Why grcov count function more than once ?

[redtankd]

For these codes, grcov count 8 functions in total and 3 functions hit. test_student() is excluded by me. Actually, only new() is hit. get_id() and get_grade() are not tested. I thought that there be 3 functions in total and 1 function hit.

Why?

pub struct Student {
    id: i32,
    grade: u8,
}

impl Student {
    pub fn new(id: i32, grade: u8) -> Student {
        Student {
            id: id,
            grade: grade,
        }
    }

    pub fn get_id(&self) -> i32 {
        self.id
    }

    pub fn get_grade(&self) -> u8 {
        self.grade
    }
}

#[test]
fn test_student() {
    let stud = Student::new(10i32, 200u8);
    assert_eq!(10i32, stud.id);
}

[calixteman]

I agree with you. Could you give us a way to reproduce please ? Anyway in inspecting the llvm ir for this code, I see 5 functions, new, get_id, get_grade, test_student and a closure for test_student. So I don't know how you can get 8 functions.

[redtankd]

The attachment is my crate. code-coverage.zip

My OS is macOS 10.15.5

My env is

CARGO_INCREMENTAL=0
RUSTFLAGS="-Zprofile -Ccodegen-units=1 -Copt-level=0 -Clink-dead-code -Coverflow-checks=off -Zpanic_abort_tests"

-Cpanic=abort doesn't work for dylib, so I remove it.

The steps are

cargo +nightly test
grcov . -o ./target/debug/lcov.info -s . -t lcov --llvm --branch --ignore-not-existing --ignore "$HOME/.cargo/*" --excl-start "#\[test\]" --excl-br-start "#\[test\]"
genhtml -o ./target/debug/coverage/ --show-details --branch-coverage --no-prefix --highlight --ignore-errors source --legend ./target/debug/lcov.info

From the report I find 8 functions for src/stud/student.rs.

_ZN13code_coverage4stud7student7Student3new17h1e31aa08edd8ed2dE | 0
-- | --
_ZN13code_coverage4stud7student7Student6get_id17h032a0defb68c635dE | 0
_ZN13code_coverage4stud7student7Student6get_id17hcd930dd96a635515E | 0
_ZN13code_coverage4stud7student7Student9get_grade17hbe4d057828c8491cE | 0
_ZN13code_coverage4stud7student7Student9get_grade17hcf71b6946d15f11bE | 0
_ZN13code_coverage4stud7student12test_student17hf23155a333cae1bdE | 1
_ZN13code_coverage4stud7student12test_student28_$u7b$$u7b$closure$u7d$$u7d$17h2ac7417fb8ab4f53E | 1
_ZN13code_coverage4stud7student7Student3new17hbd7550f0bd160415E

[SWW13]

Demangling the functions clears this a bit up:

code_coverage::stud::student::Student::new | 0
code_coverage::stud::student::Student::get_id | 0
code_coverage::stud::student::Student::get_id | 0
code_coverage::stud::student::Student::get_grade | 0
code_coverage::stud::student::Student::get_grade | 0
code_coverage::stud::student::test_student | 1
code_coverage::stud::student::test_student::{{closure}} | 1
code_coverage::stud::student::Student::new | 1

The functions hit are test_student, a closure in test_student (which is most likely related to assert_eq!) and Student::new. My guess for the duplicate functions is a cargo build which generated another binary (without test code).

[MaulingMonkey]

I've noticed that e.g. the presence of a build.rs can cause duplicate functions like this when fed something like e.g. --binary-path target\debug... presumably because it's using a different profile to build said build script?

llvm-cov accepts a demangler ala: llvm-cov show -Xdemangler=rustfilt - it might be worth having grcov do so as well, and then merge hit counts from identical fns? Or perhaps such things should be handled upstream in llvm-cov itself? Or perhaps just improving display of "identical" fns so users can understand where "duplicate" functions are coming from?