Synchronous plugin registration

[kanongil]

Hapi 9.x will require eg. inert to be registered before its file and directory handlers can be used.

This is fine but the new process seems a bit off, requiring all such routes to be embedded in the registration callback. Alternatively, the user would need to know that the registration is actually synchronous, and that the callback can be ignored and an empty function should be provided.

Ideally, it should just be something like:

var Hapi = require('hapi');
var Inert = require('inert');

var server = new Server();
server.connection(…);

server.register(Inert);

server.route(…);

instead of:

server.register(Inert, function(err) {

    if (err) {
        throw err;
    }

    server.route(…);
});

or:

server.register(Inert, function() {});

server.route(…);

[hueniverse]

This is already the case for any auth scheme or anything other than the three built-in plugins. I don't see this as a real problem. It might be annoying at first but eventually people will just get use to it (or better yet, glue will add built-in support directly).

[kanongil]

Why not remove the callback requirement from server.register and catch errors from such in the server.start callback? Plugin dependency after errors are already exposed there.

[hueniverse]

Because this is not the right way to solve the issue. hapi is designed to work with plugins and you are optimizing for the most basic use case.

[kanongil]

To me, designed to work with plugins, reads as: should be simple to use plugins, which my proposal aims at.

The most basic use case is the one new users are likely to be confronted with. Requiring nesting, or empty function arguments, is somewhat off-putting.

Anyway, I don't intend to fight on this, and the current implementation is fine for my private usage.

[hueniverse]

The only way this doesn't lead to future pain because people are just ignoring the callback is if we add a plugin attribute called sync which when true allows you to skip the callback in register. But this would require publishing all the plugins again. Would that work?

[kanongil]

It seems that you miss my point. My suggestion is that errors that should have triggered in a missing register callback are instead triggered in the server.start callback. As far as I can tell, this can be done without any breaking changes, and most of the tooling to implement this is already in place.

[hueniverse]

This is weird. You would call register and if there are errors, you would get them returned when you call start() or initialize()? This will be a nightmare to debug when things go wrong and your tests are not starting the server.