What can we do to improve the reporting?

In the previous post, we ended up with the following report when running tests.

stderr:
thread 'main' panicked at '
runit-assertions/src/lib.rs:66
    ✓ should be not equal to `0`, was `42`
    ⨯ should be not equal to `42`, was `42`
'

This is already nice but can we improve on that? First of all, it be real nice to color it. How about we use red/green colors to quickly identify which tests failed?

Colors in the terminal

To display colors in the terminal, I searched for a crate with the keyword color in crates.io (see link). ^[1]

A few of them showed up and all are done by known people ^[2] in the community:

ansi_term by Josh Triplett (owner on cargo) and Benjamin Sago (author of exa)
termcolor by Andrew Gallant also know as BurntSushi (prolific author of at least regex, csv, ripgrep to name only a few)
term by Alex Crichton (owner on cargo), Huon Wilson and Steven Allen
console by Armin Ronacher (which I’ve known through cargo-insta) and Pavan Kumar Sunkara

Ok, that is a lot of dropped names but that is also a way for me to thank them. I’ve known by name most of them, interacted with some of them. But for sure, all of them brought a lot to the Rust ecosystem: thank you.

Now, which one to choose? To be honest, I’ve picked the most downloaded one: ansi_term. It’s not the most recently updated (ansi_term: 2 years, termcolor: 1 year, term: 1 year, console: 4 months) but it is the most downloaded one by about 15 millions. In any case, it should not be to difficult to change for another one if that was necessary so I’m not spending too much time thinking about it.

Additional note, most of these libraries basically brings colors but also bold, underlined and a few other stylish stuff.

Let’s make it green and red

First of all, I’m not gonna get through code here because it’s not really interesting. The only technical part worth mentioning is that I’ve made ansi_term an optional dependency, although part of the default features. This means that I want colored output to be part of the default provided behavior of runit… but people should be able to suppress that if they believe it’s just fancy and add too much to the building time (even if it’s probably going to be test building).

First iteration

Very simple start. How about this?

Some colored checkmarks or crosses with a flashy green and red color. On top of that, some bold fonts for the location. Pretty good already, right? We can do more.

Second iteration

How about we highlight more the failed tests and make the passed tests a bit more discrete.

Colored background lines with red/green for failed/pass tests

Now, the entire background’s line is colored. The failed tests are more flashy with bold, whereas the passed tests use a more pastel color with normal font.

Can we do more? Let’s try!

Third iteration

The inspiration for this third iteration is purely inherited from pretty_assertions.

First of all, compared values are aligned to ease the comparison. I also used a diffing algorithm thanks to dissimilar to precisely show the difference between the strings. I you’re interested in the details of how complex that nice feature was to implement, it’s only about 27 lines of code. dissimilar is really pulling all the hard work for us.

Since the output is now on 2 lines, I’ve underlined some lines to create a separation between each assertion.

I like this last rendering… although, there is a bunch of problems with it.

Problems to address later

When I worked on these features, I worked head-on on the coloring improvement. Now that it’s done, I can take a step back and think a bit. So far, I’ve been exploring the rendering of only is_equal_to (and its friend is_not_equal_to). But the plan always was to expand to more assertions.

What good is going to be a diff output for something like is_greater_than? Yeah, you’re right, it doesn’t make any sense.

How about if we start using regular expression and have a matches assertion?

Or asserting on a map with has_key or contains?

And I’m sure I could find thousands of these little examples where a diff would not make any sense. It feel like a diff is only useful when comparing for equality which is exactly what PartialEq is about.

Let’s pin these and we’ll try to address them on a later post. One thing that seems to emerge is that the reporting is tightly coupled with the assertion. In term of code architecture, it probably means that the current version, that you can find on woshilapin/runit, is not well-suited for what’s coming.

1. Pretty original ugh!

2. I can even say famous for some of them.

In the rabbit hole

Other articles

Assertion reporting

Little recap

Equality assertions

Equality for other types

Fluent assertions for numbers in Rust

Introduction

Other articles

Little recap

Equality for other types

Introduction

links

social