Change struct to serializable-struct.

[LeifAndersen]

This library is a graph library, the serialize library is designed to serialize cyclic data structures. These two libraries should be talking to each other.

[mkohlhaas]

I think persistence shouldn't be mixed into the existing graph structs in the sense of making it a default behavior. If one wants to have persistence on top of the existing structs one could define a persistent struct with define-serializable-struct (or something similar) and specify the existing graph struct as a super struct.

There could be new graph types that don't use a Racket struct at all but store the graph in e.g. files, databases or any kind of external entities.

[stchang]

I agree (which is probably why I never merged this ticket). Given the generic implementation, a representation using serializable structs should be straightforward to add. This way, existing code is unaffected.

[LeifAndersen]

To clarify: I mostly was just saying the built in graph implementations in this library can be serializable with no downside. Where as not doing it requires a bunch of redundant work for any end users who want to serialize their graphs. This patch will not break any code, unless they were relying on the serialize? functions for the default graphs Stephen made to return #f. Something like:

(define graph (make-directed-graph '()))
(if (serializable? graph)
    <<launch-the-missles>>
    <<something-else>>)

@mkohlhaas I presume by persistence you are actually meaning serializability? I ask for clarification as serializability and persistence are fairly unrelated.

@stephen and @mkohlhaas Anyway, in principle, yes, anyone could construct a new struct type and make them serializable and also meet the graph API. But that would require them to implement 20 or so methods. Which just seems silly given that the implementations would be exactly the same as the default implementations built into this graph. It would also make for a bad pull request.

[bennn]

I agree with Leif, serializability would be a very useful default and seems unlikely to break things.

[mkohlhaas]

Hi Leif,

yes that's true. I mean serialization for the purpose of persistence. Would you use it for something else?

I thought one could create a new serializable struct referring to an existing structure without reimplementing all the methods. Cf maybe-super.

It's awesome having a discussion with all you Racket masters I know from YouTube! 😀 I should go to the next Racket conf and make some selfies with you guys.

Sidenote: I'm pretty new to Racket. No way I could keep up with you guys. I'm just divulging my naive opinions.

[stchang]

@mkohlhaas Thanks for resuming this conversation.

@bennn Did you have a particular use case in mind? Do other data structures use serializable structs?

@LeifAndersen Did you initially create this PR for the feature-specific profiler?

[bennn]

No, I don't have a need for this right now. But it would be useful the same way that serializing lists & hashes can be useful.

If graphs were serializable, then I could have used them instead of adjacency lists to represent module-dependence graphs for the gradual typing benchmarks (link).