... I'm trying to count digits in a scalar

That's what the OP says, but the  /\d\D*/g regex suggests misterperl is trying to count occurrences of some kind of digit-group pattern. If that's the case, I can, offhand, think of some  s/// approaches (in the vein of a  tr/// operation) that would do the trick as well as the  m// approach that misterperl seems to favor:

c:\@Work\Perl\monks>perl -wMstrict -le
"$_ = '+++1223w3433.45+34***';
 ;;
 my $ndg;
 ;;
 $ndg =()= /\d\D*/g;
 print qq{A: $ndg digit grps.;  m// change unpossible '$_'};
 ;;
 $ndg = do { (my $r = $_) =~ s/\d\D*//g };
 print qq{B: $ndg digit grps.;  s/// string unchanged '$_'};
 ;;
 $ndg = s/(\d\D*)/$1/g;
 print qq{C: $ndg digit grps.;  s/// string unchanged '$_'};
 ;;
 $ndg = s/\d\D*//g;
 print qq{D: $ndg digit grps.;  s/// string CHANGED '$_'};
"
A: 12 digit grps.;  m// change unpossible '+++1223w3433.45+34***'
B: 12 digit grps.;  s/// string unchanged '+++1223w3433.45+34***'
C: 12 digit grps.;  s/// string unchanged '+++1223w3433.45+34***'
D: 12 digit grps.;  s/// string CHANGED '+++'
[download]

My own preference would be for the  m// approach.