Emit returning errors

[whyrusleeping]

I think it provides a pretty poor UX to have Emit return errors. It complicates the calling code a bit (what do we even do with the errors? we cant use SetError). Lets discuss how to address this.

[Stebalien]

Proposal: Entirely separate errors and values.

1. Make SetError set the final error (and close the request).
2. Use Emit(err) to send "informational" errors to the user.

This would also solve #31 as, in the HTTP case, RCP-level errors would be sent via trailer headers only (never inline as content).

[dignifiedquire]

On that note, it would be really nice if Run could return an error which gets just passed to SetError if it is set, for the simple error case, the code becomes much more aligned with the way regular error handling is written:

func(req *cmds.Request, re cmds.ResponseEmitter, env cmds.Environment) {
  err := doSth()
  if err != nil {
    re.SetError(err, cmdkit.ErrNormal)
    return
  }
  re.Emit("cool")
}
// vs
func(req *cmds.Request, re cmds.ResponseEmitter, env cmds.Environment) error {
  err := doSth()
  if err != nil {
    return err
  }
  re.Emit("cool")
  return nil
}

[keks]

May I propose

func(req *cmds.Request, re cmds.ResponseEmitter, env cmds.Environment) error {
  err := doSth()
  if err != nil {
    return err
  }
  return re.Emit("cool")
}

That way we don't lose the error and the code is still clean.

ping @whyrusleeping