Monthly Archives: July 2019

UTest Harness for Rust (Setup/Test/Release)

Edit Titel should have been “setup/test/teardown”

The crate test-generator-utest implements a 3-phase utest-harness for Rust. These 3 phases are:

Setup: the setup function must initialize the context; in case the setup-function panics/aborts, the utest is aborted
fn() -> Context

Test: if the setup did return with valud context item, the context-reference is borowed to invoke the test-function
fn( & Context )

Teardown: no matter if the test-function above did panic/abort, the teardown function is invoked to release the context
fn( Context )

No matter of any panic or failing assertion in the second phase (feature testing), the teardown function is invoked. The test will either succeed, or otherwise unwinding a failure in the setup-phase, the test-phase or the teardown-phase.

This crate has been inspired by the post of Eric Opines.

## Usage

Please see the executable example mytests

#[cfg(test)]
extern crate test_generator_utest;

// demonstrating usage of utest-harness
mod testsuite {
    use std::fs::File;
    use std::io::prelude::*;

    use test_generator_utest::utest;

    // Defining a context structure, storing the resources
    struct Context<'t> { file: File, name: &'t str }

    // Setup - Initializing the resources
    fn setup<'t>(filename: &str) -> Context {
        // unwrap may panic
        Context { file: File::create(filename).unwrap(), name: filename }
    }

    // Teardown - Releasing the resources
    fn teardown(context: Context) {
        let Context { file, name } = context;

        // drop file resources
        std::mem::drop(file);

        // unwrap may panic
        std::fs::remove_file(name).unwrap();
    }

    // Test - verify feature
    fn test_write_hello_world(ctx: &Context) {
        // may panic
        let mut file = ctx.file.try_clone().unwrap();

        // may panic
        file.write_all(b"Hello, world!\n").unwrap();

        // demonstrating assertion
        assert_eq!(1,1);

        std::mem::drop(file);
    }

    // Test - verify feature
    fn test_write_hello_europe(ctx: &Context) {
        // may panic
        let mut file = ctx.file.try_clone().unwrap();

        // may panic
        file.write_all(b"Hello, Europe!\n").unwrap();

        // demonstrating assertion
        assert_eq!(1,1);

        std::mem::drop(file);
    }

    // Defining Utest formed by setup, test and teardown
    utest!(hello_world,
        || setup("/tmp/hello_world.txt"),
        |ctx_ref| test_write_hello_world(ctx_ref),
        |ctx|teardown(ctx));


    // Defining Utest formed by setup, test and teardown
    utest!(hello_europe,
        || setup("/tmp/hello_europe.txt"),
        test_write_hello_europe,
        teardown);
}

Executing the example code cargo test -p test-generator-example testsuite the testsuote above will print the following output.

running 2 tests
test testsuite::hello_europe ... ok
test testsuite::hello_world ... ok

test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured; 3 filtered out

Please let me know about possibel improvements via github-issues at test-generator-utest

Test-generator v0.3

A new version of test-generator (0.3) has been published. This crate introduces #[test_resources(PATTERN] and #[bench_resources(PATTERN)] procedural macro attributes that generates multiple parametrized tests using one body with different resource input parameters. A test is generated for each resource matching the specific resource location pattern.

Go To Statement Considered Harmful (discussed)

Talking with other developers about control-flow patterns, after a while someone will cite Edgar Dijkstra’s paper

“Go To Statement Considered Harmful” Communications of the ACM 11, 3 (March 1968), pages 147-148

Usually those developers argue, that go to statements should be abolished completely.

Just, it is important to note, that this paper is about loop-control-structures, comparing the expressiveness of recursive function calls, while/repeat-statements and goto-statements with each other (preferred pattern mentioned first).

And Dijkstra does not oppose the usage of go to statements in general, but mentioning a use case for go to statements:

“The remark about the undesirability of the go to statement is far from new. I remember having read the explicit recommendation to restrict the use of the go to statement to alarm exits, but I have not been able to trace it; presumably, it has been made by C.A.R. Hoare.”

My interpretation of this statement is, that go to statements can be tolerated for so called “alarm exits” as being used widely in C-code of OS-Kernels and embedded systems.

An “alarm exit” may look like this, realizing a single exit point for the success case and a single one for the error case, instead of spreading multiple exit-points all over the function.

/**  foo reading a file and returns with SUCCESS, otherwise with ERROR
*/
int foo(const char* filename) {
int fd = INVALID_FD;
if (filename==INVALID_STRING) {
goto alarm_exit;
}
fd = open_file(filename);
if (fd == INVALID_FD) {
goto alarm_exit;
}
/* do something, release resource and exit with success */
close_file(fd);
return SUCCESS_CASE;

alarm_exit:
/* release resources if allocated */
if (fd!=INVALID_FD) close_file(fd);
return ERROR_CASE;
}

So, next time having a discussion with fundamental developers, abolishing all occurances of go to statements (even for the “alarm exits” use case), I will be happy to cite Dijkstra and Hoare 😉

The Computer Language Benachmarks Game: Rust ranks #1 for n-body

The Computer Language Benchmarks Game is a free software project for comparing how a given subset of simple algorithms can be implemented in various popular programming languages.

I converted the fastest (dating early 2019) n-body C-implementation (#4) to Rust (#7) in a one-to-one fashion, gaining a performance encreasement by factor 1.6 to my own surprise.

The moment writing, the benachmarks are measured on quad-core 2.4Ghz Intel® Q6600® ; dating back to the year 2006. This ancient hardware supports the SSSE3 64bit-SIMD instructions; both implementations are making use them intensively.

The conversion of the implementation from C to Rust was fun and educational; listing just a few:

  • The pattern of “static global memory” of C had to be transformed to similar code just based on he Rust ownership model.
  • Dealing with memory alignment-directives in Rust
  • Using the early, stable SIMD-API of Rust, namely std::arch::x86_64::*

I must admit that newer test-hardware with support for more advanced AVX-512 SIMD-instructions, would permit to run an even faster Rust-implementation of the n-body task as part of these benchmarks.