Attached is a sample program, data, and output illustrating (what I believe is) the problem. I think that the
info gawk file states that the pieces split from the input string are returned in the first array argument (
ARRAY) and the part matched by the separation regular _expression_ are returned in the second array argument. (
SEPS) Those two roles seem to be reversed in the output of the program I attached here.
This was executed on an up-to-date Fedora "Rawhide" system, running rc4 of the 4.0.0 Linux kernel,
but I just ran the same program and data on a Fedora 21 system (3.18.9 kernel) with the same result. (I thought that the change to gcc5 in F-23 might be a problem, but apparently not.)
$ gawk --version
GNU Awk 4.1.1, API: 1.1
Copyright (C) 1989, 1991-2014 Free Software Foundation.
gawk was (probably) compiled using gcc 5.0.0, but I've not confirmed that. (Since the