We’ve posted a new proposal to deprecate mojo test. The proposal itself has a detailed background on the issues we’ve encountered with mojo test, as well as how to migrate your projects to the new Mojo TestSuite framework that the team is building.
In short, mojo test has had some problems throughout its history. Initially, it used the Mojo REPL to run tests, which led to slow execution and subtly broken behavior as code was not being run in the same way as compiled Mojo. The most recent form of mojo test is tightly coupled to the parser and located in publicly inaccessible C++ portions of the codebase. It has a number of bugs, and its emulation of a main function makes it incompatible with the test orchestration we want to do.
To provide a better path forward for testing in Mojo, the standard library team is building a new testing framework in Mojo. You may have noticed the addition of some reflection functionality in Mojo for discovering functions within a module. That has enabled a Mojo-based way to perform test discovery. The team is in the process of converting the standard library’s tests to use this new framework, and people have really liked the results so far.
We are therefore proposing to deprecate the current mojo test and migrate Mojo testing to this new test framework, along with user-defined test orchestration.
Steps to migrate off of mojo test
The first step in migration to this new framework is to add the following to your test_*.mojo files:
from testing import TestSuite
# ... The existing test functions starting with `test_`.
def main():
TestSuite.discover_tests[__functions_in_module()]().run()
If you don’t already have a custom build system or process for test orchestration (Bazel, scripts, etc.), you can use a script like the following to find and run your test files:
#!/bin/bash
set -e
# Get test directory from first argument
if [ -z "$1" ]; then
echo "Error: Test directory not provided"
echo "Usage: $0 <test_directory>"
exit 1
fi
test_dir="$1"
any_failed=false
echo "### ------------------------------------------------------------- ###"
while IFS= read -r test_file; do
if ! mojo run "$test_file"; then
any_failed=true
fi
echo "### ------------------------------------------------------------- ###"
done < <(find "$test_dir" -name "test_*.mojo" -type f | sort)
if [ "$any_failed" = true ]; then
exit 1
fi
If you add that script to your project, you can then set up a Pixi or uv command for running tests with the aid of that script.
Feedback
We feel that building a robust testing framework for Mojo in Mojo is the right path forward, but we welcome your comments about this plan. Are there common use cases that we’ve missed? Will this cause trouble that we’re not aware of? Suggestions for improvement?
Feel free to let us know your thoughts on this proposal in the comments here.