Re: [FR] Display stderr contents after executing shell blocks (even when stdout :results output is requested) (was: Org 9.6-pre and Bash sessions)

[Tim Cross]

Ihor Radchenko <yantar92@posteo.net> writes:

> Rudolf Adamkovič <salutis@me.com> writes:
>
>> Ihor Radchenko <yantar92@posteo.net> writes:
>>
>>> I do not think that it make sense to display that buffer when the code
>>> finishes successfully. I can see this kind of behaviour
>>> breaking/spamming automated scripts or export---code working in the
>>> past may throw error output into unsuspecting users.
>>
>> But the exit code has nothing to do with the standard error.
>>
>> Unix programs can, and often do, halt with non-zero exit codes while
>> producing error output containing important information, such as
>> deprecation warnings.  Further, many programs use error output as the
>> alternative "anything but the result" stream.
>>
>> Preserving user data, instead of trashing it, data does not count as
>> "spamming ... unsuspected users".  On the contrary!
>>
>> For example, I use a program for work that uploads data to a certain
>> 3rd-party server.  It exits with a zero code but also shows extremely
>> important notices on error output.  As an "unsuspecting user", if I used
>> Babel to run the program, I would end up in a trouble.
>>
>> So, we should never implicitly trash user-generated data, let alone
>> based on a "completely made up" belief that a non-zero exit code somehow
>> implies "no important error output".  It does not.
>>
>> (I speak only about Unix-like systems here.  Perhaps on other operating
>> systems, things work differently.)
>
> Dear All,
>
> As explained in the above quote, it may be reasonable to display stderr
> in the shell (and possibly other) src blocks upon execution.
>
> + Stderr may contain important information even if the code block
>   succeeds
> - Displaying stderr will raise *Error* buffer, which may or may not be
>   expected or desired.
>
> What do you think?

I think this is perhaps the tip of a more complex iceberg. Not sure if
we can address bits individually. Suspect we are better off
clarifying the basic babel model and then look at specific language
exceptions/limits and adjusting the model or documenting where back ends
need to diverge from the basic model.

Part of the challenge here is that the relevance/importance of stderr is
somewhat dependent on the language. Same holds true for handling of
error/return codes and to 'results' output.

For me, this is another symptom of our need to define a clearer model
for babel processes so that we can get some consistency across the
board. Such a model would likely also make it easier for people to
develop new babel back ends. We may even need 2 models, one for
interactive/repl based back ends and one for non-interactive/scripted
back ends. 

The 2 big questions I see are "What should be defaults be?" and "How do
we handle the options without adding lots of new switches or suffering
an option permutation blow out?".

I do agree with the other posts regarding the importance of stderr.