reset Clp termination code to "time limit reached" when seeing "iteration limit reached" in dual simplex resolve

[yuxies]

Change the return code in dual_simplex from LP_D_ITLIM to LP_TIME_LIMIT if use Clp.

• Clp may return "maximum iteration limit reached" flag when it is actually timeout (see ClpModel::hitMaximumIterations()).

[tkralphs]

@yuxies I don't think we can just always LP_TIMELIMIT here because there are checks in other places for LP_D_ITLIM (like in strong branching). Can we try to distinguish in someway, such as checking whether we have set an iteration limit and or are expecting that return code for some reason? Maybe it's better to do this on the MibS side, since we know we are only expecting a timeout there and not an iteration limit.

[yuxies]

Can we try to distinguish in someway, such as checking whether we have set an iteration limit and or are expecting that return code for some reason?

The change here is inside the function int dual_simplex(LPdata *lp_data, int *iterd), after si->resolve(); so the changed line will only affect the immediate result of the dual simplex solving itself, when Clp is used. And there would not be any misunderstanding, since any other term code from Clp has its own correct mapping, and we are not expecting LP_D_ITLIM as a Clp dual simplex term result.

[tkralphs]

Can we try to distinguish in someway, such as checking whether we have set an iteration limit and or are expecting that return code for some reason?

The change here is inside the function int dual_simplex(LPdata *lp_data, int *iterd), after si->resolve(); so the changed line will only affect the immediate result of the dual simplex solving itself, when Clp is used. And there would not be any misunderstanding, since any other term code from Clp has its own correct mapping, and we are not expecting LP_D_ITLIM as a Clp dual simplex term result.

If I understand correctly, this is not true. We are indeed expecting LP_D_ITLIM as a Clp dual simplex result in some places, as I mentioned. Just search for LP_D_ITLIM in the source code.

[yuxies]

Ok, maybe we can add another condition before this line "checking whether we have set an iteration limit". Then the change will look like:

int itlim_check = -1;
retval=si->getIntParam(OsiMaxNumIteration, itlim_check);
if(itlim_check >= LP_MAX_ITER){ term = LP_TIME_LIMIT; }

[tkralphs]

Hmm, this is probably reasonable. The only edge case would b if an iteration limit is set, but the time limit is nevertheless being reached. Can we check the actual number of iterations or the actual time to see which is being triggered?

[yuxies]

Just change the condition to check the actual iteration number.

[tkralphs]

Yes! But could you open this against master? Then I will cherry pick it over to 5.7.

[yuxies]

Rebased the changes onto the master branch.

Note: there is a similar case switch in this function int initial_lp_solve (LPdata *lp_data, int *iterd){} that may be affected by the Clp termination code issue.

[tkralphs]

Yes, I guess we need to make the same change in initial_lp_solve() and also in solve_hotstart. Can you do that and we can finally put this to bed?

[tkralphs]

OK, looks good!