#[test_script] for unit testing

[adlerjohn]

Add a #[test_script] that defines a script with no input parameters and an optional output value. Only valid for contracts. The compiler must:

1. compile the contract entirely, then
2. compile each script test independently, allowing implicit use of any ABI declarations the contract implements, then
3. output each script test's compiled output to a different .bin file.

This will be sufficient for downstream tools such as forc test to pick them up and automatically run them, as a first step to in-language testing.

---

Following added by @mitchmindtree.

TODO

• Forward attributes from ParseProgram to TyProgram #2979.
• Add #[test] to list of known attributes in sway-core #2980.
  • Support the #[test] attribute on Sway functions.
  • Add #[test] to set of known attributes.
  • Update original compile function (i.e. forc build) to ignore all #[test] functions. Likely best done during convert_parse_tree.
• Update sway-core to provide an alternative to compile called compile_tests that lazily generates bytecode for each test function, treating each as an entry point.
  1. Type check full program (including #[test] functions).
  2. Add #[test] functions as entry points for dead code analysis.
  3. Generate bytecode for each #[test] function entrypoint.
• Add a forc test command that executes the bytecode.
• Add a forc check --tests flag for use by sway-lsp to type-check all tests (without codegen or running them).

Follow-up

• Refactor E2E tests to take advantage of new #[test] feature where possible.
• Consider moving some std lib unit tests back to std library and use #[test], rather than in E2E tests?
• Introduce [dev-dependencies] table to Forc.toml.

[mitchmindtree]

@adlerjohn are you happy to go with #[test] for this over #[test_script]? I don't recall if there was a particular motivation for #[test_script] instead 🤔

[adlerjohn]

The motivation is that there needs to be a way to write tests that involve running a script transaction and calling one or more ABI methods, and a way to run a unit test that just happens without having to go through a transaction.

[mitchmindtree]

Ahhh yes, perhaps we could go with something like this?

• #[test] - base test declaration, "just happens".
• #[test(script)] - indicates a script transaction is required in order to call one or more ABI methods.

This way we only need to check for test throughout the compiler when doing things like checking for extra entry points, testing for known attribute names, etc. Then we can just do something like .contains("script") on the inner set to determine if a transaction is required.

[adlerjohn]

Interesting alternative. I'm impartial either way, so long as there's the two options. Might be a good time to get some bikeshedding comments in!