Strange indentation in libtest output

[RalfJung]

libtest output recently started indenting some but not all test results:

$./x.py test library/alloc --stage 0 --test-args "--format pretty" --no-doc
running 373 tests
test raw_vec::tests::allocator_param ... ok
test raw_vec::tests::reserve_does_not_overallocate ... ok
test raw_vec::tests::zst ... ok
test alloc::tests::allocate_zeroed ... ignored
test raw_vec::tests::zst_reserve_panic - should panic ... ok
test raw_vec::tests::zst_reserve_exact_panic - should panic ... ok
test alloc::tests::alloc_owned_small                         ... ok   <!----- notice this one!
test collections::binary_heap::tests::test_iter_rev_cloned_collect ... ok
[...]
test collections::linked_list::tests::test_drop ... ok
test collections::linked_list::tests::test_drop_panic ... ok
test collections::vec_deque::tests::bench_push_front_100     ... ok  <!---- and these!
test collections::vec_deque::tests::bench_push_back_100      ... ok
test collections::vec_deque::tests::bench_pop_back_100       ... ok
test collections::vec_deque::tests::bench_pop_front_100      ... ok
test collections::vec_deque::tests::test_swap_front_back_remove ... ok
test collections::vec_deque::tests::test_insert ... ok
test collections::vec_deque::tests::test_get ... ok
test collections::vec_deque::tests::test_get_mut ... ok
test collections::vec_deque::tests::bench_retain_whole_10000 ... ok

Seems like benchmarks are indented but regular tests are not.

[RalfJung]

Cc @thomcc any idea which change might have caused this? I think this must be fairly recent but haven't tried bisecting yet.

[m-ou-se]

Is this still reproducable?

[RalfJung]

With #108659 this can no longer be tested using x.py, since bootstrap now does test output formatting itself.

So this now needs a little cargo project with a test and a benchmark, to check what libtest actually does.

[ehuss]

It is because benches are dynamic, and they always use padding (and tests are static):

https://github.com/ehuss/rust/blob/6b57a344cba18b18178cdcf609d68880f33cc6eb/library/test/src/types.rs#L90-L97

I'm not sure why padding is different for static vs dynamic.

Here is a repro:

#![feature(test)]
extern crate test;

#[test]
fn foo() {}

#[bench]
fn short_name(b: &mut test::Bencher) {
    b.iter(|| 1);
}

#[bench]
fn this_is_a_really_long_name(b: &mut test::Bencher) {
    b.iter(|| 1);
}

[cuviper]

Seems like that could use context if the bench is only being run as a test, then don't pad.

The padding is desirable for benchmark output so the numeric values line up, for easier visual comparison. You could argue either way about tests -- maybe ok/FAILED would stand out even more with padded alignment.