Open 6bdad4c1-1e26-4f2f-a442-a01a2292c181 opened 14 years ago
In case of a SparseGraph
, the SparseGraphBackend
has self._cg
and self._cg_rev
. They can merely be swapped. In the case of a NetworkX
graph, I believe their digraphs have succ
and pred
dictionaries storing the adjacency information. These could also be swapped. For DenseGraphs
, this is equivalent to switching between row ordering and column ordering in bitsets, which is inevitably costly.
This function should be rewritten much more efficiently :
First, there should be a way to reverse the arcs "in place" ( without building a copy, by modifying the current graph -- I do not know if the expression used in frech translates in this case ). Such a function, for c_graphs, should be written in Cython and consist in the case of sparse graph in reverting the two copies kept of the graph. In the end, this function should consist in an (optional) copy of the graph (=fast) and a call to the functionr reverting the arcs ( O(1) )
If possible and if deemed useful, the same should be made for NetworkX graphs.
CC: @rlmill
Component: graph theory
Issue created by migration from https://trac.sagemath.org/ticket/7720