Re: [PATCH] Re: [BUG] ob-shell: :shebang changes interpretation of :cmdline

[Max Nikulin <manikulin@gmail.com>]

On 21/04/2024 22:09, Matt wrote:
> The proposed solution assumes we intend to parse the characters
> following :cmdline as space delimited and grouped by quotes.  However,
> AFAICT, the parsing issue makes this solution ambiguous.

Matt, I am sorry, but I do not agree with your proposal. I do not think 
that `split-string-and-unquote' will solve all issues.

Certainly issues with formatting of output should be treated separately.

I figured out there is at least one more issue. Consider

#+property: header-args:bash :results verbatim
#+begin_src bash :cmdline $LANG :shebang #!/bin/bash
   printf '"%s" ' "$0" "$@"
   printf '\n'
   tr '\000' '\n' </proc/$$/cmdline
#+end_src

#+RESULTS:
: "/tmp/babel-Is56Ki/sh-script-8FluOx" "$LANG"
: /bin/bash
: /tmp/babel-Is56Ki/sh-script-8FluOx
: $LANG

#+begin_src bash :cmdline $LANG
   printf '"%s" ' "$0" "$@"
   printf '\n'
   tr '\000' '\n' </proc/$$/cmdline
#+end_src

#+RESULTS:
: "/tmp/babel-Is56Ki/sh-script-g0cQ7R" "en_GB.UTF-8"
: /usr/bin/bash
: -c
: /tmp/babel-Is56Ki/sh-script-g0cQ7R $LANG

First line is argv as it is represented for a script, next lines are 
exec arguments at lower level (actual executable may be obtained from
"readlink /proc/$$/exe")

Notice that in the former case "$LANG" is passed literally, but in the 
latter it is expanded. I am in favor of dropping `shell-command-switch' 
in the latter case to pass arguments literally in both cases.

I think, it would be more consistent with :var to specify multiple 
arguments using elisp lists
#+header: :cmdline '("first 1" "second 2")

However looking wider, I do not like that :cmdline for ob-shell has 
different meaning than for other languages, see e.g. ob-sql. Only for 
shell this parameter is treated as arguments of a *script*. In other 
cases :cmdline is used to specify arguments of *interpreter* and I think 
ob-shell should follow this convention.

Actually script arguments (and :stdin) might be applied to python and at 
least some other languages, so support of this feature should be moved 
from ob-shell to common org-babel code.

My point:
- header arguments should have as close as possible meaning across 
various languages.
- withing ob-shell variation of ways to execute a script should be 
minimized either some parameters (:cmdline, :shebang, :stdin) are 
specified or not.

Finally a note on tests
> +(ert-deftest test-cmdline-alone-and-with-shebang-have-same-result ()
> +  "Pass arguments to a block.  Don't use shebang.  Then use
> +shebang set to the same language as the block.  The result should
> +be the same."
> +  (should (equal
> +           (org-test-with-temp-text
> +               "#+begin_src bash :cmdline 1 2 3
> +echo \"$1\"
> +<point>
> +#+end_src"
> +             (org-babel-execute-src-block))
> +           (org-test-with-temp-text
> +               "#+begin_src bash :cmdline 1 2 3 :shebang #!/usr/bin/env bash
> +echo \"$1\"
> +<point>
> +#+end_src"
> +             (org-babel-execute-src-block)))))
I believe, that starting point of the discussion is that the results 
should be same and decision what is more correct is the result of the 
discussion. Unit tests should check both cases independently and should 
fix particular treatment of arguments.