Yes, Ambrus. I think the best way is to recognize a full record and make whatever in between as one record. The input file is in fact shown as below, The first element of one record should be a file name:

file1.xml;Description A, B
    C, D, E...;SystemNumber
file2.txt;Description X,Y,Z;SystemNumber
file3.xml;Description M,N,
          O,P,Q;SystemsNumber
[download]

I think it might be easy if I first delete all the white spaces in each line; then read all lines into an array as the elements; then split the array by the ".xml" or ".txt" files and put them as the first element of a child array. Is that the right track?