because with huge files (e.g. 17Gb) the process is killed by the kernel OOM killer

and produce a category tree containing only the converted datasets

reading "FROM_DATETIME" env variable,
instead of opening the previous table of contents XML file in Git