Re: Asynchronous blocks for everything (was Re: [BUG] Unexpected result when evaluating python src block asynchronously [9.7-pre (release_9.6.17-1131-gc9ed03.dirty @ /home/yantar92/.emacs.d/straight/build/org/)])

[Bruno Barbier <brubar.cs@gmail.com>]

Hi Ihor,

Ihor Radchenko <yantar92@posteo.net> writes:

>
> Thanks for the code!
> It is a lot more that I expected.

Note that only the first 5 patchs are real patchs.  The remaining things
are just a demo how it could be used.  The current async (ob-comint)
depends on writing UUIDs in org files, and, that's why I couldn't use it
as a demo.  I'm thinking about dropping the patch:

   'ob-core async: Add org-babel--async tools [2/5]'

and just provide an other new keyword (feedbacks-with) so that anybody
may plug its own feedback process.

> I have many questions ;)

Thanks! They are welcome :-)


>> The new API itself is more about how to wait for and display one block
>> result.  So, it's not really aware of sessions.
>
> I think that it is important to think about sessions. For non-sessions,
> we may execute the code in parallel, disregarding already running
> execution. In contrast, for sessions, we need to maintain some queue and
> wait until previous execution finishes before running next (if multiple
> session blocks are executed in quick succession).

I agree.  My proposed patch leaves this open to further work.

My proposed patch allows to have one way, in ob-core, to display
progress of asynchronous executions: like not started, 10%, success,
failed; and a way to plug in asynchronous executions in ob-core.

I've included some demo code just to show how that API could be used.


> It may also be necessary to provide an UI listing the queued and running
> execution. Users should be able to stop "stale" processes if they are
> defunc (consider infinite loop).

That looks nice but that may be too limiting.  Like it will probably forbid
to execute something asynchronously using elisp threads.  Anyway, that
outside the intended scope of my set of patchs.


> What is more important is when users, for example, remove the whole
> subtree containing the place where the execution result is to be
> inserted. Or when users perform edits at or around that place where the
> result is to be inserted. Or consider subtree with pending result
> refiled elsewhere (different file or different place in the same file);
> or archived.

My patch definitely needs to address those, even if only by raising an
error.

> Or consider user running an src block, quickly editing it, and then
> running again. What should we do then? Should we stop the first running
> process?

To keep things simple and generic, I would just forbid the user to
reschedule the task until the previous outcome is available.


> I have several general questions:
>
> - what if append/prepend to result asynchronously?

You mean if org is executing several times the same code concurrently?
I think we should forbid it.

> - what if we replace the result, call async evaluation two times, and they arrive in opposite order (first called is calculated after the second)?

"One execution at a given time" will solve this too :-)

> - on failure, Org babel sometimes uses ~org-babel-eval-error-notify~. How will it interact with asynchronous evaluation? What if we have multiple simultaneously arriving error notifications?

In the asynchronous case, I've decided to leave the overlay in place
in case of errors, with the error description.  Clicking on that error
pops up the same kind of popups as in the synchronous case.  You can
give it a try using examples of errors my demo file:

   scratch/bba-ob-core-async/my-async-tests.org

(Please use the updated set of patchs that I've sent today, else, you
will have to fix the required libraries).


> Note that we already have a WIP an asynchronous API in the works.
> Check out `org-async-call' in
> https://code.tecosaur.net/tec/org-mode/src/branch/dev/lisp/org-macs.el#L377
> If possible, we should reuse it.

Thanks. I will have a look at it.


Bruno