Ihor Radchenko <yanta...@posteo.net> writes: > Thanks! > Attaching the two patches combined with some fixed to my patch. > > Please check if these two patches solve the discussed bug.
The original bug for async sessions looks fixed. But I encountered a complication regarding the non-async bug of changing :dir. In particular the following example: #+begin_src python :dir otherdir :session pysession :return figname :results file value :mkdirp yes import matplotlib.pyplot as plt plt.figure(figsize=(1, 1)) plt.plot([1, 2]) figname = 'fig.svg' plt.savefig(figname) #+end_src #+RESULTS: [[file:otherdir/fig.svg]] #+begin_src python :dir otherdir2 :session pysession :return figname :results file value :mkdirp yes import matplotlib.pyplot as plt plt.figure(figsize=(1, 1)) plt.plot([1, 2]) figname = 'fig5.svg' plt.savefig(figname) #+end_src #+RESULTS: [[file:otherdir2/fig5.svg]] As you can see the second result points to the wrong directory. However, if replacing ":session pysession" with ":session *pysession*", then it works. It's because ob-python starts the session in buffer "*pysession*" (it adds earmuffs around the session name when missing). So the following doesn't find the inferior Python: > + ((when-let ((session (cdr (assq :session params)))) > + (when (org-babel-comint-buffer-livep session) Honestly, I'm not sure why ob-python insists on the session buffer having the earmuffs, that behavior is from before my time. In particular, I'm not yet sure if the motivation is just cosmetic (because of the general convention like *shell*, *Python*, *R*, etc), or whether it actually affects python.el behavior. But if it's just cosmetic, then we might consider removing this behavior to simplify things. Though note that ob-python is not the only one with this sort of behavior -- looks like ob-lua might behave similarly, and until recently ob-R also had some surprising behavior depending on whether the session name had earmuffs.
