Re: Bug or not a bug? dot expansion in ob-shell

[Tim Cross <theophilusx@gmail.com>]

It seems to me that two separate issues have been mixed up and causing
some confusion here. However, I think it is actually quite simple once
we consider the issues separately.

Issue 1: Defining the meaning of :result value and :result output
Issue 2: Specifying what the default :result setting would be i.e.
:result without a value/output specifier.

Issue 1 seems the most straight-forward to me
- output :: Whatever goes to stdout/stderr
- value :: whatever would be returned by the execution of the code. This
might be a return value (i.e. for some shell commands) or the value
returned by a function (including shell functions i.e. bash function) or
the last command executed etc. Essentially, anything returned (including
nil) by the block. STDOUT/STDERR is never a return value (though some
languages may send output to STDOUT/STDERR as well as returning it, in
which case it would also be in :return value).

Note that I don't agree that the only 'useful' result from shell blocks
is output. You might have a complex shell script which uses source
blocks to define shell functions
that return values which you use via oweb expansion to include in other
blocks etc.s

With this definition, essentially :result value is essentially anything
except whatever is sent to stdout/stderr.

Issue 2 is a little more difficult. It is likely that a 'standard'
default for :result that is the same for all languages is not
possible/desired. The default will likely be a combination of what seems
most natural for that language and what is most common usage for the
majority of users. It could be that for some languages, the default for
:result should be :result output and for others it should be :result
value.

Personally, I care less about issue 2 than issue 1. In the worst case, I
will need to change my header arguments for some blocks and that would
be easily automated. Far more critical is that :result value and :result
output are clear, unambiguous and consistent. Any proposal to change
these meanings because of different uses cases in languages is a bad
idea. Instead, changing the 'default' would be preferable. Having shell
source blocks return STDOUT/STDERR output for :result value is IMO a bad
idea. Having shell blocks default to :result output when only specifying
:result while having other languages, like python or clojure default to
:result value seems far more preferable (provided differences are
clearly documented of course).

The current situation seems inconsistent to me e.g.

#+begin_src shell :results value
  echo .

#+end_src

#+RESULTS:
: .


#+begin_src shell :results output
  echo .

#+end_src

#+RESULTS:
: .

You get the same results value regardless of whether :results value or
:results output. I think :results value should return 0 (return code for
echo). Whether shell blocks defaults to :results output or :results
value is a different questions (it probably should default to :result
output). e.g.


#+begin_src shell
  echo .

#+end_src

#+RESULTS:
: .


#+begin_src shell :results value
  echo .

#+end_src

#+RESULTS:
: 0


#+begin_src shell :results output
  echo .

#+end_src

#+RESULTS:
: .

Tom Gillespie <tgbugs@gmail.com> writes:

> Hi all,
>     Sorry to be late to this thread (and for a wall of text), but as a
> heavy user of ob-shell I wanted to chime in. First a disclosure, I
> would be very unhappy if option 1 were selected, it would require me
> to make a whole bunch of changes and try to find an option to revert
> to the current default behavior. When I run a bash block in org mode,
> I want what is going to stdout, why else would I run it in org? If I
> don't want to capture the output then it is either in a pipeline
> inside the block, or I will silence the results. Option two is by far
> the most reasonable answer in my opinion and there is a consistent
> principle which can be applied in many cases so that it would not be
> an exception. The principle is that block defaults for results should
> default to the value that the language itself makes the most
> accessible. Shell return codes are esoteric from this point of view, I
> had been writing bash scripts for years before I learned about $?. By
> far the most actionable and accessible thing to a user of a shell
> program is the stdout -- ignore the naming for a moment, because the
> meaning is not in the name, it is in how it interacts with other
> programs in the larger environment. If we apply the logic that says
> that option 1 should be the default, then python blocks set to results
> value should return nothing unless an error is raised and then they
> should return the error. This is clearly absurd, no python programmer
> cares about the absence of an error and making the default behavior of
> python blocks verbosely report their non-errorness by default would
> likely incite a riot. What has happened with shells is that the
> nomenclature has been switched with respect to how users and other
> code consume and deal with 'outputs' aka return values in any other
> langues, shell return values are error signals and should be treated
> as such, unfortunately (or perhaps fortunately?) org-mode doesn't have
> an error handling/signaling system, but if a python code block fails
> emacs will yell at us, shells don't do this, it is up to the user to
> detect and handle the error case themselves, but that is a language
> internal matter. In summary, option 2 seems to me to be the most
> consistent with how users interact with shell commands and shell
> scripts, return codes are more akin to error handling in other
> languages, so ignoring the naming issues, if `my-org-block` behave
> like a shell function, then pipe would be the more appropriate default
> behavior than `$?`. I think that the underlying principle can be
> applied to other languages as well to arrive at sane defaults.
> Thoughts?
> Tom
>
>
> On Sat, Feb 29, 2020 at 7:41 AM Jack Kamm <jackkamm@gmail.com> wrote:
>>
>> Sorry, I was confused about this:
>>
>> > According to my reading of this thread, most of the commenters were in
>> > agreement that we should keep the original behavior and return the exit
>> > code, as we do in 9.3.
>>
>> Actually, it looks like ":results value" does return the exit code. I
>> just got confused because shell blocks now return the output when no
>> ":results" is specified.
>>


--
Tim Cross