I do not think after()  should be called with an error at all. We should bail off the plugin loading.

Why not, in this way the user can handle the error inside the after.
Example:
If plugin 1 fails => load plugin 2.

that should be in the register callback, not in after IMHO.

Agree, but at the moment this is the behaviour:

test('error', (t) => {
  t.plan(2)

  const server = { my: 'server' }
  const app = boot(server)

  app.use(function (s, opts, done) {
    done(new Error('err'))
  }, err => {
    t.ok(err)
    app.use(function (s, opts, done) {
      done()
    })
  })

  app.ready(function (err) {
    t.ok(err)
  })
})

As you can see, we have not that capability, because the error will arrive to the ready callback in every case.
With after I can "block" that error ina specific point, as you can see here:

test('error', (t) => {
  t.plan(3)

  const server = { my: 'server' }
  const app = boot(server)

  app.use(function (s, opts, done) {
    done(new Error('err'))
  }, err => {
    t.ok(err)
  })

  app.after(function (err) {
    t.ok(err)
    app.use(function (s, opts, done) {
      done()
    })
  })

  app.ready(function (err) {
    t.error(err)
  })
})

This is why yesterday I was saying that we should raise the error in the ready callback only if the error is not handled in the use callback.

OK for this behavior you are describing right know.

Sorry, I didn't understand, do you want me to implement this behaviour?

This is why yesterday I was saying that we should raise the error in the ready callback only if the error is not handled in the use callback.

This is why yesterday I was saying that we should raise the error in the ready callback only if the error is not handled in the use callback.

yes.