Perl, 1428 1099
This has 1193 ASCII characters (including 960 permuted binary digits).
1193 - 94 = 1099
$s='010011100001100010101100111111101001101011101000100000101011011010100110111111011111101011101000100110111111011100101000011101011110100000101000100101011111111110101100101101011010011100100100011110110001011100100001011010100111100000011110111110011100101000100110111111101001011110101011100110101110101101011110101100111111100010101101101100011110100101011111111111101101101000111111011110100111011100101000011101011110111111011010111111101100101101101011100010100111100000111110';$_=q{$i=join'',A..Z,a..z,0..9,'. ';print map({substr$i,oct'0b'.$_,1}$s=~/.{6}/g),$/;chop($s=<>);$s=join'',map{sprintf"%06b",index$i,$_}$s=~/./g;$t=join'',map{$_ x(480-(()=$s=~/$_/g))}0,1;print"\$s='$s';\$_=q{$_};eval#$t"};eval#000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111
My first design
Before I took a suggestion from Dennis to switch to binary, my program permuted octal digits.
My first design encodes each string in 160 octal digits, with 2 digits per character. This encoding has 1008=64 different characters. The octal system has 8 different digits. The program must have 160 copies of each digit, so it permutes 8×160=1280 digits.
I keep 160 digits in $s
and the other 1120 digits in $t
. I start with a program that is not a quine, but only prints the assignments to $s
and $t
for the next run. This is it:
$s = '2341425477515350405332467737535046773450353640504537765455323444366134413247403676345046775136534656553654774255543645377755507736473450353677327754555342474076';
$t = '0000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000011111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222223333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333334444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666667777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777';
# $i = character map of 64 characters, such that:
# substr($i, $_, 1) is the character at index $_
# index($i, $_) is the index of character $_
$i = join '', 'A'..'Z', 'a'..'z', '0'..'9', '. ';
# Decode $s from octal, print.
# 1. ($s =~ /../g) splits $s into a list of pairs of octal digits.
# 2. map() takes each $_ from this list.
# 3. oct() converts $_ from an octal string to a number.
# 4. substr() on $i converts number to character.
# 5. print() outputs the characters from map() and a final "\n".
print map({ substr $i, oct, 1 } $s =~ /../g), "\n";
# Read new $s, encode to octal.
# 1. ($s = <>) reads a line.
# 2. chop($s) removes the last character of $s, the "\n".
# 3. ($s =~ /./g) splits $s into characters.
# 4. map() encodes each character $_ as a pair of octal digits.
# 5. join() concatenates the pairs from map().
chop($s = <>);
$s = join '', map { sprintf "%02o", index $i, $_ } $s =~ /./g;
# Make new $t.
# 1. map() takes each $_ from 0 to 7.
# 2. $_ x (160 - (() = $s =~ /$_/g)) makes a string where $_ repeats
# 160 times, minus the number of times that $_ appears in $s.
# 3. join() concatentates the strings from map().
$t = join '', map { $_ x (160 - (() = $s =~ /$_/g)) } 0..7;
# Print the new assignments for $s and $t. This is not yet a quine,
# because it does not print the rest of the program.
print "\$s = '$s';\n\$t = '$t';\n";
(() = $s =~ /$_/g))
is an assignment to an empty list of variables. I take this trick from the context tutorial at PerlMonks. It forces list context on the match operator =~
. In scalar context, the match would be true or false, and I would need a loop like $i++ while ($s =~ /$_/g)
to count the matches. In list context, $s =~ /$_/g
is a list of matches. I put this list in the scalar context of a subtraction, so Perl counts the list elements.
To make a quine, I take the form $_=q{print"\$_=q{$_};eval"};eval
from the Perl quines at Rosetta Code. This one assigns a string q{...}
to $_
and then calls eval
, so I can have my code in a string and also run it. My program becomes a quine when I wrap my third to last lines in $_=q{
and };eval
, and change my last print
to print "\$s = '$s';\n\$t = '$t';\n\$_=q{$_};eval"
.
Finally, I golf my program by changing the first assignment to $t
to a comment, and by removing extra characters.
This has 1522 ASCII characters (including 1280 permuted octal digits).
1522 - 94 = 1428
$s='2341425477515350405332467737535046773450353640504537765455323444366134413247403676345046775136534656553654774255543645377755507736473450353677327754555342474076';#0000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000011111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222222223333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333333334444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444444555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555555666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666666667777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777777
$_=q{$i=join'','A'..'Z','a'..'z','0'..'9','. ';print map({substr$i,oct,1}$s=~/../g),"\n";chop($s=<>);$s=join'',map{sprintf"%02o",index$i,$_}$s=~/./g;$t=join'',map{$_ x(160-(()=$s=~/$_/g))}0..7;print"\$s='$s';#$t\n\$_=q{$_};eval"};eval
The switch to binary
In the comments, Dennis noticed that 960 permuted binary digits would be fewer than 1280 octal digits. So I graphed the number of permuted digits for each base from 2 to 16.
Maxima 5.29.1 http://maxima.sourceforge.net
using Lisp ECL 13.5.1
...
(%i36) n : floor(x);
(%o36) floor(x)
...
(%i41) plot2d(n * ceiling(log(64) / log(n)) * 80, [x, 2, 16],
[xlabel, "base"], [ylabel, "number of permuted digits"]);
(%o41)

Though base 8 is a local minimum, bases 2 and 3 and 4 tie for best base, at 960 permuted digits. For code golf, base 2 is best because Perl has conversions for base 2.
Replacing 1280 octal digits with 960 binary digits saves 320 characters.
Switching code from octal to binary costs 8 characters:
- Change
oct
to oct'0b'.$_
costs 7.
- Change
/../g
to /.{6}/g
costs 2.
- Change
"%02o"
to "%06b"` costs 0.
- Change
160
to 480
costs 0.
- Change
0..7
to 0,1
saves 1.
I learned some Perl golf tips. They save 14 characters:
- Change
'A'..'Z','a'..'z','0'..'9'
to A..Z,a..z,0..9
, using barewords and bare numbers, saves 12 characters.
- Change
"\n"
to $/
saves 2 characters.
I save 3 characters by moving the #$t
comment to the end of the file. This removes the newline that ends the comment, and a literal \n
in the quine.
These changes save a total of 329 characters, and reduce my score from 1428 to 1099.