Re: [O] python sessions

[Andreas Röhler]

Am 26.03.2013 22:41, schrieb Gary Oberbrunner:
> I did some further testing.  With my patch, my real org-mode python file is
> now actually working.  There were a few gotchas I didn't understand about
> session mode and python and matplotlib:
>
> 1. In session mode, you are essentially running an interactive python.
>   (This is made explicit by my patch, which passes -i).
> 2. In interactive mode, blank lines are special: they indicate the end of
> an indented block.  So you have to write your org-mode python a little
> differently when using session mode.
> 3. matplotlib uses an "interactive" backend when started from an
> interactive python (sensibly).  So you have to set the backend explicitly
> when using session mode, or it hangs forever with no indication what's
> wrong.
> 4. ob-python does not handle python errors at all in session mode: it
> ignores them silently.  This makes it really hard to debug the python code.
>
> I'd like to document #s 1-3; where should that go?
> I'd like to work on a patch for #4; it seems like the only thing is to
> search in the session buffer for likely error strings.  Is there any better
> method anyone can think of?

AFAIU the error string is missing, because it's sent the wrong 
path-to-interpreter,
i.e. using shell-command.
Shell returns 0 or 1, not the Python errors.

Also the interactive spec is related. Usually Python shell as started from 
py-shell or run-python include this already.
When send from a python-mode, also the empty lines you mentioned and related 
stuff is treated.

Andreas