Fix GetString memory leak

[markaren]

GetString eats memory. This PR fixes it.

Closes #69

[markaren]

Mhh. Indeed this fails with PyFMI, but works with FMI4j... Don't know what to do, because it does leak. PyUnicode_AsEncodedString returns a new reference, which is never decrefed.

[markaren]

I'm starting to think this could be a FMI4j bug? I need to check the standard on string handling.

For refernce, this work with PyFMI

PyObject* value = PyList_GetItem(refs, i);
PyObject* strValue = PyUnicode_AsEncodedString(value, "utf-8", nullptr);
char* src = PyBytes_AsString(strValue);
char* dst = reinterpret_cast<char*>(malloc(strlen(src) + 1));
strcpy_s(dst, strlen(src) + 1, src);
values[i] = dst;
Py_DECREF(strValue);

But I dunno if thats correct. It uses less memory in FMI4j, but still seams to leak. Is the caller responsible for deleting dst ?

[fcollonval]

Hey I don't think this is a bug of FMI4J.

I have seen there is a built-in package for tracing Python allocation. Maybe you can get more inside with it?

https://docs.python.org/3/library/tracemalloc.html

Not: there is another one to correct too 😉

PythonFMU/pythonfmu/pythonfmu-export/headers/pythonfmu/PyException.hpp

Line 27 in 6c9daaf

oss << PyBytes_AsString(PyUnicode_AsEncodedString(pRepr, "utf-8", nullptr));

[fcollonval]

From the specification:

The strings returned by fmi2GetString must be copied in the target environment because the allocated memory for these strings might be deallocated by the next call to any of the fmi2 interface functions or it might be an internal string buffer that is reused.

So from your latest proposal, could it be that the copied string dst are not released neither?

[markaren]

dst leaks for sure.

So how I read your excerpt is that we must deallocate the string in another FMI call. Say the next invocation of GetString. Ok. Makes the code more complex, but doable.

[markaren]

This should do it !

[markaren]

oss << PyBytes_AsString(PyUnicode_AsEncodedString(pRepr, "utf-8", nullptr));

I'll fix that once #71 is merged

[markaren]

We should probably release a patch including the two latest bugfixes soon.

[fcollonval]

Thanks for catching and fixing this one.

We are almost good. I have just one concern about the buffer clearance not being called within a GIL captured lambda (see comments).

[fcollonval]

Thank you!