D'oh! I did omit a couple of bits didn't I (It was (obviously) untested -- I don't have an /etc/passwd).

I tried to remove the $b=~ & $a=~ bits when I noticed that reverse was one character less -- ignoring/forgetting that $_ cannot be both at the same time :(.

I don't think the (...)[0] bits are necessary, without /g only the regex will only return one (the first) match.

This should work? (still untested):

perl -E"say sort{$b=~/:(\d+):/<=>$a=~/:(\d+):/}<>" /etc/passwd
[download]

No, unfortunately that doesn't work either. A regex match in scalar context returns 1 or 0. It needs to be called in list context to return the value.

>perl -E"print 'foo'=~/(o+)/;"
oo
[download]

>perl -E"print scalar 'foo'=~/(o+)/;"
1
[download]

No, no. It does work (tested and confirmed). See my reply to thundergnat.

NOTE- remember you have to omit the typos though, as I already pointed out. So far, the winner remains:

UPDATE: thundergnat is right that the code below DOESN'T work as intended

perl -e 'print reverse sort{/(:\d+:)/<=>/(:\d+:)/}<>' /etc/passwd
Tommy
A mistake can be valuable or costly, depending on how faithfully you pursue correction

No, no. It doesn't work (tested and confirmed).

The sort block operates on the global variables $a and $b. A match without an explicit variable operates on $_. There is nothing mapping $a and $b to $_ there, so the sort block is just comparing 0 to 0 and leaving the order alone. The fact that /etc/password is in a "sorted" order makes it SEEM like it is working... but it isn't.

print reverse sort{/(:\d+:)/<=>/(:\d+:)/}<DATA>;

__DATA__

aaa:5:
bbb:3:
ccc:1:
ddd:7:
eee:8:
[download]

eee:8:
ddd:7:
ccc:1:
bbb:3:
aaa:5:

It's quite possible that I'm being exceedingly dense, but how is this any different from just:
perl -E"say reverse<>" /etc/passwd

UPDATE- we only wanted it to work as described below-- with implicit reassignment of $_ to $a and also to $b.

It's not the same. BrowserUK's solution uses two regexes, each with a capture group. The capture groups implicitly assign to what would be the equivalents of "$a" and "$b". Since the point of this excercise was to save typing (potentially at the loss of clarity), BrowserUK didn't bother wasting precious keystrokes in explicitly assigning the result of the regex capture groups to an "$a" or "$b". Instead, he/she (I don't know BrowserUK personally) just let Perl handle those details. BrowserUK definitely meant what was posted ;~)