Re: Lazy load of org-protocol

[Jim Porter]

On 2/6/2022 8:42 AM, Max Nikulin wrote:
> On 06/02/2022 01:27, Jim Porter wrote:
> > As a result, I think a good first step might be to add support for 
> > "--funcall" to emacsclient, just like the regular emacs binary. (The 
> > "-f" shorthand won't work though, since emacsclient already uses that 
> > for "--server-file"). This would simplify the `message-mailto' case 
> > above and would also allow org-protocol to do something similar:
> >
> >    emacsclient --funcall org-protocol-capture %u
>
> No, --funcall is just a sugar for --eval '(func)' that does not contain 
> arbitrary input, but func has no access to other arguments and it is the 
> real problem.

Oh right, of course. The emacsclient case with `(message-mailto \"%u\")' 
threw me off, but `message-mailto' works differently depending on 
whether you pass it an arg or not.

> I think, the solution is to add -arg command to emacs server protocol 
> that pushes its argument to a list and extend -exec command that would 
> make such list available as argv or as `command-line-args-left' for 
> evaluated expression. Of course, emacsclient option parser should be 
> modified as well to support --arg option
>       emacsclient --eval '(func)' --arg 1 2 3
>       emacsclient --eval '(func)' --arg -- 1 2 3
> and maybe even for multiple eval+arg pairs
>       emacsclient --eval '(f1)' --arg 'a1' --eval '(f2)' --arg 'a2' 'a3'

I think something like this could work, so long as the arguments can be 
forwarded correctly in the alternate editor case. If I understand how 
emacsclient handles --eval, I think that might work, but it would still 
require writing Lisp functions that know how to handle argv or 
`command-line-args-left'. I'm also not totally sure how safe/sensible 
that is when the arguments aren't from the invocation of emacs itself, 
but from emacsclient (and thus, the args could be updated multiple times 
throughout a session). It probably wouldn't actually break anything, but 
it does seem a bit surprising to me...

Maybe another option would be to add an --apply argument that really 
*does* consume the other command-line args and turns them into a 
properly-quoted function call. Roughly speaking, it would turn this:

  emacs --apply func foo bar baz

into this:

  (apply #'func '(foo bar baz))

This would work effectively like how I momentarily thought --funcall 
works. Then you could say:

  emacsclient --apply org-protocol-capture %u
  # or ...
  emacs --apply org-protocol-capture %u

This would also mean that the `message-mailto' function from Gnus 
doesn't need to care about `command-line-args-left' anymore, which would 
simplify it somewhat. And this could be useful in general for 
programmatically interacting with Emacs, since users would be able to 
pass arbitrary strings to Emacs Lisp functions without having to worry 
(as much) about sanitizing the strings first.

> The proper place to discuss idea is emacs-devel list, but I am afraid 
> that without a patch it will be just buried.

I don't know if an actual patch would be a strict prerequisite for 
posting to emacs-devel, but it's probably wise to have a clear, detailed 
proposal first. I think it's easier to get everyone on board with an 
idea if it's easy for them to see how all the details work up-front.

- Jim