-
Notifications
You must be signed in to change notification settings - Fork 5.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add a #[test(should_revert)]
attribute to indicate a unit test should revert
#3260
Comments
OK, this should be good to go! ## Highlights - Adds support for multiple entry points to IR. Adds a tests for serializing entry functions to/from IR. - Enables ASM generation for libraries (to support test function entry points). - Track entry points through ASM generation so we can return entry point metadata as a part of the compilation result. This doesn't affect ASM generation, but allows tools using `sway-core` as a library to work with different entry points. - Updated E2E test harness with a new test category "UnitTestsPass". - Added E2E tests with multiple unit tests for each type of Sway program (library, script, predicate, contract). - Gets `forc test` working with pretty output! Here's the output from the new `lib_multi_test`: ![Screenshot from 2022-11-03 19-00-57](https://user-images.githubusercontent.com/4587373/199700362-ba32f90d-1f0f-4f76-bf89-b191de9589af.png) ## Test running implementation Currently, it seems like there's no publicly accessible approach to execute a script directly from a custom entry point. As a result, for each test, we patch the bytecode with a `JI` instruction that jumps from after the data section setup to the test's entry point. This is a bit hairy, but works for now! As a follow-up, we may want to consider adding support upstream in `fuel-vm` for executing scripts directly from a given entry point. Even if this was only exposed from the interpreter API, this would make the `forc-test` implementation quite a bit cleaner and avoid the need to patch the bytecode. Alternatively, when building projects we could return more metadata along with the compiled output (whether in memory or bytecode header) to indicate how to work with different entry points in a more reliable manner (rather than the magic const offset currently used in `forc-test`). # TODO - [x] Add `include_tests` flag to `BuildConfig`, allowing `forc` to trigger compilation of test functions. - [x] Include `#[test]` fns as entry points within dead code analysis. - [x] IR and ASM generation for test entry points. - [x] Add `forc build --tests`. - [x] Add `forc test`. - [x] Always include test fns in `TyProgram` (for DCA), but omit from IR if not building tests. - [x] Work out how to iterate over different entry points during `forc test`. - [x] Only generate tests for top-level "members" (not all dependencies). ### Follow-up: - #3260 - #3261 - #3262 - #3263 - #3264 - #3265 - #3266 - #3267 - #3268 Closes #1832.
cc @kayagokalp IIRC, using the IR metadata could be the way to go here w.r.t. passing "revertness" information through to the compiled asm result. #3266 is related, in that it's also about passing info through to the asm-gen stage. Perhaps ideally, we'd have some way of mapping the generated test entry points to their matching function declarations - then rather than having to pass loads of information through IR metadata, we would just need to pass through the TypeId or the DeclId or something like that, and then use that ID to look up that function. cc @otrho in case you have any thoughts/tips on this :) In implementing basic initial support for |
Nice, I was basically implementing this but collecting the whole declaration struct, collecting type/decl id instead sounds cool! cc @mitchmindtree |
Tests decorated with
#[test(should_revert)]
should "pass" in the case that they exited withProgramState::Revert(_)
.The text was updated successfully, but these errors were encountered: