Unpickling a graph can change the backend

[ed359]

Steps To Reproduce

In a standard install of sagemath 10.4, run the following code.

G = Graph(1, data_structure="dense")
H = loads(dumps(G))

print(G == H) # True
print(isinstance(G._backend, sage.graphs.base.dense_graph.DenseGraphBackend)) # True
print(isinstance(H._backend, sage.graphs.base.sparse_graph.SparseGraphBackend)) # True

On my system, I get

True
True
True

Expected Behavior

The backend data structure of a graph is preserved upon pickling and unpickling.

Actual Behavior

The backend is not preserved and falls back to the default (sparse) backend.

Additional Information

This is a bug, as I want to use process-level parallelism in some loops that generate graphs. With standard tools such as joblib, multiprocessing, concurrent.futures, this requires some pickling and unpickling. But my code needs the backends of the graphs to be the DenseGraph implementation and this bug prevents me from using these tools.

Environment

• OS: MacOS Sequoia 15.0.1
• Sage Version: 10.4 (binary distribution from 3-manifolds project)

Checklist

• I have searched the existing issues for a bug report that matches the one I want to file, without success.
• I have read the documentation and troubleshoot guide

[maxale]

Related #35592

[dcoudert]

See #38919 for a possible fix.