We're already actually building with the llvm library every time, so this isn't needed. Also, this is made more explicit here: #2034

So we can just remove this option altogether (and the conflict_handler option above).

Thanks! Removed in 18f4e77.

We should name this option --no-use-booster, not --no-booster. The convention is to prefix the exact same argument name with no-.

Renamed to --no-use-booster in 45680be; added this flag to the test-prove-foundry CI job, as suggested below.

This --no-use-booster flag logic should be added somewhere, you could either:

• rename all uses of use_booster in pytest tests and invert their logic
• handle the logic in the Makefile by writing --use-booster if the flag NO_USE_BOOSTER is false
  As you want to make the booster the default, 1. might be a better choice and a less confusing one

Actually, what about writing

diff --git a/kevm-pyk/src/tests/conftest.py b/kevm-pyk/src/tests/conftest.py
index 7a2b381d3..58cd4dcd6 100644
--- a/kevm-pyk/src/tests/conftest.py
+++ b/kevm-pyk/src/tests/conftest.py
@@ -30,4 +30,4 @@ def update_expected_output(request: FixtureRequest) -> bool:

 @pytest.fixture(scope='session')
 def use_booster(request: FixtureRequest) -> bool:
-    return request.config.getoption('--use-booster')
+    return not request.config.getoption('--no-use-booster')

Thanks @iFrostizz for the suggestion, looks very good!

@iFrostizz @ehildenb I realized that the --use-booster option is now also used in the CI Haskell proof tests (test-prove-haskell-booster), where the booster is not used by default. Changing the interface for use_booster in pytest tests will affect Haskell proofs tests too, i.e., we'll have to add --no-use-booster to no-booster test-prove-haskell in this line.

As far as I understand, the CLI's (--no)-use-booster behavior doesn't impact the usage of booster in tests, as the latter relies on the pytest --use-booster argument defined in conftest.py:

def pytest_addoption(parser: Parser) -> None:
...    
    parser.addoption(
        '--use-booster',
        action='store_true',
        default=False,
        help='Use the kore-rpc-booster binary instead of kore-rpc',
    )

@pytest.fixture
def update_expected_output(request: FixtureRequest) -> bool:
    return request.config.getoption('--update-expected-output')

It's further supported by CI job timings: the "legacy" Foundry proofs job still takes more than an hour to complete, while the booster job took less than 20 minutes.

I agree that it'd be better to unify the behavior between CLI and pytest arguments, but, considering different default booster behavior between kevm prove and kevm foundry-prove (and, therefore, test-prove-haskell and test-prove-foundry), I'm thinking about leaving the CI commands as they are. @iFrostizz @ehildenb could you please let me know what you think? Thanks!

I agree that it's okay to leave the CI commands as they are @palinatolmach , because they are not user facing, and because we may drop non booster CI jobs in the future. Also, the --use-booster option is more straightforward, having --no-use-booster features a double negation unless you edit further the tests, it looks like unnecessary effort!