Better output printing.

[DRMacIver]

This introduces draw_named as the primitive for when we are actually printing, which provides much nicer output for values.

It then makes the hegel::test macro automatically rewrite draw() to draw_named where it is safe to do so.

The result is that if you have a test that looks something like:

use hegel::{generators as gs};

#[hegel::test]
fn my_test(tc: hegel::TestCase){
    let mut running_total = 0;

    let iters = tc.draw(gs::integers().min_value(0).max_value(10));

    for _ in 0..iters {
        let summand = tc.draw(gs::integers().min_value(0).max_value(100));
        running_total += summand;
        assert!(running_total <= 100);
    }
}

You will see output like:

let iters = 2;
let summand_1 = 1;
let summand_2 = 100;
thread 'my_test' (2) panicked at examples/new_format.rs:12:9:
assertion failed: running_total <= 100

If you were to write it like this without the intermediate variable:

use hegel::{generators as gs};

#[hegel::test]
fn my_test(tc: hegel::TestCase){
    let mut running_total = 0;

    let iters = tc.draw(gs::integers().min_value(0).max_value(10));

    for _ in 0..iters {
        running_total += tc.draw(gs::integers().min_value(0).max_value(100));
        assert!(running_total <= 100);
    }
}

You would see:

let iters = 2;
let unnamed_1 = 1;
let unnamed_2 = 100;
thread 'my_test' (2) panicked at examples/new_format.rs:11:9:
assertion failed: running_total <= 100

Users are able to name things directly if they want. It's a bit unergonomic but not awful. e.g.

use hegel::{generators as gs};

#[hegel::test]
fn my_test(tc: hegel::TestCase){
    let mut running_total = 0;

    let iters = tc.draw(gs::integers().min_value(0).max_value(10));

    for _ in 0..iters {
        running_total += tc.draw_named(gs::integers().min_value(0).max_value(100), "summand", true);
        assert!(running_total <= 100);
    }
}

is equivalent to the first example.

[Liam-DeVoe]

My initial reaction: I think this is too much magic right now and we will regret going down this path. Doing AST-rewriting into draw_named for better output is clever, but I'd prefer to start with rewriting into something less magic, for example printing the file and line number in a format amenable to jump-to-definition cmd+click in an editor.

[DRMacIver]

My initial reaction: I think this is too much magic right now and we will regret going down this path. Doing AST-rewriting into draw_named for better output is clever, but I'd prefer to start with rewriting into something less magic, for example printing the file and line number in a format amenable to jump-to-definition cmd+click in an editor.

I think this is a relatively modest amount of magic for a huge ergonomics improvement. It is a well-defined rewrite between two public facing APIs that fails gracefully when it cannot be applied, and in exchange it gives you stable names for all your drawn values, which is most of the upside of the up front method that we lost with the inline method.

I think it's also a precondition for good usability on #144, so I'm extremely keen on getting it in.

[DRMacIver]

To put it into context: I think this is less magic than we routinely perform in Hypothesis, and it gives us better ergonomics than the equivalent features in Hypothesis.

[Liam-DeVoe]

I'm going through this now, but one thought at the top of my mind - I don't want draw_named to be part of our public API surface yet. I'd prefer it to be __draw_named so it doesn't show up in autocomplete when someone types tc.draw.

[Liam-DeVoe]

LGMT! Excited to get this in.

I renamed draw_named to __draw_named, because I don't want to make this API public yet. Happy to debate that when you're back 🙂

Did a huge cleanup pass of the tests. They now assert stronger statements about the actual printed lines.

Fixed one bug - we previously had this test:

#[test]
fn test_macro_top_level_same_name_panics() {
    let code = r#"
use hegel::generators as gs;

#[hegel::test(test_cases = 1)]
fn test_dup(tc: hegel::TestCase) {
    let x = tc.draw(gs::booleans());
    let x = tc.draw(gs::booleans());
}
"#;
    TempRustProject::new()
        .test_file("test_dup.rs", code)
        .expect_failure(r#"draw_named.*"_x".*more than once"#)
        .cargo_test(&["--test", "test_dup"]);
}

which should clearly be passing instead with names x_1 and x_2.