Add codable interface for python objects, (Ingress|Egress)Ports python bindings, and other elements required for multi-segment. #18
Conversation
There was a problem hiding this comment.
Unfortunately, this requested change is to revert segment::Object back to having a pure virtual name() method, which was neither your change nor mine, but I feel like we need to fix the change that ultimately caused protected member variable to pop up.
The other option would be to make m_name private on segment::Object and provided a protected setter. I'm maybe OK with duplicating the std::string storage of the name property for some object that already store a name like Egress/IngressPort, but it still feels silly, since segment::Object::get_node is still a pure virtual method, meaning the classes that derive from segment::Object can provide name storage if needed and provide the override on name().
I would expect you'd see hangs when they are not connected as the "stop" signal, i.e. the closure of the channel is not allowed to propagate. We could probably do more to detect this condition earlier, i.e. incomplete connections, but we need to work under the constraint that Morpheus allows for unconnected Sources as part of Stage construction, i.e. there are no true Sinks. |
Double checking that its not specific to py::object ingress/egress types; but, if its not, this does seem like an oversight. I think it makes sense to emit a warning if something isn't connected, but an infinite hang shouldn't be possible. |
drobison00
added
improvement
drobison00
changed the title
python/srf/_pysrf/include/pysrf/module_wrappers/shared_memory.hpp
Outdated
Show resolved
Hide resolved
|
@gpucibot merge |
ghost
merged commit 
Blocked by #75
pipeline.make_segment(segment, ["a", "b", "c"], ["x", "y", "z"], init)-- does not directly support ingress / egress only signatures because we can't differentiate on type and using something like argument count has an ambiguous interpretation.builder.make_edgebehavior in Python.