backglue transition

[Aaron Richton <richton@nbcs.rutgers.edu>]

The move to integrate slapo-glue has had some interesting side effects.
I'm curious how many of these are intentional and how many of these are not:

1a. 2.3.0 - 2.3.6 "glue-sub" and such is no longer recognized.
1b. of course you get no runtime message about this because no conf bailout!
2. The older "subordinate" can be made to work. However, in the past:

suffix "b,a"
subordinate

suffix "c,a"
subordinate

suffix "a"

resulted in { b,a c,a } results.

The same today results in { c,a b,a } results -- reversed. So if the
goal was behavior compatibility with earlier versions, we've missed that.
(Note that 'glue-sub' statements callback the same way as today's
'subordinate'. That didn't bother me, though, because it was a new
syntax.)

So a few things bother me:

1. The slapo-glue config that worked yesterday with 2.3.6 doesn't work--at
all--today with 2.3.7.

2. The config that we tried to be compatible with from, say, 2.2.28
doesn't work the same way with 2.3.7.

3. If I try to use either of those old configs, I get no
warnings/errors/etc. despite the fact there's a behavior change.


Finally, although I realize there were reasons for moving things into core
and making 'subordinate' the keyword, I really liked the slapo-glue
syntax. In my ideal world, the new backglue would be able to handle either
style of config. I don't know if the new design makes this
possible/plausible.