✨ Expose DD serialization/deserialization in Python API

[mohammadSafwanAlam]

Problem Statement

MQT Core exposes a useful Python API for constructing and manipulating DDs, but compressed DD persistence currently appears unavailable from Python. The available
Python paths are either dense (get_matrix / from_matrix) or visualization-oriented (to_dot), both of which are unsuitable for persisting large compressed DDs.

The C++ DD package already supports native serialization/deserialization. Would you be open to exposing that through mqt.core.dd, ideally as bytes-based APIs?

Proposed API shape:

  data = matrix_dd.to_bytes(binary=True)
  restored = dd_package.matrix_from_bytes(data, binary=True)

  data = serialize_matrix_dd(matrix_dd, binary=True)
  restored = deserialize_matrix_dd(dd_package, data, binary=True)

The bytes-based form avoids filesystem policy questions and lets downstream packages manage metadata/versioning themselves. This would let Python users cache verified
DD contracts without materializing exponential dense matrices or maintaining separate pybind shims.

Proposed Solution

Proposed API shape:

  data = matrix_dd.to_bytes(binary=True)
  restored = dd_package.matrix_from_bytes(data, binary=True)

  data = serialize_matrix_dd(matrix_dd, binary=True)
  restored = deserialize_matrix_dd(dd_package, data, binary=True)

[burgholzer]

Hey @mohammadSafwanAlam 👋🏼
Thanks for opening up this issue and for the feature suggestion.
What you propose sounds reasonable and we'd be open to accepting a PR that adds this feature.

I like the to_bytes signature for serialization as an additional member function of the VectorDD and MatrixDD classes that are exposed.

As for the deserialization, one could probably also do something like

VectorDD.from_bytes(dd_package, data, binary=True)
MatrixDD.from_bytes(dd_package, data, binary=True)

That feels fairly clean to me and feels like a nice symmetry to VectorDD.to_bytes etc.

Does that make sense?