When is registration of new sockets allowed? about dmtcp

[gc00]

@rohgarg: The implicit principle was that a plugin could create a new resource at checkpoint time if it removes the resource during checkpoint or at least before resume or restart. But this principle needs to be refined. As you've observed, we also need to extend the principle to say whether DMTCP is allowed to be aware of these new resources, or should the new resource be invisible to DMTCP. Making that precise will take some thought. There are several corner cases to worry about.

I think it would be good to document this (at least for the core plugins) and add some assertions in the code to make debugging easier.

Very impressive. This is also what @karya0 and I had independently concluded when finishing up the plugin document of @karya0, but we ran out of time. I'd be glad to brainstorm with you about what would be the right way to do that.

[rohgarg]

I can imagine an optional "debug" plugin that defines wrappers with assertions for violations of invariants. This plugin would get loaded last in the library search order. This way we avoid having to touch any of the existing code. Generation of a new wrapper could be a matter of adding a few lines to a template file.

[gc00]

I can imagine an optional "debug" plugin that defines wrappers with assertions ...
I like that idea a lot. I hadn't thought of that possibility!