Greetings,
Following the closing of #7965 which dealt with IDA* implementation, I wonder whether there is a necessity for such algorithm in NetworkX. Generally, combinatorial problems that are solved with heuristic search represent the state space as start state and a successor function which allows the algorithm to explore the graph. Since A* saves all nodes discovered, its memory usage is linear in the state space. This can be a problem with graphs that do not fit into memory, hence IDA* which is linear in solution length. But since NetowrkX assumes the entire state space fits into memory, I doubt IDA* will provide any benefit over A*. I would also like to point out there quite a few issues (in my opinion) with the suggested stack based approach which makes the algorithm not exactly IDA* while also being less efficient. But, since the PR was closed, I did not write a comment there.
There exists an algorithm called BAE*¹² which is a bidirectional heuristic algorithm which assumes heuristic consistency (unlike A*, where this assumption was fixed in #3508). It does require two heuristics, the regular one from node to goal, and the other is from start to node. Would there be interest for such algorithm? It's mostly a question of how often NetworkX is used as a heuristic search framework, since BAE* is significantly better than A*, but mostly on exponential state spaces.