Unfortunately, changing from the current synchronous to asynchronous
processing probably requires changes to the API that would require changes
to every existing language mode.

A better way forward may simply be to implement a new block type and then
let people gradually convert their language bindings to that. I'll have a
look at it for Python.

Tom

On Thu, Nov 19, 2015 at 5:38 AM, John Kitchin <jkitchin@andrew.cmu.edu>
wrote:

> Your suggestions sounds possible to me. If you are up for it, I suggest
> trying to implement it, and offering it as a patch.
>
> Tom writes:
>
> > On Wed, Nov 18, 2015 at 2:55 PM, John Kitchin <jkitchin@andrew.cmu.edu>
> > wrote:
> >
> >> I am pretty sure this is not directly possible right now.
> >>
> >> Some approaches that resemble it could be:
> >> 1. write a src block that will be tangled to a script.
> >> 2. tangle the block
> >> 3. Run the script in a shell src block with an & so it runs
> >> non-blocking.
> >>
> >> or, use an elisp block like:
> >>
> >> (org-babel-tangle)
> >> (async-shell-command "your script" some-output-buffer)
> >>
> >> I don't know a way to get continuous updated output in an org-buffer
> >> though.
> >>
> >
> > Thanks for the response. I didn't necessarily expect continuous output
> into
> > the org-buffer itself to work, but I don't see why the Python subprocess
> > can't display output as it occurs. After all, it uses comint, and comint
> > certainly has facilities for collecting output incrementally while still
> > displaying it (cf comint-output-filter-functions).
> >
> > It looks to me like the problem is that org-babel-comint-with-output
> uses a
> > "while" loop to collect process output (ob-comint.el:92). At least, it
> > could insert the output into the subprocess buffer and make redisplay
> > happen.
> >
> > But I'm not sure why the code is written that way anyway. Long running
> > "while" loops in Emacs code don't seem like a good idea to begin with.
> > Wouldn't the more natural way for this code to be written in Emacs be the
> > following?
> >
> > - an output filter gets added to the subprocess that collects output
> > - the code is sent to the subprocess for execution
> > - the command returns
> > - the output filter inserts any data it gets into the subprocess buffer,
> > into its "results" data structure, perhaps even into the org-buffer
> > - when the output filter gets the eoe-indicator, it removes itself from
> the
> > output filter list and sends a notification that execution has completed
> >
> > If the user schedules a second block for execution, the simplest thing to
> > do is return an error if there is already a block executing for that
> > subprocess; alternatively, it could be queued somewhere.
> >
> > Thanks,
> > Tom
>
> --
> Professor John Kitchin
> Doherty Hall A207F
> Department of Chemical Engineering
> Carnegie Mellon University
> Pittsburgh, PA 15213
> 412-268-7803
> @johnkitchin
> http://kitchingroup.cheme.cmu.edu
>