I have a C program that calls a C++ wrapper around a large C++ library. One of the functions of the C++ library is to provide some predictions, and those predictions will be used within C. I’m wondering where I should free or delete the memory. Here is some example code.
cpp_wrapper.cpp
#include "cpp_wrapper.h"
uint8_t * run_model(struct_with_model model_struct, const double * input_data) {
// the model_struct is new'd in C++ in an earlier method
// I have a method to convert the input data to std::vector for use in C++ code
std::vector<double> data = get_data(input_data);
model_struct->model->LoadData(data);
model_struct->model->run();
std::vector<uint8_t> predictions = model_struct->model->GetPredictions();
std::unique_ptr<uint8_t[]> classValues (new uint8_t[predictions.size()]);
memcpy(classValues.get(), predictions.data(), sizeof(uint8_t) * predictions.size());
model_struct->model->ClearData(); //clears data from model for future runs, if necessary
return classValues.release()
}
void model_delete(model_struct) {
// method to delete the model_struct when necessary
delete model_struct;
}
I have a cpp_wrapper.h header file that declares these functions and calls extern C as necessary. On the C side, c_code.c:
#include "cpp_wrapper.h"
/*
There's a bunch of code here for ingesting data, initializing the model_struct, etc
*/
uint8_t * predictions = run_model(model_struct, input_data);
// Do stuff with predictions, as necessary
model_delete(model_struct);
free(predictions); // HERE is the question
I new the initial variable classValues in the C++ code, however I release the std::unique_ptr as the return of the C++ run_model function. My understanding was that when I release the memory to C, that the C code was responsible for freeing the memory. Is this the case? Should I have a C++ method to delete the predictions variable (from C), much like I have a C++ method to delete the model_struct variable? I’m confused about how to best manage the memory here. I was advised to not use malloc (or calloc) in C++, but since I’m passing the values back to C, maybe this is a better option…? I’m not sure.
>Solution :
The simplest solution is to divide the C++ call into two calls. The first one determines how big the needed buffer is. Then the memory can be allocated in the C part. In the second call the data is copied into this buffer. This way the memory management remains completely in the C part.
Alternatively, the C++ interface must provide a function to clear the memory. The C code calls this when the memory is to be released. The usual rule applies:
malloc->freenew->deletenew[]->delete[]
If I see this correctly, you are implementing the second approach at the moment, but are incorrectly using new[] -> delete!