[O] Using buffers instead of string when processing and passing org-babel results

[Bjarte Johansen]

Hi,

I have been wondering: Is there a reason org-babel passes around results as 
strings instead of buffers? I understand there will be more to keep track of 
and right now there is a great chance of breaking code (org-modes code and 
other people who depend on the current behaviour).

However, I just did an elementary test with the following

#+BEGIN_SRC bash :results verbatim
  cat file.txt
#+END_SRC

I created the file with the command `base64 /dev/urandom | head -c 100M > 
file.txt`.

If I execute the script in the normal way it takes more then half a minute to 
complete. If I on the other hand convert every point on the path that expect it 
to be a string to be a buffer instead it takes less than 5 seconds to complete 
the script and insert the result into the buffer.

I understand that it might not be normal to insert these big files into the 
org-buffer, but the problem is also there if you want to silence the result and 
pass it from script to script. The creation of big strings is really slow.

I also feel that buffers are more natural vessels for the results. They are 
easier to manipulate than strings in Emacs and are much much faster. A big 
problem however is how do we handle the buffers that are created? They won’t be 
garage collected, so we need a way to safely kill them when we no longer need 
them.

Is this something that org-mode is interested in? I could try to work on it, 
but if org-mode isn’t interested in the change I am not sure I will bother.

Regards,
Bjarte