Pasting commands into iESS[julia] buffers might fail

[frederic-santos]

Hi,

This might be a question rather than a bug report. However, there is a strange behavior from iESS[julia] buffers, that I don't manage to explain. Here is a minimal reproducible example.

1. Open a terminal (so, don't use Emacs!), run julia, and paste the following code into the Julia REPL:

   a8 = repeat(collect(4:-1:1),
           inner=[1],
           outer=[2])

   Everything should be okay, and your array should be displayed normally. (For Emacs: this will also work if julia is started from eshell rather than within the ESS framework.)

2. Now, open Emacs, do M-x julia, paste and submit the same piece of code into the iESS[julia] buffer. Result: the Julia process hangs forever and no result is displayed. Basically any command written on several lines will result in a frozen process. This is even stranger, because the same command can perfectly be executed when doing a C-c C-c on an ESS[julia] source code buffer.

This is not harmless, because I'm writing a Julia backend for org-mode based on ESS, and this behavior makes simply impossible to evaluate some Julia code blocks into iESS[julia] buffers with org-babel-comint-with-output.

Thanks!

(P.-S.: I'm on Linux, with the latest melpa version of ESS.)

[frederic-santos]

Well, if this can help:

When one single Julia command is separated with newlines, the iESS process reads the first (incomplete) line, and then (ess-process-get 'busy) is now t. At this stage, it does not even try to read the following lines. It seems that there is something like an ess-wait-for-process that is triggered after the first incomplete line, and it obviously hangs the whole process since the rest of the command is not passed to the iESS buffer. I couldn't find where is the problem, since there is no such instruction in ess-julia.el. However, this problem does not exist in iESS[R] buffers, so there is really something wrong specifically with the Julia part.

Here is a concrete illustration. That piece of code does not work:

(julia)
(with-current-buffer "*julia*"
  (insert "a8 = repeat(collect(4:-1:1),\n        inner=[1],\n        outer=[2])")
  (inferior-ess-send-input)) ; hangs

For my ob-julia backend, I wrote a piece of code that concatenates Julia commands into one single line, so that they can be executed in iESS[julia] buffers. Now this kind of thing works within my backend: actually, removing the newlines is enough to make things work:

(with-current-buffer "*julia*"
  (insert "a8 = repeat(collect(4:-1:1), inner=[1], outer=[2])")
  (inferior-ess-send-input)) ; okay, works

Don't know whether you can (or want to) change that directly in ESS. It's not a blocking bug for me any longer, but it's stil quite a strange behavior.

However, there is no such problem with R:

(R)
(with-current-buffer "*R*"
  (insert "d = data.frame(c(1, 2, 3),\n c(4, 7, 8),\n c(9, 10, 15))")
  (inferior-ess-send-input)) ; okay, works

[lionel-]

Thanks for investigating. I think the issue is that julia doesn't define a secondary prompt: https://github.com/emacs-ess/ESS/blob/a694b2627992bda5489c1b4b5bb750c590aa8d85/lisp/ess-julia.el#L290

This makes it impossible to implement busy detection for that language.

Possibly we should define the secondary prompt as 7 spaces, i.e. the number of signs in julia>. But ESS should also handle the case of no secondary prompt gracefully.

[tecosaur]

I'm interested in ob-ess-julia replacing ob-julia, but it would be a real pain to do this while multi-line code blocks aren't supported in sessions. Is there a good chance this could be implemented in the near future?

[frederic-santos]

Thanks @tecosaur . Indeed, if we can find a way to allow for busy detection in Julia and/or defining a secondary prompt, this will obviously help to achieve a correct ob-julia. I think I can do without that, but clearly, this will be much more painful, and will require some dirty workarounds.