Confused by the general flow that capnproto expects

[chbug]

Hey,

I'm investigating using capnproto serialization as a replacement for protobufs. I've looked at the examples, but I'm confused by what the API is trying to push me to do. I see the point of the various Builders, to set up an object that has the appropriate memory layout. But then in my use-case I'd expect to pass around the structured data in read-only format, and only eventually send it / store it. So what do people do in similar cases? Turn the Builder into a reader? Is this equivalent to serializing and passing it to a reader? The types are very bulky and none of the examples seem to show the life of an object besides immediate serialization/deserialization...

Thanks for your help!

[dwrensha]

Have you tried passing around a message::Builder or message::TypedBuilder? Those have get_root_as_reader() methods that allow read-only access.

[chbug]

I think one thing that confused me is that I expected .into_reader() to be the standard way of "closing" a builder, but actually you'd keep the builder around as the "live" object, right?

Unfortunately I have many more questions :) Let me know if this is not the right place to ask the followups...

• I assumed the favored approach was to use the Typed variants of Builder and Reader but I notice that serialize::write_message takes a plain Builder, and even the addressbook example use the plain variants. What's the recommended API?
• What's the difference between message.init_root() and message.get_root() when I'm first filling in an object? For Lists, initializing the size seems required, but for an object it seems that both work?

Thanks for your patience...

[dwrensha]

I assumed the favored approach was to use the Typed variants of Builder and Reader but I notice that serialize::write_message takes a plain Builder, and even the addressbook example use the plain variants. What's the recommended API?

Maybe we should generalize the serialize::write_message() function so that it can accept any of message::Builder, message::TypedBuilder, message::Reader, or message::TypedReader. It looks like the message::ReaderSegments trait is all that write_message() actually needs, and message::Builder<A> already implements that.

What's the difference between message.init_root() and message.get_root() when I'm first filling in an object? For Lists, initializing the size seems required, but for an object it seems that both work?

There is some explanation of get vs init in the C++ docs: https://capnproto.org/cxx.html#structs

The main thing is that if there is an existing value in the field, get retains that value, while init clobbers it.

[chbug]

Thanks for the get vs init link, very useful.

What are the typed variants solving? Granted I only wrote toy code, but I don't see a big difference compared to the plain builder/reader. So if it's not buying much and is not fully integrated, I'm happy to stay clear, but I'm curious what the intent is.

(and after that I think I'll be good for a little while :P)

[chbug]

Ok, I could have tried a bit harder. So the typed variants are just here to help avoid passing around completely typeless message::Builder. In which case, indeed generalizing the various serializers to accept them would make sense, as I think one needs to do an into_inner() to do most things, which consumes the typed struct :)