Crate lang_tester

This crate provides a simple language testing framework designed to help when you are testing things like compilers and virtual machines. It allows users to express simple tests for process success/failure and for stderr/stdout, including embedding those tests directly in the source file. It is loosely based on the compiletest_rs crate, but is much simpler (and hence sometimes less powerful), and designed to be used for testing non-Rust languages too.

For example, a Rust language tester, loosely in the spirit of compiletest_rs, looks as follows:

use std::{env, fs::read_to_string, path::PathBuf, process::Command};

use lang_tester::LangTester;
use tempfile::TempDir;

fn main() {
    // We use rustc to compile files into a binary: we store those binary files
    // into `tempdir`. This may not be necessary for other languages.
    let tempdir = TempDir::new().unwrap();
    LangTester::new()
        .test_dir("examples/rust_lang_tester/lang_tests")
        // Only use files named `*.rs` as test files.
        .test_path_filter(|p| p.extension().and_then(|x| x.to_str()) == Some("rs"))
        // Treat lines beginning with "#" inside a test as comments.
        .comment_prefix("#")
        // Extract the first sequence of commented line(s) as the tests.
        .test_extract(|p| {
            read_to_string(p)
                .unwrap()
                .lines()
                // Skip non-commented lines at the start of the file.
                .skip_while(|l| !l.starts_with("//"))
                // Extract consecutive commented lines.
                .take_while(|l| l.starts_with("//"))
                .map(|l| &l[COMMENT_PREFIX.len()..])
                .collect::<Vec<_>>()
                .join("\n")
        })
        // We have two test commands:
        //   * `Compiler`: runs rustc.
        //   * `Run-time`: if rustc does not error, and the `Compiler` tests
        //     succeed, then the output binary is run.
        .test_cmds(move |p| {
            // Test command 1: Compile `x.rs` into `tempdir/x`.
            let mut exe = PathBuf::new();
            exe.push(&tempdir);
            exe.push(p.file_stem().unwrap());
            let mut compiler = Command::new("rustc");
            compiler.args(&["-o", exe.to_str().unwrap(), p.to_str().unwrap()]);
            // Test command 2: run `tempdir/x`.
            let runtime = Command::new(exe);
            vec![("Compiler", compiler), ("Run-time", runtime)]
        })
        .run();
}

This defines a lang tester that uses all *.rs files in a given directory as test files, running two test commands against them: Compiler (i.e. rustc); and Run-time (the compiled binary).

Users can then write test files such as the following:

// Compiler:
//   stderr:
//     warning: unused variable: `x`
//       ...unused_var.rs:12:9
//       ...
//
// Run-time:
//   stdout: Hello world
fn main() {
    let x = 0;
    println!("Hello world");
}

lang_tester is entirely ignorant of the language being tested, leaving it entirely to the user to determine what the test data in/for a file is. In this case, since we are embedding the test data as a Rust comment at the start of the file, the test_extract function we specified returns the following string:

Compiler:
  stderr:
    warning: unused variable: `x`
      ...unused_var.rs:12:9
      ...

Run-time:
  stdout: Hello world

Test data is specified with a two-level indentation syntax: the outer most level of indentation defines a test command (multiple command names can be specified, as in the above); the inner most level of indentation defines alterations to the general command or sub-tests. Multi-line values are stripped of their common indentation, such that:

x:
  a
    b
  c

defines a test command x with a value a\n b\nc. Trailing whitespace is preserved.

String matching is performed by the fm crate, which provides support for ... operators and so on. Unless lang_tester is explicitly instructed otherwise, it uses fm’s defaults. In particular, even though lang_tester preserves (some) leading and (all) trailing whitespace, fm ignores leading and trailing whitespace by default (though this can be changed).

Each test command must define at least one sub-test:

• status: <success|error|signal|<int>>, where success and error map to platform specific notions of a command completing successfully or unsuccessfully respectively. signal checks for termination due to a signal on Unix platforms; on non-Unix platforms, the test will be ignored. <int> is a signed integer checking for a specific exit code on platforms that support it. If not specified, defaults to success.
• stderr: [<string>], stdout: [<string>] match <string> against a command’s stderr or stdout. The special string ... can be used as a simple wildcard: if a line consists solely of ..., it means “match zero or more lines”; if a line begins with ..., it means “match the remainder of the line only”; if a line ends with ..., it means “match the start of the line only”. A line may start and end with .... Note that stderr/stdout matches ignore leading/trailing whitespace and newlines, but are case sensitive. If not specified, defaults to ... (i.e. match anything). Note that the empty string matches only the empty string so e.g. stderr: on its own means that a command’s stderr muct not contain any output.

Test commands can alter the general command by specifying zero or more of the following:

• env-var: <key>=<string> will set (or override if it is already present) the environment variable <key> to the value <string>. env-var can be specified multiple times, each setting an additional (or overriding an existing) environment variable.
• exec-arg: <string> specifies a string which will be passed as an additional command-line argument to the command (in addition to those specified by the test_cmds function). Multiple exec-args can be specified, each adding an additional command-line argument.
• stdin: <string> specifies text to be passed to the command’s stdin. If the command exits without consuming all of <string>, an error will be raised. Note, though, that operating system file buffers can mean that the command appears to have consumed all of <string> without it actually having done so.

Test commands can specify that a test should be rerun if one of the following (optional) is specified and it matches the test’s output:

• rerun-if-status follows the same format as the status.
• rerun-if-stderr and rerun-if-stdout follow the same format as stderr and stdout.

These can be useful if tests are subject to intermittent errors (e.g. network failure) that should not be considered as a failure of the test itself. Test commands are rerun at most n times, which by default is specified as 3. If no rerun-if- is specified, then the first time a test fails, it will be reported to the user.

The above file thus contains 4 meaningful tests, two specified by the user and two implied by defaults: the Compiler should succeed (e.g. return a 0 exit code when run on Unix), and its stderr output should warn about an unused variable on line 12; and the resulting binary should succeed produce Hello world on stdout.

A file’s tests can be ignored entirely with:

• ignore-if: <cmd> defines a shell command that will be run to determine whether to ignore this test or not. If <cmd> returns 0 the test will be ignored, otherwise it will be run. <cmd> will have its directory set to CARGO_MANIFEST_DIR.

lang_tester’s output is deliberately similar to Rust’s normal testing output. Running the example rust_lang_tester in this crate produces the following output:

$ cargo run --example=rust_lang_tester
   Compiling lang_tester v0.1.0 (/home/ltratt/scratch/softdev/lang_tester)
    Finished dev [unoptimized + debuginfo] target(s) in 3.49s
     Running `target/debug/examples/rust_lang_tester`

running 4 tests
test lang_tests::no_main ... ok
test lang_tests::unknown_var ... ok
test lang_tests::unused_var ... ok
test lang_tests::exit_code ... ok

test result: ok. 4 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

If you want to run a subset of tests, you can specify simple filters which use substring match to run a subset of tests:

$ cargo run --example=rust_lang_tester var
   Compiling lang_tester v0.1.0 (/home/ltratt/scratch/softdev/lang_tester)
    Finished dev [unoptimized + debuginfo] target(s) in 3.37s
     Running `target/debug/examples/rust_lang_tester var`

running 2 tests
test lang_tests::unknown_var ... ok
test lang_tests::unused_var ... ok

test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured; 2 filtered out

Integration with Cargo.

Tests created with lang_tester can be used as part of an existing test suite and can be run with the cargo test command. For example, if the Rust source file that runs your lang tests is lang_tests/run.rs then add the following to your Cargo.toml:

[[test]]
name = "lang_tests"
path = "lang_tests/run_tests.rs"
harness = false

Structs

• LangTester