Am 03.01.2017 um 16:15 schrieb Allin Cottrell:
 Ah, I've found the remaining issue: it's a Windows thing. The Microsoft
 C library doesn't recognize either "nan" or "NaN" as meaning
 not-a-number. We've had a work-around in place to handle "nan", as
 written by gretl in its matrix files, but this wasn't dealing with
 "NaN", as written by R. The work-around is now generalized to handle
 both, in git and today's snapshots. 
Yes, by coincidence I was already using the brand new snapshot an hour 
ago or so, and Jan's script now ran fine.
Jan, if you don't want to unnerve your admin again, in general it might 
be advisable to keep the missings out of the stuff that is passed to R 
in the first place (if that's doable). My feeling is that it would be 
cleaner anyway, apart from avoiding this particular bug.
cheers,
sven