This document has thousands of identically named headers. I believe that is the problem. Translation slows down as it goes, and it is almost certainly because `registerHeader` uses a linked list of the previously used header identifiers. Each time pandoc parses a header, it needs to traverse this linked list. If the list has thousands of elements, this can start to take some time. At least this hypothesis is consistent with the behavior I'm seeing running a conversion to HTML. There may also be a laziness-related space link. All this can be solved by using better data structures, probably. I didn't really think anyone would have more than a few identically named headers -- and they won't appear in real-world cases. +++ Pablo Rodríguez [Jan 22 16 02:26 ]: