Re: python sessions

[Ivan Andrus <darthandrus@gmail.com>]

On Mar 25, 2013, at 11:27 AM, Andreas Röhler <andreas.roehler@easy-emacs.de> wrote:
> Am 25.03.2013 17:43, schrieb Eric Schulte:
>> John Hendy <jw.hendy@gmail.com> writes:
>> 
>>> On Mon, Mar 25, 2013 at 11:01 AM, Ista Zahn <istazahn@gmail.com> wrote:
>>>> On Mon, Mar 25, 2013 at 11:40 AM, Eric Schulte <schulte.eric@gmail.com> wrote:
>>>>> John Hendy <jw.hendy@gmail.com> writes:
>>>>> 
>>>>>> On Sun, Mar 24, 2013 at 9:38 PM, Nick Dokos <nicholas.dokos@hp.com> wrote:
>>>>>>> Eric Schulte <schulte.eric@gmail.com> wrote:
>>>>>>> 
>>>>>>>>> 
>>>>>>>>> From participating in evaluating code throughout the discussion and
>>>>>>>>> catching the comments throughout, I'd say yes, at least in terms of
>>>>>>>>> how other babel languages function. In other words =#+begin_src R
>>>>>>>>> :session foo= creates an R session named "foo" whereas doing the same
>>>>>>>>> with =python= instead of =R= does not yield a named session.
>>>>>>>>> 
>>>>>>>>> From what others experienced, however, the functionality was working
>>>>>>>>> correctly (results were persistent across blocks and two differently
>>>>>>>>> names blocks created two different sessions), just not named
>>>>>>>>> correctly.
>>>>>>>>> 
>>>>>>>> 
>>>>>>>> See the cond form starting at line 169 in ob-python.el.  Different
>>>>>>>> session functionality is used based on the `org-babel-python-mode'
>>>>>>>> variable, and on the version of Emacs in use (prior to 24.1 or not).
>>>>>>>> 
>>>>>>>> The branch taken when `org-babel-python-mode' equals 'python is
>>>>>>>> certainly broken, as it never saves the name of the newly created
>>>>>>>> buffer, so session re-use and use of multiple named sessions probably
>>>>>>>> works only when `org-babel-python-mode' equals 'python-mode.
>>>>>>>> 
>>>>>>> 
>>>>>>> That's me: org-babel-python-mode's value is python, so it's no wonder
>>>>>>> it's broken given what Eric says. I'm on emacs 24.3.50 where there is
>>>>>>> python.el but no python-mode.el. I tried the "cheap" workaround of
>>>>>>> switching the value to python-mode, but that does a (require
>>>>>>> 'python-mode) somewhere, so that option is out as well.
>>>>>> 
>>>>>> I'm on Emacs 24.3.1 and have no python-mode.el, either (only
>>>>>> python.el). My setup is working correctly (again, with the caveat of
>>>>>> not having named sessions).
>>>>>> 
>>>>> 
>>>>> It sounds like we have the same setup, and the following un-named
>>>>> session example does not work for me.  The first code block evaluates
>>>>> successfully, but it doesn't appear to be having any impact on the
>>>>> default session (e.g., in the *Python* buffer).
>>>>> 
>>>>>    Returns the value of x as expected.
>>>>> 
>>>>>    #+begin_src python :session
>>>>>      x = 1
>>>>>      return x
>>>>>    #+end_src
>>>>> 
>>>>>    #+RESULTS:
>>>>>    : 1
>>>>> 
>>>>>    #+begin_src python :session
>>>>>      return x
>>>>>    #+end_src
>>>>> 
>>>>>    #+RESULTS:
>>>>> 
>>>>> The second code block /should/ have access to the x variable defined
>>>>> previous, but instead it throws an error because x is undefined.
>>>>> 
>>>>> Currently I'd say session support for python is completely broken.
>>>> 
>>>> As of this morning I've joined the "it does not work" crowd. Python
>>>> sessions worked for me last week, but are now completely broken for me
>>>> in the way Eric and others describe.
>>> 
>>> Interesting... checked out back to that commit
>>> (eff59a15d76647ce8282626b9eb463dc3706d56e) and it still doesn't work.
>>> On a whim, I checked my pacman log (Arch's install system) and
>>> coincidentally on Mar 20 /after/ I wrote that post in which things
>>> work, I ran a system package update.
>>> 
>>> $ grep -i emacs /var/log/pacman.log
>>> 
>>> [2013-03-20 12:51] upgraded emacs (24.2-4 -> 24.3-1)
>>> 
>>> Using the Arch Rollback Machine, I downloaded Emacs 24.2.4 and
>>> downgraded (also required downgrading imageMagick from 6.8.3.10 ->
>>> 6.8.2.3). Now it works again (refer to the reproducible example from
>>> the mailing list post):
>>> - http://www.mail-archive.com/emacs-orgmode@gnu.org/msg68238.html
>>> 
>>> Eric, your example fails for me. I get:
>>> 
>> 
>> Yes, because my example only works in external (non session) execution
>> with the current buggy code, where as your example works with session
>> execution in the old working code.
>> 
>>> 
>>>>>> x = 1
>>>>>> return x
>>>  File "<stdin>", line 1
>>> SyntaxError: 'return' outside function
>>> 
>>> This works, hoever:
>>> 
>>> #+begin_src python :session
>>>      x = 1
>>>      x
>>> #+end_src
>>> 
>>> #+RESULTS:
>>> : 1
>>> 
>>> #+begin_src python :session
>>>      x
>>> #+end_src
>>> 
>>> #+RESULTS:
>>> : 1
>>> 
>>> So, with emacs 24.2.4 and current Org-mode (pulled just now) and clean
>>> make, *both* named and un-named sessions work for me on Arch Linux.
>>> 
>> 
>> Aha! Thanks for sleuthing this out.  So the problem lies in changes to
>> the python.el distributed with Emacs.  I don't suppose we can ask
>> whoever made these changes to python.el to fix the breakage they've
>> caused in Org-mode?
>> 
>> Thanks,
>> 
> 
> Please give me some time still to investigate. Still doubt it's python.el
> But if yes, probably will be able to tell more.

I think 24.3 is where they changed python.el to fgallina's python.el.  So I'd be 
willing to bet that it _is_ the problem since it's a complete rewrite and many 
things changed.

-Ivan