correct posixly shell quoting

[vtjnash]

Ensures that addprocs arguments will be passed through a posix shell correctly, (including those with shell-metacharacters such as being added in #21849 for -C, but also weird cases like -- and files named \r).

fix #10120

[yuyichao]

You do make me wonder if I should have chosen another character instead.
OTOH though, , is what llvm use for features so I'd like to keep that, / is a bit strange since it usually mean nesting (path) which this isn't, : looks similar to ; but it feel like a lower evel of separation than ,........................

[vtjnash]

You do make me wonder if I should have chosen another character instead.

This was not the motivation for the PR, I just thought it would be interesting to note current known cases that this fixes (passing other exeflags is usually pretty rare, as is having shell metacharacters in the path).

[yuyichao]

This was not the motivation for the PR

Yes, I know. I just wonder if this can cause other random script to error. Though this is just a compile flag so you normally won't use it in a script or addproc.

[vtjnash]

Note: these disabled tests are left as motivation for someone to finish the implementation of #20401 🙂

[JeffBezanson]

Do these tests not pass anymore?

[vtjnash]

With the current deprecation, they're partially broken. Hence the motivation for someone to remove the deprecation and finish implementing the #20401

[StefanKarpinski]

Would it be possible to just make shell_escape POSIX-compatible by default? If so, I think we should just do that since it seems good to just default to shell escaping in a way that works in POSIX shells.

[vtjnash]

Would it be possible to just make shell_escape POSIX-compatible by default?

Perhaps, since I think currently shell_parse will get the same result when called on the result of shell_escape_posixly (that's partially of the idea of the function - to be an overly conservatively approximation of all shells). I don't think there's any particular reason to remove shell_escape however, since it's useful for its intended purpose (as a pretty-printer for Cmd objects), and I'm not sure what #20401 will want for changes to the parser.

[vtjnash]

since I think currently shell_parse will get the same result

Actually, this isn't true, but upon further review, I think it may actually be due to other bugs in shell_escape, rather than being due to divergence between shell_parse and posix sh:

julia> Base.shell_escape(` '$a"' `)
"'\$a\"'"

julia> Base.shell_escape(` '\$a\"' `)
"'\\\$a\\\"'"