/rusty-fork

Run Rust tests in isolated subprocesses

Primary LanguageRustApache License 2.0Apache-2.0

rusty-fork

Build Status

Rusty-fork provides a way to "fork" unit tests into separate processes.

There are a number of reasons to want to run some tests in isolated processes:

  • When tests share a process, if any test causes the process to abort, segfault, overflow the stack, etc., the entire test runner process dies. If the test is in a subprocess, only the subprocess dies and the test runner simply fails the test.

  • Isolating a test to a subprocess makes it possible to add a timeout to the test and forcibly terminate it and produce a normal test failure.

  • Tests which need to interact with some inherently global property, such as the current working directory, can do so without interfering with other tests.

This crate itself provides two things:

  • The rusty_fork_test! macro, which is a simple way to wrap standard Rust tests to be run in subprocesses with optional timeouts.

  • The fork function which can be used as a building block to make other types of process isolation strategies.

Quick Start

If you just want to run normal Rust tests in isolated processes, getting started is pretty quick.

In Cargo.toml, add

[dev-dependencies]
rusty-fork = "0.3.0"

Then, you can simply wrap any test(s) to be isolated with the rusty_fork_test! macro.

use rusty_fork::rusty_fork_test;

rusty_fork_test! {
    #[test]
    fn my_test() {
        assert_eq!(2, 1 + 1);
    }

    // more tests...
}

For more advanced usage, have a look at the fork function.

How rusty-fork works

Unix-style process forking isn't really viable within the standard Rust test environment for a number of reasons.

  • While true process forking can be done on Windows, it's neither fast nor reliable.

  • The Rust test environment is multi-threaded, so attempting to do anything non-trivial after a process fork would result in undefined behaviour.

Rusty-fork instead works by spawning a fresh instance of the current process, after adjusting the command-line to ensure that only the desired test is entered. Some additional coordination establishes the parent/child branches and (not quite seamlessly) integrates the child's output with the test output capture system.

Coordination between the processes is performed via environment variables, since there is otherwise no way to pass parameters to a test.

Since it needs to spawn new copies of the test runner executable, rusty-fork does need to know about the meaning of every flag passed by the user. If any unknown flags are encountered, forking will fail. Please do not hesitate to file issues if rusty-fork fails to recognise any valid flags passed to the test runner.

It is possible to inform rusty-fork of new flags without patching by setting environment variables. For example, if a new --frob-widgets flag were added to the test runner, you could set RUSTY_FORK_FLAG_FROB_WIDGETS to one of the following:

  • pass — Pass the flag (just the flag) to the child process
  • pass-arg — Pass the flag and its following argument to the child process
  • drop — Don't pass the flag to the child process
  • drop-arg — Don't pass the flag to the child process, and ignore whatever argument follows.

In general, arguments that affect which tests are run should be dropped, and others should be passed.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.