It's probably best to use an appropriate module for this. Net::LDAP::LDIF appears to normalize the entries. I haven't used perl-ldap much myself but as far as I know it's a very good module for LDAP.

use warnings;
use strict;
use Net::LDAP::LDIF;

my $ldif = Net::LDAP::LDIF->new("test.ldif", "r", onerror=>'die');
while ( not $ldif->eof ) {
    my $entry = $ldif->read_entry;
    print $entry->ldif;
}
$ldif->done;
[download]

Given the following input, the last three lines before the dash are transformed into one line "foo: BarTest123Blah", which I guess is what you want?

dn: cn=Test,ou=Foo,o=Bar,c=US
changetype: modify
replace: foo
foo: Bar
 Test123
 Blah
-
[download]

You may want to have a look at this post Re^2: Grab 3 lines before and 2 after each regex hit (sliding window).

The update shows a generic solution how to parse a sliding window of 6 lines and test the 4th one.

use strict; 
use warnings; 
use Data::Dump; 

my @window; 
while ( my $line = <DATA> ) { 
    push @window, $line; 
    next if @window < 6; # init 
    if( $window[3] =~ m/[^\d]+\d+/ ){ 
         dd \@window; 
    } 
    shift @window; 
}
[download]

You could easily adapt this to your needs.

What code have you already written and where does it fail to do what you need?

Personally, I would read the file line-by-line and accumulate all data into an array until I encounter a line with only a dash on it. Then I would look at the three last lines in the array and reformat them if needed. After that, I would output and reset the array.

#!/usr/bin/perl

# http://perlmonks.org/?node_id=1201076

use strict;
use warnings;

print +(join '', <DATA>) =~ s/^ .*\K\n (?=.*\n-)//gmr;

__DATA__
a data line
 Test123
 Blah
-
something else
 do not change this line
-
a data line
 combine th
 is line
-
something else
[download]

I am impressed by your suggestion tybalt89! It works wonders. Thank you so much!