DECLARE_EXPORT PyObject* SolverMRP::getattro(const Attribute& attr)
{
if (attr.isA(Tags::tag_constraints))
return PythonObject(getConstraints());
if (attr.isA(Tags::tag_autocommit))
return PythonObject(getAutocommit());
if (attr.isA(Tags::tag_userexit_flow))
return getUserExitFlow();
if (attr.isA(Tags::tag_userexit_demand))
return getUserExitDemand();
if (attr.isA(Tags::tag_userexit_buffer))
return getUserExitBuffer();
if (attr.isA(Tags::tag_userexit_resource))
return getUserExitResource();
if (attr.isA(Tags::tag_userexit_operation))
return getUserExitOperation();
if (attr.isA(Tags::tag_plantype))
return PythonObject(getPlanType());
// Less common parameters
if (attr.isA(tag_iterationthreshold))
return PythonObject(getIterationThreshold());
if (attr.isA(tag_iterationaccuracy))
return PythonObject(getIterationAccuracy());
if (attr.isA(tag_lazydelay))
return PythonObject(getLazyDelay());
if (attr.isA(tag_planSafetyStockFirst))
return PythonObject(getPlanSafetyStockFirst());
// Default parameters
return Solver::getattro(attr);
}
DECLARE_EXPORT void SolverMRP::solve(const Demand* l, void* v)
{
// Set a bookmark at the current command
SolverMRPdata* data = static_cast<SolverMRPdata*>(v);
CommandManager::Bookmark* topcommand = data->setBookmark();
// Create a state stack
State* mystate = data->state;
data->push();
try
{
// Call the user exit
if (userexit_demand) userexit_demand.call(l, PythonData(data->constrainedPlanning));
short loglevel = data->getSolver()->getLogLevel();
// Note: This solver method does not push/pop states on the stack.
// We continue to work on the top element of the stack.
// Message
if (loglevel>0)
{
logger << "Planning demand '" << l->getName() << "' (" << l->getPriority()
<< ", " << l->getDue() << ", " << l->getQuantity() << ")";
if (!data->constrainedPlanning || !data->getSolver()->isConstrained())
logger << " in unconstrained mode";
logger << endl;
}
// Unattach previous delivery operationplans, if required.
if (data->getSolver()->getErasePreviousFirst())
{
// Locked operationplans will NOT be deleted, and a part of the demand can
// still remain planned.
const_cast<Demand*>(l)->deleteOperationPlans(false, data);
// Empty constraint list
const_cast<Demand*>(l)->getConstraints().clear();
}
// Track constraints or not
data->logConstraints = (getPlanType() == 1);
// Determine the quantity to be planned and the date for the planning loop
double plan_qty = l->getQuantity() - l->getPlannedQuantity();
Date plan_date = l->getDue();
if (plan_qty < ROUNDING_ERROR || plan_date == Date::infiniteFuture)
{
if (loglevel>0) logger << " Nothing to be planned." << endl;
data->pop();
return;
}
if (plan_qty < l->getMinShipment()) plan_qty = l->getMinShipment();
// Temporary values to store the 'best-reply' so far
double best_q_qty = 0.0, best_a_qty = 0.0;
Date best_q_date;
// Select delivery operation
Operation* deliveryoper = l->getDeliveryOperation();
// Handle invalid or missing delivery operations
{
string problemtext = string("Demand '") + l->getName() + "' has no delivery operation";
Problem::const_iterator j = Problem::begin(const_cast<Demand*>(l), false);
while (j!=Problem::end())
{
if (&(j->getType()) == ProblemInvalidData::metadata
&& j->getDescription() == problemtext)
break;
++j;
}
if (!deliveryoper)
{
// Create a problem
if (j == Problem::end())
new ProblemInvalidData(const_cast<Demand*>(l), problemtext, "demand",
l->getDue(), l->getDue(), l->getQuantity());
// Abort planning of this demand
throw DataException("Demand '" + l->getName() + "' can't be planned");
}
else
// Remove problem that may already exist
delete &*j;
}
// Planning loop
do
{
// Message
if (loglevel>0)
logger << "Demand '" << l << "' asks: "
<< plan_qty << " " << plan_date << endl;
// Store the last command in the list, in order to undo the following
// commands if required.
CommandManager::Bookmark* topcommand = data->setBookmark();
// Plan the demand by asking the delivery operation to plan
double q_qty = plan_qty;
data->state->curBuffer = NULL;
data->state->q_qty = plan_qty;
data->state->q_date = plan_date;
data->planningDemand = const_cast<Demand*>(l);
data->state->curDemand = const_cast<Demand*>(l);
data->state->curOwnerOpplan = NULL;
deliveryoper->solve(*this,v);
Date next_date = data->state->a_date;
if (data->state->a_qty < ROUNDING_ERROR
&& plan_qty > l->getMinShipment() && l->getMinShipment() > 0)
{
bool originalLogConstraints = data->logConstraints;
data->logConstraints = false;
try
{
// The full asked quantity is not possible.
// Try with the minimum shipment quantity.
if (loglevel>1)
logger << "Demand '" << l << "' tries planning minimum quantity " << l->getMinShipment() << endl;
data->rollback(topcommand);
data->state->curBuffer = NULL;
data->state->q_qty = l->getMinShipment();
data->state->q_date = plan_date;
data->state->curDemand = const_cast<Demand*>(l);
deliveryoper->solve(*this,v);
if (data->state->a_date < next_date)
next_date = data->state->a_date;
if (data->state->a_qty > ROUNDING_ERROR)
{
// The minimum shipment quantity is feasible.
// Now try iteratively different quantities to find the best we can do.
double min_qty = l->getMinShipment();
double max_qty = plan_qty;
double delta = fabs(max_qty - min_qty);
while (delta > data->getSolver()->getIterationAccuracy() * l->getQuantity()
&& delta > data->getSolver()->getIterationThreshold())
{
// Note: we're kind of assuming that the demand is an integer value here.
double new_qty = floor((min_qty + max_qty) / 2);
if (new_qty == min_qty)
{
// Required to avoid an infinite loop on the same value...
new_qty += 1;
if (new_qty > max_qty) break;
}
if (loglevel>0)
logger << "Demand '" << l << "' tries planning a different quantity " << new_qty << endl;
data->rollback(topcommand);
data->state->curBuffer = NULL;
data->state->q_qty = new_qty;
data->state->q_date = plan_date;
data->state->curDemand = const_cast<Demand*>(l);
deliveryoper->solve(*this,v);
if (data->state->a_date < next_date)
next_date = data->state->a_date;
if (data->state->a_qty > ROUNDING_ERROR)
// Too small: new min
min_qty = new_qty;
else
// Too big: new max
max_qty = new_qty;
delta = fabs(max_qty - min_qty);
}
q_qty = min_qty; // q_qty is the biggest Q quantity giving a positive reply
if (data->state->a_qty <= ROUNDING_ERROR)
{
if (loglevel>0)
logger << "Demand '" << l << "' restores plan for quantity " << min_qty << endl;
// Restore the last feasible plan
data->rollback(topcommand);
data->state->curBuffer = NULL;
data->state->q_qty = min_qty;
data->state->q_date = plan_date;
data->state->curDemand = const_cast<Demand*>(l);
deliveryoper->solve(*this,v);
}
}
}
catch (...)
{
data->logConstraints = originalLogConstraints;
throw;
}
data->logConstraints = originalLogConstraints;
}
// Message
if (loglevel>0)
logger << "Demand '" << l << "' gets answer: "
<< data->state->a_qty << " " << next_date << " "
<< data->state->a_cost << " " << data->state->a_penalty << endl;
// Update the date to plan in the next loop
Date copy_plan_date = plan_date;
// Compare the planned quantity with the minimum allowed shipment quantity
// We don't accept the answer in case:
// 1) Nothing is planned
// 2) The planned quantity is less than the minimum shipment quantity
// 3) The remaining quantity after accepting this answer is less than
// the minimum quantity.
if (data->state->a_qty < ROUNDING_ERROR
|| data->state->a_qty + ROUNDING_ERROR < l->getMinShipment()
|| (plan_qty - data->state->a_qty < l->getMinShipment()
&& fabs(plan_qty - data->state->a_qty) > ROUNDING_ERROR))
{
if (plan_qty - data->state->a_qty < l->getMinShipment()
&& data->state->a_qty + ROUNDING_ERROR >= l->getMinShipment()
&& data->state->a_qty > best_a_qty )
{
// The remaining quantity after accepting this answer is less than
// the minimum quantity. Therefore, we delay accepting it now, but
// still keep track of this best offer.
best_a_qty = data->state->a_qty;
best_q_qty = q_qty;
best_q_date = plan_date;
}
// Delete operationplans - Undo all changes
data->rollback(topcommand);
// Set the ask date for the next pass through the loop
if (next_date <= copy_plan_date
|| (!data->getSolver()->getAllowSplits() && data->state->a_qty > ROUNDING_ERROR)
|| (data->state->a_qty > ROUNDING_ERROR && plan_qty - data->state->a_qty < l->getMinShipment()
&& plan_qty - data->state->a_qty > ROUNDING_ERROR))
{
// Oops, we didn't get a proper answer we can use for the next loop.
// Print a warning and simply try one day later.
if (loglevel>0)
logger << "Warning: Demand '" << l << "': Lazy retry" << endl;
plan_date = copy_plan_date + data->getSolver()->getLazyDelay();
}
else
// Use the next-date answer from the solver
plan_date = next_date;
}
else
{
// Accepting this answer
if (data->state->a_qty + ROUNDING_ERROR < q_qty)
{
// The demand was only partially planned. We need to do a new
// 'coordinated' planning run.
// Delete operationplans created in the 'testing round'
data->rollback(topcommand);
// Create the correct operationplans
if (loglevel>=2)
logger << "Demand '" << l << "' plans coordination." << endl;
data->getSolver()->setLogLevel(0);
double tmpresult = 0;
try
{
for(double remainder = data->state->a_qty;
remainder > ROUNDING_ERROR; remainder -= data->state->a_qty)
{
data->state->q_qty = remainder;
data->state->q_date = copy_plan_date;
data->state->curDemand = const_cast<Demand*>(l);
data->state->curBuffer = NULL;
deliveryoper->solve(*this,v);
if (data->state->a_qty < ROUNDING_ERROR)
{
logger << "Warning: Demand '" << l << "': Failing coordination" << endl;
break;
}
tmpresult += data->state->a_qty;
}
}
catch (...)
{
data->getSolver()->setLogLevel(loglevel);
throw;
}
data->getSolver()->setLogLevel(loglevel);
data->state->a_qty = tmpresult;
if (tmpresult == 0) break;
}
// Register the new operationplans. We need to make sure that the
// correct execute method is called!
if (data->getSolver()->getAutocommit())
{
data->getSolver()->scanExcess(data);
data->CommandManager::commit();
}
// Update the quantity to plan in the next loop
plan_qty -= data->state->a_qty;
best_a_qty = 0.0; // Reset 'best-answer' remember
}
}
// Repeat while there is still a quantity left to plan and we aren't
// exceeding the maximum delivery delay.
while (plan_qty > ROUNDING_ERROR
&& ((data->getSolver()->getPlanType() != 2 && plan_date < l->getDue() + l->getMaxLateness())
|| (data->getSolver()->getPlanType() == 2 && !data->constrainedPlanning && plan_date < l->getDue() + l->getMaxLateness())
|| (data->getSolver()->getPlanType() == 2 && data->constrainedPlanning && plan_date == l->getDue())
));
// Accept the best possible answer.
// We may have skipped it in the previous loop, awaiting a still better answer
if (best_a_qty > 0.0 && data->constrainedPlanning)
{
if (loglevel>=2) logger << "Demand '" << l << "' accepts a best answer." << endl;
data->getSolver()->setLogLevel(0);
try
{
for (double remainder = best_q_qty;
remainder > ROUNDING_ERROR && remainder > l->getMinShipment();
remainder -= data->state->a_qty)
{
data->state->q_qty = remainder;
data->state->q_date = best_q_date;
data->state->curDemand = const_cast<Demand*>(l);
data->state->curBuffer = NULL;
deliveryoper->solve(*this,v);
if (data->state->a_qty < ROUNDING_ERROR)
{
logger << "Warning: Demand '" << l << "': Failing accepting best answer" << endl;
break;
}
}
if (data->getSolver()->getAutocommit())
{
data->getSolver()->scanExcess(data);
data->CommandManager::commit();
}
}
catch (...)
{
data->getSolver()->setLogLevel(loglevel);
throw;
}
data->getSolver()->setLogLevel(loglevel);
}
// Reset the state stack to the position we found it at.
while (data->state > mystate) data->pop();
}
catch (...)
{
// Clean up if any exception happened during the planning of the demand
while (data->state > mystate) data->pop();
data->rollback(topcommand);
throw;
}
}