Tensorflow C++ - Avoiding overlapping names when loading two graphs - c++

Question

As a result of earlier parts of a pipeline, I have two .pb files containing a frozen optimized Tensorflow graph for slightly different architectures for the same production model. I would like to load them both in the same C++ program into the same session for inference, but of course the graph nodes have lots of conflicting names.
In Python, I was under the impression that you could load the two graphs under into the same session within different variable scopes, but in C++ I'm not sure how to do this.
So I've been doing essentially the following, which seems to work, but also seems a bit hacky and fragile to do manually like this, particularly handling the caret in the name for control edges. Is this a reasonable way to do this, and/or is there a pre-existing function in the C++ api that I can call instead that accomplishes the same thing?
Status status;
GraphDef graphDef1;
GraphDef graphDef2;
status = ReadBinaryProto(Env::Default(), string("frozen_graph_optimized1.pb"), &graphDef1);
CHECK_STATUS(status,"reading graph1");
status = ReadBinaryProto(Env::Default(), string("frozen_graph_optimized2.pb"), &graphDef2);
CHECK_STATUS(status,"reading graph2");
auto addPrefixToGraph = [](GraphDef& graphDef, const string& prefix) {
for(int i = 0; i<graphDef.node_size(); ++i)
{
auto node = graphDef.mutable_node(i);
string* name = node->mutable_name();
*name = prefix + *name;
int inputSize = node->input_size();
for(int j = 0; j<inputSize; ++j) {
string* inputName = node->mutable_input(j);
if(inputName->size() > 0 && (*inputName)[0] == '^')
*inputName = "^" + prefix + inputName->substr(1);
else
*inputName = prefix + *inputName;
}
}
};
addPrefixToGraph(graphDef1,"g1/");
addPrefixToGraph(graphDef2,"g2/");
status = session->Create(graphDef1);
CHECK_STATUS(status,"adding graph1 to session");
status = session->Extend(graphDef2);
CHECK_STATUS(status,"adding graph2 to session");