Confusing error message when options for `python` are erroneously quoted

[za3k]

Bug report

Bug description:

#!/bin/python -q -i

I was writing an unrelated post about shebang lines tricks+tips, and happened to pick python as an example, only to discover some weird behavior. The above gets parsed as /bin/python "-q -i" <path/to/script>

But the message python prints is quite confusing:

Unknown option: - 
usage: python [option] ... [-c cmd | -m mod | file | -] [arg] ...
Try `python -h' for more information.

What's happening is that the "unknown option" is a single space, being parsed as a short-form option, and printed. But the output is pretty opaque, since the space is not quoted, and it's just a trailing invisible space.

Aside: Several other invalid one-byte options like \n get printed wrongly. This doesn't seem worth fixing.

---

Proposed fix:

I think the shebang case happens to users often enough that it would be good to have some kind of better error message if a short-form options list contains either of

• - (this would require scanning ahead!), which either indicates some quoting problem, or an incorrect long-form option (-check-hash-based-pycs rather than --check-hash-based-pycs).

I think a good error message could be unknown option: -q -i or unknown option: "-q -i". Quoting the argument when it contains a space seems clearer to me.

The current long-form options have pretty clear error messages either way, mostly by happenstance.

Currently, -help-* parses as -h which then prints help, stopping option parsing. I have no strong opinion about whether that behavior is better or worse than printing an error message.

CPython versions tested on:

3.13

Operating systems tested on:

Linux

[picnixz]

What's the behavior of other CLIs like ls, cp, etc in this case?

[za3k]

Oh, sensible question.

$ ls "-l -a"
ls: invalid option -- ' '
$ ls "-q"
ls: invalid option -- 'q'
$ ls -a-s-d
ls: invalid option -- '-'
$ ls --a-sdsd
ls: unrecognized option '--a-sdsd'
$ cp "-v -v"
cp: invalid option -- ' '

$ which "-a -a" # note: zsh builtin
which: bad option: - 

$ /bin/which "-a -a"
/bin/which: invalid option -- ' '
/bin/which: invalid option -- '-'

$ perl "-p -u"
perl # unquotes the argument and executes normally

GNU coreutils programs like cp and ls use seem to unconditionally quote the argument. cp and ls print the first failure, which seems to print all of them.

zsh's which builtin matches python behavior. Note that it doesn't support long-form options.

perl is doing its own thing.

[picnixz]

I think we should follow GNU coreutils as we would have less ambiguities, namely "invalid option -- '<FAILURE>'".

Let's hear the opinion of people who worked on some accessibility REPL issues @hugovk @pablogsal @serhiy-storchaka

[hugovk]

I think GNU coreutils is an okay default, but can we do better?

invalid option -- doesn't really tell me anything about what I need to do to get it working. Can we give an actionable message?

Or is it safe to act like perl and unquote the arguments and execute normally?

---

btw I use https://github.com/lsd-rs/lsd as an alias for ls and it tries to be more useful, but doesn't really help:

❯ ls "-l -a"
error: unexpected argument '- ' found

  tip: to pass '- ' as a value, use '-- - '

Usage: lsd [OPTIONS] [FILE]...

[serhiy-storchaka]

An actionable message is just a guess. It can be correct, it can be wrong. This is not common issue, and we cannot be sure that special message will be helpful and not add more confusion.

Acting like perl is unsafe. If the guess is not correct, it will silently do wrong things instead of just printing (possibly unclear) message.

[pablogsal]

I agree with @serhiy-storchaka

[za3k]

Right, to summarize (there are a lot of thumbs-ups) everyone asked is in agreement to follow the coreutils example of

/bin/python: invalid option -- '<FAILURE>'

Although both of the following have some of the same virtues for the same input

/bin/python: invalid option -- ' '
/bin/python: invalid option -- '-q -i'

I'm pretty sure we prefer the exact option coreutils uses (the first), and which has therefore been (1) tested for a few decades and (2) will be familiar to users.