swarms/ideas/63-refactor-geth-packages.md

4.7 KiB

Preamble

Idea: 63-refactor-geth-packages
Title: Refactor status-go geth packages
Status: Done
Created: 2017-12-20

Summary

geth/ packages require refactoring as they are tightly coupled, hard to maintain and extend. In this idea, we would like to fix all these issues and make status-go architecture easier to work with for new core members and external contributors.

Swarm Participants

  • Lead Contributor: @adambabik
  • Contributors:
    • @dshulyak (20h/week)
    • @adriacidre
  • Evaluator: @adambabik

Product Overview

We have a bunch of problems related to the geth/ packages:

  • NodeManager contains a lot of complexity, especially related to starting and stopping a node,
  • geth/account unnecessarily depends on packages like geth/rpc,
  • decouple geth/transactions from geth/account,
  • remove huge, unnecessary interfaces,
  • get rid of geth/common package.

In this idea, our goal is to come up with a better architecture design for these packages, removing unnecessary dependencies and huge interfaces. Overall, the whole repo will be easier to browse through for audit companies and external contributors.

Product Description

It does not add any new features from the user perspective but it's a part of making status-go better tested, easier to extend and reason about, also by external contributors.

Requirements & Dependencies

All existing tests should pass. If we need to change the interface in library.go, we need to communicate that and work with status-react team to figure out the best way to do it.

Minimum Viable Product

Goal Date: 2018-02-02

Description: Minimal implementation of Node that can be started and stopped and an RPC client can be connected to it. It should be preceded with code examples how the new interface of Node will look like.

Iteration 1

Goal Date: 2018-02-09

Description: The rest of the Node features is implemented.

Iteration 2: AccountManager cleanup

Goal Date: 2018-03-23

Description: AccountManager should know nothing about Whisper or node.NodeManager, so on this iteration we will remove these dependencies.

Iteration 3: TransactionManager cleanup

Goal Date: 2018-03-30

Description: TransactionManager shouldn't directly depend on AccountManager and NodeManager. In this iteration we want to remove these direct dependencies.

Iteration 4: common.NodeManager removal

Goal Date: 2018-03-30

Description: common.NodeManager interface and all its mocks have to be removed from status-go. This interface is too large and is not useful at all.

Iteration 5: geth/common package removal

Goal Date: 2018-04-06

Description: get/common package contains everything. In this iteration, the goal is to remove it complete or at least move most of the functions to appropriate packages.

Iteration 6: Simplify node.NodeManager API

Goal Date: 2018-04-13

Description:

  1. Change its name to something more appropriate as it does not manage multiple nodes
  2. Rename methods indicating there is only support for one node: StartNode -> Start, StopNode -> Stop etc.
  3. Unify EnsureSync on t/utils/utils.go and StatusManager

Iteration 6: A PR in status-react

Goal Date: 2018-04-13

Description: A PR in status-react is created with updated status-go version.

Success Metrics

  1. A tree of module dependencies is hugely simplified,
  2. All tests pass,
  3. There are no race conditions in tests.

Exit criteria

  1. All iterations are done. There is no threat that they can't be finished.
  2. New version of status-go is merged into status-react develop branch.

Summary after finish

The first and second success metrics have been achieved. We managed to reduce dependencies between packages and get rid of common package. No change did negatively impact tests reliability and all tests pass properly. With regard to the third metric, we haven't implemented automated race condition checks yet so it's hard to tell but there are no race condition issues reported so far.

Copyright and related rights waived via CC0.