`DesktopEntry::get_args` (private, used in `DesktopEntry::parse_exec*`) does not handle quoting correctly

[quantenzitrone]

Try this abomination desktop entry:

[Desktop Entry]
Name=Default Terminal
Type=Application
Exec=printf "|||%%s|||\\\\n" "quoting terminal" "with 'complex' arguments," "quotes \\"," "" 	"empty args," "new\nlines," "and \\"back\\\\slashes\\""
Categories=TerminalEmulator;

from Vladimir-csp/xdg-terminal-exec

parse_exec() parses this to:

[
    "printf",
    "\"|||%%s|||\\\\\\\\n\"",
    "\"quoting",
    "terminal\"",
    "\"with",
    "'complex'",
    "arguments,\"",
    "\"quotes",
    "\\\\\",\"",
    "\"\"",
    "\"empty",
    "args,\"",
    "\"new\\nlines,\"",
    "\"and",
    "\\\\\"back\\\\\\\\slashes\\\\\"\"",
]

which should be parsed to:

[
    "printf",
    "|||%s|||\\\\n",
    "quoting terminal",
    "with 'complex' arguments,",
    "quotes \\\",",
    "",
    "empty args,",
    "new\nlines",
    "and \\\"back\\\\slashes\\\"",
]

(if i unescaped and re-rust-escaped this manually correctly)

so basically what the spec says about the Exec attribute

Quoting must be done by enclosing the argument between double quotes and escaping the double quote character, backtick character ("`"), dollar sign ("$") and backslash character ("") by preceding it with an additional backslash character. Implementations must undo quoting before expanding field codes and before passing the argument to the executable program.
[...]
Literal percentage characters must be escaped as %%.

[quantenzitrone]

I have successfully parsed this Exec field with

let exec: &str = <the exec arg from that file>;
let exec = exec
    .replace("\\\"", "\"")
    .replace("\\`", "`")
    .replace("\\$", "$")
    .replace("\\\\", "\\")
    .replace("%%", "%");
shlex::split(&exec).unwrap().iter().map(|s| s.replace("\\n", "\n")).collect()

not sure if this will cover all cases

[aleb2000]

Looking at the source, the exec parsing is extremely simplistic and assumes that the quotes are always at the beginning and end of the string, which often is not the case at all. I came around the same problems. Would be nice to have proper parsing of the exec string.

After that, an additional feature could be to not just split the string but to recognize the various parts of the binary invocation, for example the environment variables that are being set (if any) could be grouped together, the parser could then figure out which part is the binary that is being invoked and which part are the arguments to the binary. Of course this would be nice to have but not required. What is most important right now is to at least properly split the string with correct handling of quotes and spaces.

[wiiznokes]

Maybe with the recent commit it will be correct. Would be nice to add test for this