message.el and browse-url.el interaction [was: Re: BUG: notmuch-emacs-mua wrong signing MML placement]

[Tomi Ollila]

On Thu, Dec 12 2019, David Edmondson wrote:

> On Wednesday, 2019-12-11 at 13:55:16 -05, Robbie Harwood wrote:
>
>> My emacs config contains `(add-hook 'message-setup-hook
>> 'mml-secure-sign-pgpmime)` as recommended in the emacstips
>> https://notmuchmail.org/emacstips/#index14h2
>>
>> However, when I run (something like):
>>
>>     notmuch-emacs-mua 'mailto:testto?body=testbody'
>>
>> I get an email in which the requested body contents appear *before* the
>> signing MML.  If I understand correctly how this all fits together, that
>> won't do the right thing - but at any rate, it's unexpected.
>
> This is an unfortunate interaction between message.el and browse-url.el
> and isn't specific to notmuch or notmuch-emacs-mua.
>
> You can see the same behaviour in “emacs -Q -q” if you execute:
>
> (progn
>  (add-hook 'message-setup-hook 'mml-secure-sign-pgpmime)
>  (browse-url-mail "mailto:fred?body=testbody"))
>
> Given that, it's not something that we should be fixing in notmuch
> itself, as it's an emacs bug.
>
> The problem is that `browse-url-mail' always inserts the body after the
> header separator, which in this case happens after
> `mml-secure-sign-pgpmime' has already inserted the MML header.
>
> Having `browse-url-mail' insert the body at the end of the message would
> work for this case, but would presumably make a mess of situations where
> someone inserts their signature block via `message-setup-hook'.
>
> Perhaps `browse-url-mail' should skip past any MML headers before
> inserting the body? That's a bit messy, as browse-url.el currently
> doesn't know anything about MML.

I've never found `browse-url-mail' adequate for handling mailto: links,
so couple of years ago I wrote mailto: "wrapper" of my own (*). It,
among other things, spits the following elisp for emacs to swallow:

(with-demoted-errors
(require 'notmuch)
(notmuch-mua-mail "fred" nil nil nil
                  (notmuch-mua-get-switch-function))
 (cd "/home/too/vc/projects/nottoomuch")
 (message-goto-body) (insert "testbody")
 (goto-char (point-max))
 (if (/= (point) (line-beginning-position))
    (newline))
 (set-buffer-modified-p nil) (message-goto-to))

Surprise! this has exactly the same problems as David explains above; I
could add code to check for this [#part sign=pgpmime] (**) and skip it if
it exists (and, as first step, not look for anything else to skip). That
is just so hacky and error prone I would not like to do so (at least as
a general solution)...

>
> The nicest fix would perhaps be to have `compose-mail' accept the
> initial body as a parameter, which would eventually flow into the
> relevant code in message.el which could do the right thing (which I
> think would just be to skip past MML headers before inserting the body).

This suggestion of an optional parameter to take initial body sounds
very good solution (as first sight =D), so if I don't come up with
any other suggestions +1 for that..

>
> Any advice from emacs-devel on how to move forward? I'm happy to produce
> a patch for whatever solution is considered most appropriate.
>
> dme.
> -- 
> If the girls start moving the boys will start moving.

Tomi

(*) https://github.com/domo141/nottoomuch/blob/master/nottoomuch-emacs-mailto.rst

(**) replace [...] with <...>