Copying objects in C++/CLI and message passing in multithreading - multithreading

Question

I'm trying to transfer a command line code that I have to a more visual program with a
GUI to enable easier use. The original code was in C++, so I'm using Visual C++ that is
available in Visual Studio Express 2012, but I have problems understanding the "new"
managed C++/CLI way of handling objects. Being new to CLI and managed C++, I was wondering
if someone can explain what I am doing wrong, and why it doesn't work. Now here is a
description of the code and the problem.
The program is essentially an optimization program:
There are multiple boxes (modes) in a system, each mode, depending on its type has a
few numerical coefficients that control its behavior and the way it responds to outside
excitation.
The program asks the user to specify the number of boxes and the type of each box.
Then tries to find the numerical coefficients that minimize the difference between
the system response with those obtained experimentally.
So, the UI has means for user to open the experimental result files, specify the number
of modes, and specify the type of each mode. Then, the user can initiate the processing
function by clicking on a start button, that initiates a background worker.
Following the example given in MSDN, I created a class that performs the work:
ref class curveFit
{
public: ref class CurrentState{
public:
int percentage;
int iterationNo;
int stage;
bool done;
multimode systemModel;
};
public:
int modes;
int returncode;
array<double> ^expExcitations;
array<double> ^expResults;
multimode systemModel;
private:
void fcn(int, int, double*, double*, int*);
double totalError(std::vector<double> &);
public:
delegate void fcndelegate(int, int, double*, double*, int*);
public:
curveFit(void);
curveFit^ fit(System::ComponentModel::BackgroundWorker^, System::ComponentModel::DoWorkEventArgs^, Options^);
};
multimode is just a container class: a list of different boxes.
ref class multimode
{
private:
Collections::Generic::List<genericBoxModel ^>^ models;
int modes;
public:
multimode(void);
multimode(const multimode%);
int modeNo(void);
void Add(genericBoxModel^);
void Clear();
genericBoxModel^ operator[](int);
multimode% operator=(const multimode%);
double result(double);
bool isValid();
std::vector<double> MapData();
void MapData(std::vector<double> &);
};
multimode::multimode(void)
{
models = gcnew Collections::Generic::List<genericBoxModel ^>();
modes = 0;
}
multimode::multimode(const multimode% rhs)
{
models = gcnew Collections::Generic::List<genericBoxModel ^>();
for(int ind = 0; ind < rhs.modes; ind++)
models->Add(rhs.models[ind]);
modes = rhs.modes;
}
int multimode::modeNo(void)
{
return modes;
}
void multimode::Add(genericBoxModel^ model)
{
models->Add(model);
modes++;
}
void multimode::Clear()
{
models->Clear();
modes = 0;
}
genericBoxModel^ multimode::operator[](int ind)
{
return models[ind];
}
multimode% multimode::operator=(const multimode% rhs)
{
models->Clear();
for(int ind = 0; ind < rhs.modes; ind++)
models->Add(rhs.models[ind]);
modes = rhs.modes;
return *this;
}
double multimode::result(double excitation)
{
double temp = 0.0;
for(int ind = 0; ind < modes; ind++)
temp += models[ind]->result(excitation);
return temp;
}
bool multimode::isValid()
{
bool isvalid = true;
if(modes < 1)
return false;
for(int ind = 0; ind < modes; ind++)
isvalid = (isvalid && models[ind]->isValid());
return isvalid;
}
std::vector<double> multimode::fullMap()
{
//Map the model coefficients to a vector of doubles
...
}
void multimode::fullMap(std::vector<double> &data)
{
//Map a vector of doubles to the model coefficients
...
}
and genericBoxModel is an abstract class that all box models are based on.
The curvefit::fit function does the optimization based on the options passed to it:
curveFit^ curveFit::fit(System::ComponentModel::BackgroundWorker^ worker, System::ComponentModel::DoWorkEventArgs^ e, Options^ opts)
{
fcndelegate^ del = gcnew fcndelegate(this, &curveFit::fcn);
std::vector<double> data;
CurrentState^ state = gcnew CurrentState;
state->done = false;
state->stage = 0;
state->percentage = 0;
state->systemModel = systemModel;
worker->ReportProgress(state->percentage, state);
switch(opts->optimizationMethod)
{
case 0:
while(iterationNo < maxIterations)
{
data = systemModel.MapData();
OptimizationMethod0::step(some_parameters, data, (optmethods::costfunction)Runtime::InteropServices::Marshal::GetFunctionPointerForDelegate(del).ToPointer());
systemModel.MapData(data);
iterationNo++;
state->percentage = 0;
state->systemModel = systemModel;
worker->ReportProgress(state->percentage, state);
}
...
}
}
I'm passing the system model inside the state so that I can display the results of the
latest step on the screen, which doesn't work, but that is another question :-)
The start button calls the curvefit::fit function after initializing the system model:
private: System::Void btnStart_Click(System::Object^ sender, System::EventArgs^ e) {
systemModel.Clear();
for(int mode = 0; mode < modes; mode++)
{
switch(model)
{
case 0:
systemModel.Add(gcnew model0);
systemModel[mode]->coefficients[0] = 100.0 / double(mode + 1);
...
break;
...
}
}
btnStart->Enabled = false;
stStatusText->Text = "Calculating!";
Application::UseWaitCursor = true;
curveFit^ cf = gcnew curveFit;
fitCurve->RunWorkerAsync(cf);
}
private: System::Void fitCurve_DoWork(System::Object^ sender, System::ComponentModel::DoWorkEventArgs^ e) {
System::ComponentModel::BackgroundWorker^ worker;
worker = dynamic_cast<System::ComponentModel::BackgroundWorker^>(sender);
curveFit^ cf = safe_cast<curveFit^>(e->Argument);
cf->expExcitations = gcnew array<double>(expExcitations.Count);
expExcitations.CopyTo(cf->expExcitations);
cf->expResults = gcnew array<double>(expResults.Count);
expResults.CopyTo(cf->expResults);
cf->systemModel = systemModel;
cf->modes = modes;
e->Result = cf->fit(worker, e, options);
}
This works perfectly! But, in order to make the optimization process faster and more
successful, I wanted to use the results of previous optimizations as the initial guess
for the next run (if possible):
multimode oldmodel(systemModel);
systemModel.Clear();
for(int mode = 0; mode < modes; mode++)
{
switch(model)
{
case 0:
if(mode < oldmodel.modeNo() && oldmodel.isValid() && (oldmodel[mode]->model == 0))
systemModel.Add(oldmodel[mode]);
else
{
systemModel.Add(gcnew model0);
systemModel[mode]->coefficients[0] = 100.0 / double(mode + 1);
...
}
break;
...
Now, my problem is, after this change, it seems that the messages don't get passed
correctly: the first time the start button is clicked everything functions as it should,
but from then on, if the statement systemModel.Add(oldmodel[mode]); gets executed,
results remain the same as the initial guesses, and don't get updated after the fit
function is called.
So, why should these two lines(Add(oldmodel[mode]) and Add(gcnew model0)) give
such different results?