[Coin-discuss] Solver hints - Yes!

[Lou Hafer]

	Yes! Definitely, I think we should do this --- it provides a clean
way to take advantage of the strengths of individual solvers. I think that we
should go with with the `two or more fields' idea. Specifically, I'd propose
one field to specify the hint code, one field to specify actions to be taken
(along the lines John proposes --- try to do, try not to do, throw exception,
etc.), and a final field which is a void pointer to an optional parameter
block. This should provide sufficient generality for future needs, at a fairly
minimal cost.

	There is an issue of code space --- given we have a full integer to
work with, perhaps we can delineate a common code space (known across all
solvers, individual codes not to be overloaded) and then a separate code
space for individual solvers, for things that are sufficiently unique that
they don't really deserve a code in the common code space. (Limits our
exposure to the ICANN business of administering a globally unique code
space.)

	The immediate benefit would be to allow the various solver interfaces
to enable constraint manipulations by default, without having to remember the
unmodified system in order to pass the unitTest. A few hints would serve to
suppress them there. And we could give hints in the context of MIP solvers,
where you want a lot of preprocessing at the root, less when recovering a
subproblem, even less when hot starting.

							Lou