Put integration tests behind feature flag

[lexnv]

The substrate binary is used internally for integration testing purposes.

This PR ensures that building/running tests do not rely on the existence of the
SUBSTRATE_NODE_PATH in the environment, which is guarded under the
integration-tests crate.

The subxt tests that rely on the substrate binary are moved to a dedicated crate.
At the same time, any dependency on the test-runtime is removed to avoid any
cyclic dependencies. The only crate that requires the substrate binary to exist in the
environment is the integration-tests crate.

Testing Done

cd subxt
SUBSTRATE_NODE_PATH="" cargo build --release
SUBSTRATE_NODE_PATH="" cargo test

Closes #351.

Next Steps

• Move integration tests to their own crate to avoid substrate binary dependency on other crates
• Construct manual metadata for subxt caching tests
• Add metadata.scale blob for examples and metadata crate testing
• Ensure #[cfg] is valid with feature flag

[ascjones]

An alternative to the test-runtime feature-flag approach would be having the
dependency declared as optional = true and included for the feature flag only.
Unfortunately, this option does not work for dev-dependencies. Having the
test-runtime as simply dependency would cause cyclic reference.

Since tests/integration is really it's own crate, this could be moved to the top level and made into an explicit crate which would allow using a normal dependency and make it optional?

[jsdw]

// Currently the caching does not take into account different pallets
// as the intended behavior is to use this method only once.
// Enforce this behavior into testing.
let hash_old = metadata.metadata_hash(&["Balances"]);
assert_eq!(hash_old, hash);

I don't really understand what this line does? (also doesn't look like there is a balances pallet in the fake metadata?)

(alas I couldn't comment on the line itself; sorry!)

[lexnv]

Have updated the comment. If we call metadata_hash with an argument of pallets: ["System"] it would produce a hash that is cached internally. Therefore, if we call the same method again, but with different arguments (ie pallets: ["Balances"]), it would fetch the previously cached value 😄

[jsdw]

Thanks, that is clearer! :)

[jsdw]

Looks good to me offhand, and I think the separate crate is a nice organisational improvement, and gives people the opportunity to avoid the integration tests if they don't have a suitable substrate binary to hand while still making it the default, which I think is right.

Just a couple of small comments and it's all good from me :)

[ascjones]

How about moving the contents of integration-tests/tests/ (which at the moment is an implicit sub-crate) into integration-tests/src (renaming main.rs to lib.rs?

[lexnv]

@ascjones Moving the/tests directory directly under /src makes the crate more ergonomic for me. Thanks for the feedback!

[ascjones]

LGTM

[jsdw]

Looking good to me, nice one!