Re: solution wanted for break-on-spaces (w/specifics)

[vr]

use strict;
use warnings;
use feature 'say';

# use Regexp::Common;       
# ^^^ Not used. I'm so lazy, I just peeked at $RE{quoted}
# to construct the "$quoted" expression below, by slightly
# modifying it (see "$") to satisfy the third clause.
# And actually 2nd test case below is to test how it works,
# it seems there's not a similar one among your 18.

my $quoted = qr/
(?:(?|
    (?:(?<!\\)\")(?:[^\\\"]*(?:\\.[^\\\"]*)*)(?:\"|$)|
    (?:(?<!\\)\')(?:[^\\\']*(?:\\.[^\\\']*)*)(?:\'|$)
))
/x;

my $re = qr/(?:$quoted|[^ ])+\K(?: |$)/;

my @tests = ( 
    q(This 'isn\'t nice.'),
    q(This 'isn\'t nice.),
    q(This \"isnt unnice.\"),
);

for my $t ( @tests ) {
    say "[$_]" for split $re, $t;
}

__END__

[This]
['isn\'t nice.']
[This]
['isn\'t nice.]
[This]
[\"isnt]
[unnice.\"]

10 minutes update: aargh, added negative look-behind to cover your 14th case (and added my third). Maybe there are more to add. Further: it's more tricky, 6 (and 7) are split in 3, but wrong, groups. Will look into that later. False alarm? Will see yet later :)

Next morning update. As LanX pointed out, negative look-behind for just a single backslash isn't enough. Then to save this answer (I like how the "keep" \K meta-character helps in regexp for split, it's kind of interesting), maybe it's easier to revert $quoted to as it was borrowed from $RE{quoted}, and tweak the $re:

my $quoted = qr/
(?:(?|
    (?:\")(?:[^\\\"]*(?:\\.[^\\\"]*)*)(?:\"|$)|
    (?:\')(?:[^\\\']*(?:\\.[^\\\']*)*)(?:\'|$)
))
/x;
my $re = qr/
    (?:
        (?:\\\\)+
        |
        (?:\\[^ ])
        |
        $quoted
        |
        [^ ]
    )+ \K
    (?:
        \ | $
    )
/x;

I hope it works now, my 1st attempt at this "update" was broken (see, but better not -- nothing interesting -- below. Sorry for the mess.). But further, it's unclear whether to split on escaped space, or several spaces in a row.

my $quoted = qr/
(?:(?|
    (?:
        (?:[^\\\'\ ]*(?:\\[^\ ][^\\\'\ ]*)*)
        \"
    )
    (?:
        [^\\\"]*
        (?:
            \\
            .
            [^\\\"]*
        )*
    )
    (?:\"|$)
    
    |
    
    (?:(?:[^\\\' ]*(?:\\[^ ][^\\\' ]*)*)\')(?:[^\\\']*(?:\\.[^\\\']*)*
+)(?:\'|$)
))
/x;

And later (final(?)) update: Sigh... damn lack of practice. So this:

my $quoted = qr/
(?:(?|
    (?:\")(?:[^\\\"]*(?:\\.[^\\\"]*)*)(?:\"|$)
    |
    (?:\')(?:[^\\\']*(?:\\.[^\\\']*)*)(?:\'|$)
))
/x;

my $re = qr/
    (?:
        (?:\\.)+
        |
        $quoted
        |
        [^ \\"']+
    )* \K
    (?:
        \ | $
    )+
/x;

# and later:
    my $got = [ split $re, $str ];

passes all tests in LanX's later answer except #2 and is somewhat optimized.

About test #2: consensus is "the brief is unclear", must split-like behaviour generate an empty leading field for #2? Expression to split on is definitely not missing nor space literal. If, nevertheless, it must not (as my solution does, failing #2), then my bad, but still, yeah, this regexp is "working" and can be used to literally split on. :)

[LanX]

I'm not sure about this

(?:(?<!\\)\")

I read it as doublequote which is not preceded by backslash

But what about an escaped backslash \\" or two \\\\" ... ?

I'd rather try something like (Untested pseudocode)

s/^(?:$escaped|$quoted|\S)*\K\s+/\n/g

and

$escaped = qr/\\./;

$quoted  = 
   qr/
     (['"])                  # start
     (?: $escaped | [^\1] )* # inside 
      \1                     # end, probably \g-1 better
    /x;

NB: I didn't cover the case of unclosed quotes, which is unclear anyway.

update

tested - fails - good night! :)

update

see Re: solution wanted for break-on-spaces (w/specifics) for "working" example