Changed the online event to `once` and set it to be bound every time the node goes offline.
The above changes handle the case where:
1) `embark run` runs and starts geth.
2) geth is killed manually
3) `embark blockchain` is run in separate process to restart geth
4) the `embark run` process detects this change and restarts the web3 provider and recompiles/deploys/builds
Every time `embark blochain` is restarted, an error is appended and all are emitted from the `eth-block-tracker`. This is a bug but can't figure out where it originates. The downside is that if, for example, `embark blockchain` is restarted 4 times, there will be 4 errors emitted from the `eth-block-tracker`. Because of this, errors emitted from `eth-block-tracker` have been reduced to trace to avoid clogging the logs.
First case - run `embark run` which starts a blockchain node, then manually kill the `geth` process. Would throw `{ [Error: connect ECONNREFUSED 127.0.0.1:8543] message: 'connect ECONNREFUSED 127.0.0.1:8543', code: -32603 }` error and ruins the dashboard.
Second case, 1) run `embark blockchain` 2) run `embark run` 3) kill `embark blockchain` throws the error `{ [Error: connect ECONNREFUSED 127.0.0.1:8543] message: 'connect ECONNREFUSED 127.0.0.1:8543', code: -32603 }` and ruins the dashboard.
The first case was solved by having the child blockchain process that spawns geth listen for geth exit, then kill itself.
The second case required updating of `eth-block-tracker` to v4.0.1 inside of the `embark-web3-provider-engine`. v4.0.1 was a major version update and introduced breaking changes. Those changes were handled inside of `embark-web3-provider-engine`, covered in **blocker** PR https://github.com/jrainville/provider-engine/pull/1.
Added a “factory pattern” to DevFunds to allow for async calls to be made during object construction. This allows for `web3.eth.getAccounts` to be called and the `accounts` property assigned on instantiation.
This modification allows each unit test case to be run independently of the others.
Finished added unit tests for dev_funds. These are weak tests as the FakeIpcProvider does not provide enough support for keeping track of accounts and balances and instead we are injecting the provider responses in the tests, which kind of defeats the purposes of the tests. Instead, the FakeIpcProvider should be a little smarter and do more of what a real node would do.
Iuri Matias
Michael Bradley, Jr
Anthony Laibe
Andre Medeiros
Jonathan Rainville
emizzle
Richard Ramos