/$regex2/ will have the same interpolation as qr/$regex2/, so we can look more closely at what you are getting by examining the resulting regex2 object:

my $regex1 = qr/(?:(?|(?:\")([^\\\"]*(?:\\.[^\\\"]*)*)(?:\")))/;
my $regex2 = qq/(?:(?|(?:\")([^\\\"]*(?:\\.[^\\\"]*)*)(?:\")))/;

print $regex1, "\n", qr/$regex2/, "\n";
[download]

The output:

(?^:(?:(?|(?:\")([^\\\"]*(?:\\.[^\\\"]*)*)(?:\"))))
(?^:(?:(?|(?:")([^\"]*(?:\.[^\"]*)*)(?:"))))
[download]

The single quoted construct is gobbling the escapes.

FWIW, single-quoted regex notation gets you closer (update: than double-quoting) to the desired final form, but still not all the way and there are some pitfalls. Better, IMHO, to stick to qr//. (BTW: morgon: It's not necessary to escape a  " (double-quote) in a qr// expression.)

File rx_qr_qq_q_1.pl:

use warnings;
use strict;

my $rx_qr = qr/(?:(?|(?:\")([^\\\"]*(?:\\.[^\\\"]*)*)(?:\")))/;
my $rx_qq = qq/(?:(?|(?:\")([^\\\"]*(?:\\.[^\\\"]*)*)(?:\")))/;
my $rx__q =  q/(?:(?|(?:\")([^\\\"]*(?:\\.[^\\\"]*)*)(?:\")))/;

print qq{raw qr: $rx_qr \n};
print qq{qr/qq/: }, qr/$rx_qq/, "\n";
print qq{qr/q/ : }, qr/$rx__q/, "\n";
[download]

Output:

c:\@Work\Perl\monks\morgon>perl rx_qr_qq_q_1.pl
raw qr: (?-xism:(?:(?|(?:\")([^\\\"]*(?:\\.[^\\\"]*)*)(?:\"))))
qr/qq/: (?-xism:(?:(?|(?:")([^\"]*(?:\.[^\"]*)*)(?:"))))
qr/q/ : (?-xism:(?:(?|(?:\")([^\\"]*(?:\.[^\\"]*)*)(?:\"))))
[download]

---

Give a man a fish:  <%-{-{-{-<

Neat trick, thanks a lot.

So I need to double all the backslashes.

So I need to double all the backslashes.

I would say that you need to use  qr// to build Regexp objects. These objects behave essentially like strings in the situations in which you seem to be interested in using them.

---

Give a man a fish:  <%-{-{-{-<

So I need to double all the backslashes.

Probably not. I think that you need to use qr//

Have you printed the two values you get? They are different because qq treats the backslash differently from what qr does.

The output is above in the end block.

What I need is string that behaves just as the compiled regex as I want to feed that into a perl5-compatible regex-engine for go.

Have you printed the two values you get?

The output is above in the end block.

I'm pretty sure that by "the two values", Corion meant $regex1 and $regex2.

By the way, the only quoting construct that doesn't interpolate at all is here docs with a single-quoted delimiter:

chomp( my $regex2 = <<'ENDREGEX' );
(?:(?|(?:\")([^\\\"]*(?:\\.[^\\\"]*)*)(?:\")))
ENDREGEX
print $regex2, "\n";

my ($match2) = q|"hubba \"bubba\""| =~ /$regex2/;
print $match2, "\n";

__END__

(?:(?|(?:\")([^\\\"]*(?:\\.[^\\\"]*)*)(?:\")))
hubba \"bubba\"
[download]

... string that behaves just as the compiled regex ...

That's what  qr// is for; see perlop. The  Regexp object returned by qr// is almost a string; it interpolates as a string in  m// s/// qr// expressions or when print-ed or string interpolated. It is not, however, the same as what is produced by the  qq// q// operators.

---

Give a man a fish:  <%-{-{-{-<