, David Edmondson <dme@dme.org> wrote:
>> Address completion entirely in elisp.
>>
>> I grew frustrated with having to use an external command to provide
>> address completion, as they all had annoyances (up front scanning,
>> requiring python bindings, etc.). This is an attempt to provide
>> something similar to jkr's notmuch-addresses.py (which I was
>> previously using) entirely in elisp, relying only on the `notmuch'
>> command.
>
> Just a few quick comments: the first is relevant to others trying this
> patch.
>
> 1) You seem to be missing a (require 'std11) somewhere. I did this via M-:
> and then it ran fine.

My apologies. Will fix (and a compiler warning at the same time).

> 2) It is not quick on a spinning rust disk. This may not be relevant as
> the delay is probably notmuch so would also be the case if I were using
> notmuch-addresses.py (i normally just use a trivial script that parses
> my .mailrc)

It's not always as fast as I would like on SSD either :-) The mechanism
is very similar to the equivalent Python program, so I think that it's
probably about the same.

> 3) Have you tried
> id:1407771091-12651-1-git-send-email-sojkam1@fel.cvut.cz and do you have
> any comments on the comparison?

No, I will dig it out and look.

> 4) Finally, I wonder if we would be worth approaching the backend
> notmuch use slightly differently: if we added a
> notmuch_messages_collect_from function which was very similar to
> notmuch_messages_collect_tags, and added a corresponding --output=from
> to notmuch search then you would get the information you need very
> quickly. I think it might be a lot faster as I think the from header is
> stored in the database but some other headers are not, so that the
> current method the show --body=false needs to look at the actually
> messages

Extending notmuch to help with this was next on my list of things to
do. At the moment I just needed a solution that worked.

> I should emphasise that none of the above means I am opposed to the
> patch: having respectable built in address-completion support would be
> very nice.

Cool, thanks!