As I understand, the more "up-to-date" 2017 version is not as comprehensive as the 2014 version. I got in contact with Bruce Ravel (2014 author) and he says he's standing by, but I told him to wait to see what the org-mode side can do first about the :session issue. So in general I'm supposing that when a language's REPL session is not started, babel sometimes goes straight the executable on the system and comes back with results. But then others specifically need a :session named and started -- or at least start a session.

I guess I'm being manic about this due to the overall difficulty of producing graphs and diagrams in general in the STEM world. If you're good, you can transcribe JIT, say, a math lecture on your laptop with org-mode -- prose and LaTeX formulae. But if you need diagrams you're blown away. . . 

I've got a copy of Martin Weissman's An Illustrated Theory of Numbers which utilizes LaTeX Tufte with diagrams in TikZ/PGF. Quite a beautiful book. In general, it's just too damned hard to produce diagrams relative to prose and formulae. . .  My two farthings. . . .

On Tue, Apr 30, 2019 at 12:10 PM Fraga, Eric <e.fraga@ucl.ac.uk> wrote:
For the list: Lawrence and I have followed this through a bit more.  The
solution, for those that have a newer version of gnuplot-mode (2017
version instead of the 2014 version), is to set :session to "none".

I would suggest that there is a bug in ob-gnuplot.el.  Specifically,
:session is initialized to nil but all the code that checks for session
assumes that it has to be set to "none" to not use a session.

What hasn't been resolved is how to get sessions to work with the more
up-to-date gnuplot-mode.
--
Eric S Fraga via Emacs 27.0.50, Org release_9.2.3-327-g3375f0