In https://github.com/embark-framework/embark/commit/87d92b6091 we've introduced a feature in our test runner
to report exact gas costs used when calling Smart Contract methods that
cause computation.
Embark keeps a list of deployed contracts in memory and for all transactions
happening during tests, it'll match them to the contracts that performed them
to produce the report logs.
Unfortunately, we've been [resetting that memory](https://github.com/embark-framework/embark/commit/87d92b6091#diff-92b4f79a0473160fe700440b1ced5204R140) of deployed contracts
after every test, making it practically impossible for Embark to find
matching contracts for any transactions, because contracts aren't necessarily
redeployed per spec, resulting in no additional transaction logs whatsoever.
This commit ensures that the memory is never erased and at the same time
ensures it's not leaking infinitely in case multiple contracts are
deployed multiple times in the same process over several specs.
When running unit tests inside a project that configures ENS subdomains using Smart Contract directives,
the tests will output an error because of that particular Smart Contract's `deployedAddress` being `undefined`.
This happens only when running tests, not when deploying the Smart Contracts including the custom ENS setup.
It turns out that Embark attempts to compile and deploy the Smart Contracts of the project *twice* - once
before tests are executed, and another time **after** tests are done executing.
Both compilations/deployments are triggered through our custom `config()` function within test context,
which ensures all Smart Contracts are deployed before tests are executed.
This explains a few things:
1. There's no such issue when running `embark run`, in fact the custom ENS subdomains work perfectly fine
2. That's also why the tests are passing fine as well as the first compilation/deployment doesn't have any issues.
The errors only appear *after* the tests have been executed.
Still, this raises a few more questions, mainly
- Why is the Smart Contract's `deployedAddress` property `undefined` when `config()` is executed a second time?
- Why is `config()` executed a second time in the first place?
Let's look into both of these.
Assuming that there's a valid reason that `config()` is called twice, it's remains unclear why that property
of a Smart Contract object is `undefined` in the second run. The reason for that is that Embark determines whether
or not a particular Smart Contract should be deployed. One of the routines ensures that only the Smart Contracts
configured for deployment are actually attempted to be deployed (as opposed to just deploying all Smart Contracts
found in the file system).
It turns out that the second `config()` call is done without any Smart Contract configuration, resulting basically in no deployment
for any Smart Contract of the application. The `address` and `deployedAddress` of a Smart Contract object are however only set
*after* deployment, resulting in them being `undefined`.
**All this makes sense.**
`config()` is simply not designed for being used with an empty `contracts: {}` config. This raises another question:
Why is `config()` called with an empty configuration? This leads us to the second point.
It does seem a bit weird that Embark tries to configure compile *and* deploy a DApp's Smart Contracts again right **after**
the tests have been executed. So why is that?
A quick `git blame` (no blame intended here) shows us that this routine has been introduced in https://github.com/embark-framework/embark/commit/12cbb7bdd.
Notice the second list point of the commit:
> Contracts that had been compiled in the JS tests were being deleted at the end of the JS test run, which caused errors
> with the files not being found. The fix was to reset the contracts config to no longer require the non-test contracts
> to be compiled/deployed.
The above is a little tricky to understand without a little bit of context: Embark runs two types of tests, JavaScript tests
and Solidity tests. It does that as part of a single process, keeping state in memory in-between those two test runs.
With that in mind, the commit mentioned above says the following:
1. Artifacts that Embark generates as part of its compilation process are erased after JS tests are done executing.
This makes a lot of sense as we don't want to leave any side effects undone when tests are finished.
2. The commit ensures that not only the artifacts are removed from disc, but also updates Embark's state in memory
for `contractFiles`. The reason for that is that otherwise, in the second test run for Solidity tests, Embark will
throw errors as it tries to look up the path for the artifacts in memory for all the Smart Contracts that had been
compiled before.
3. Last but not least, there's *another* state that needs a reset and that's the Smart Contract configuration. If Embark
doesn't reset the memory it'll assume in the second run that all of those Smart Contracts left in memory
"have no code associated to it", while in reality, they shouldn't be in memory in the first place.
So it seems that `config()` is called a second time with an empty Smart Contracts configuration just to ensure that memory
is reset and no error messages are shown.
As discussed earlier, this unfortunately also introduced a lot of side effects along the way as Embark tries to reregister
ENS subdomains from Smart Contracts that are marked as undeployed and therefore don't have an address to interpolate.
While there's probably different ways to go about it, the most straight forward fix is to simply not call `config()`
a second time when it's really not needed. If the goal is to just reset the memory, we can take advantage of
Embark's internal `config:contractsConfig:set` event, which is what this commit is doing.
This enables removing unnecessary `core/fs` dependencies which can be
replaced with `fs-extra`.
This commit also fixes a bug introduced in f868d1216d
where methods from `embark.fs` weren't available.
The reason for that is because we have several *Process instances
that are created through child process communication, specifically
process.send() APIs. Using those APIs, we can only send data structures
however, methods attached on any of those will get lost.
This is the case when sending embark.fs through process.send().
We still need fs in those places though, mostly because they are relying
on embarkPath and dappPath().
These places are now importing those functions from `embark-core`. Other
API such as writeFile or mkdirp() can be accessed through fs-extra
or fs modules.
If a package uses only `.js` then supply `--extensions ".js"`. If a package
uses only `.ts` then supply `--extensions ".ts"`. If a package uses both, then
supply `--extensions ".js,.ts"`.
The reason for this is that adding/removing TS/JS support ought to occasion
revising a package's `package.json` file and adjusting other scripts as well,
e.g. for linting. With these changes, it won't work to simply start adding
`.ts` files in a package's `src/` directory, which should prompt the developer
to review `package.json` and make the appropriate changes, and/or other
developers may realize changes need to be made during code review.
Michael Bradley, Jr
Jonathan Rainville
Pascal Precht
Iuri Matias