I'll poke Richard Clamp and see if he remembers - it was his commit and he usually has a good reason for this sort of thing.
It's probably completely antithetical to the spirit of the project but you could implement the line end guessing heuristic in a file reading XS module - it's dead easy to have a state machine that gets its act together on the first line end it encounters and then expects the same sequence for every subsequent line ending and it adds nothing like the overhead it would in Perl. But, as I say, I imagine putting a lump of C in there doesn't fit well with the project ethos :)
-- Andy Armstrong, hexten.net
