GetPdf method in AmDiagGmm

[cffan]

Hi, I note that the GetPdf method in AmDiagGmm isn't implemented. Have you found a workaround? If so, could you briefly explain it here and I can help adding it. Thanks!

[bmilde]

What's the clif error message if you uncomment "# def GetPdf(self, pdf_index: int) -> DiagGmm"?

My guess is that it caused some problems with clif that weren't easy to resolve.

[asumagic]

FWIW, I am down a similar rabbit hole with Nnet.get_component.

I don't have a need for this particular function but it was easy enough for me to run a build with that GetPdf line uncommented, and the build succeeds. I cannot easily check if it actually works, though. However...
CLIF has seemingly decided that returning T& and const T& should result into a copy, which strikes to me as an astonishingly bad idea for default enforced implicit behavior.

This can be verified by checking at the generated wrapper code:

static PyObject* wrapGetPdf(PyObject* self, PyObject* args, PyObject* kw) {
// ...
  ::kaldi::DiagGmm ret0;
// ...
    ret0 = c->GetPdf(std::move(arg1));
// ...
  return Clif_PyObjFrom(std::move(ret0), {});

I am not familiar with what this class/method does, so could it be that the line was commented because this behavior was deemed misleading and undesirable in this case?
If so, adding a new C++ method returning a DiagGmm* and binding that instead may do the trick.
I don't know how concerned PyKaldi is about avoiding dangling references, though.