Sections, or subcases are a cool feature of unit testing frameworks, such as (awesome) C++ libraries Catch2 and doctest. Subcases provide an easy way to share code between tests, like fixtures do, but without needing to move setup and teardown code outside of your tests' meat, without hassles of object orientation.
How do they work? Subcases allow you to fork function execution to into different paths which will have common code in the places you want them to.
Let's look at an example.
use subcase::with_subcases;
with_subcases! {
#[test]
fn my_test_case() {
let mut v = vec![1,2,3];
subcase! { ~"single push"
v.push(9);
assert_eq!(v[3], 9);
}
subcase! { ~"clear then push"
v.clear();
assert!(v.is_empty());
for _i in 0..4 { v.push(1); }
}
assert_eq!(v.len(), 4);
assert!(v.capacity() >= 4);
}
}
my_test_case
's body will be executed twice, first time
with first subcase!{...}
block, ignoring the second,
and vice versa.
That's not all! Subcases can be nested!
use subcase::with_subcases;
with_subcases! {
#[test]
fn my_tremendous_test_case() {
let mut v = vec![1,2,3];
subcase! { ~"single push"
v.push(9);
}
subcase! { ~"clear, push, pop"
v.clear();
v.push(100);
subcase! { ~"push in for loop"
for _i in 0..5 { v.push(1); }
}
subcase! { ~"extend from slice"
v.extend_from_slice(&[4,5,6,7,8]);
}
assert_eq!(v.len(), 6);
v.pop();
v.pop();
}
assert_eq!(v.len(), 4);
}
}
Test function body is executed 3 times: once for each of leaf subcases (i.e. not containing more nested subcases), while the big parent subcase is entered twice.
You can write only one subcase or no subcases at all, function will run as usual.
Indeed, there are already a few crates that implement the concept of subcases:
What distinguishes subcase crate from each of them, is that
subcase only uses lightweight declarative (i.e. macro_rules!
)
macros and has zero dependencies. Also, with_subcases
macro stuffs
all execution paths inside one function, instead of generating
many. These making it very easy on Rust compiler, in comparison
to the mentioned crates.
In subcase
's approach, subcases discovery and switching between them
happens serially at runtime.
One consequence of this is that different branches of a test case can't run in parallel. This may or may not slow your tests down. If you have a lot of fine-grained test cases, you should be fine.
Another consequence is that you generally cannot resume a test case
when one of the execution paths failed. If it failed with a panic,
subcase
will report what chain of subcases caused that.
You can read the changelog here. It follows Common Changelog style guide and is written with the help of hallmark tool.
Licensed under MIT License.