Hello,

Attached please find my revised patch based on Kyle's feedback. I simplified the behavior to send the whole block to the tmpfile (including the lastline), but restricted this change to ":results output" only (":results value" retains its old, broken behavior). I also added a test and fixed minor style issues.

I haven't heard back from the FSF yet, any idea how long it takes to hear from them?

Given how broken ":session :results value" is, and the difficulty in implementing it correctly, I agree with Kyle's suggestion to remove it. Should we start another thread to discuss this?

Jack

On Tue, Nov 21, 2017 at 8:28 AM, Jack Kamm <jackkamm@gmail.com> wrote:
Yes, I'm starting to see now how difficult it is to properly support ":session :results value". I would vote to remove it from ob-python...

I think the patch still improves ":session :results output" so I will simplify it and restrict to that case, leaving ":session :results value" unchanged for now.

Sorry for sending this twice Kyle, forgot to reply all.

On 21 Nov 2017 4:04 am, "Kyle Meyer" <kyle@kyleam.com> wrote:
Jack Kamm <jackkamm@gmail.com> writes:

> In response to this:
>
>> I can't think of a good solution, though.  Stepping back a bit, I think
>> it's unfortunate that python blocks handle ":results value" differently
>> depending on whether the block is hooked up to a session or not.  For
>> non-sessions, you have to use return.  Using the same approach
>> (org-babel-python-wrapper-method) for ":session :results value", we
>> could then get the return value reliably, but the problem with this
>> approach is that any variables defined in a ":results value" block
>> wouldn't be defined in the session after executing the block because the
>> code is wrapped in a function.
>
> How about if we used the "globals()" and "locals()" functions in Python?
>
> Something like this at the end of the wrapper block, before return:
>
> for k, v in locals().items():
>     globals()[k] = v

Hmm, placing that code "before return" is a problem.  Like with
non-session ":results value" blocks, the user would be responsible for
inserting the return (or even multiple return's), so we can't know where
to insert the above code without parsing the block :/

> Another bug with the current approach is that it breaks if common idioms
> like "for _ in range(10)" are used. ("_" is used to inspect the last output
> of the shell, an obscure feature I hadn't known about until now).

Right.  Also, IIRC the built-in interactive python and ipython treat
multiline blocks differently.  With

    if True:
        "ipython ignores my existence"

the built-in shell binds "_" to the string's value, but ipython doesn't.

--
Kyle