Using Text::CSV would probably be better. However, this regex approach, while more verbose, is perhaps more maintainable. Needs 5.10+ Special Backtracking Control Verbs. (I've made a guess at the proper regex for a A423.23 thingy.)
>perl -wMstrict -le
"my $s = 'The fox did it[ at 12.23 ] well, Cf. 23 A423.23. The '
. 'swallow was even better';
print qq{''$s''};
;;
my $parens = qr{ \[ [^]]* \] }xms;
my $cf = qr{ (?i) cf \. }xms;
my $ref = qr{ [[:alpha:]]+ \d+ (?: \. \d+)+ }xms;
my $splitter =
qr{ (?: $parens | $cf | $ref) (*SKIP)(*FAIL)
|
\.
}xms;
;;
my @ra = split $splitter, $s;
print qq{'$_'} for @ra;
"
''The fox did it[ at 12.23 ] well, Cf. 23 A423.23. The swallow was eve
+n better''
'The fox did it[ at 12.23 ] well, Cf. 23 A423.23'
' The swallow was even better'
