Inconsistent branch coverage reports for generic functions

[NicoHaefele]

👋 Hi,

first of all, thanks for the great work on cargo-llvm-cov.

I noticed that the branch coverage report is not consistent across different output formats.
I am aware that branch coverage is still unstable and relies on nightly rustc,
but I couldn't find an issue for this specific problem.
Hence, I am opening this issue to discuss the observed behaviour
and to provide a minimal example to reproduce the issue.

Given the following Rust code:

pub fn maybe_increment_by<T>(mut value: T, increment: Option<T>) -> T
where
    T: std::ops::AddAssign + Copy,
{
    if let Some(increment) = increment {
        value += increment;
    }
    value
}

#[test]
#[cfg(test)]
fn test_maybe_increment_by() {
    assert_eq!(maybe_increment_by::<u32>(1, Some(2)), 3);
}

The expected behaviour is that the branch coverage is 50% or 1/2,
because the test currently does not account for the branch where increment is None.
All output formats seem to show the correct branch coverage.

However, if I change the test case to cover the None case like this:

#[test]
#[cfg(test)]
fn test_maybe_increment_by() {
    assert_eq!(maybe_increment_by::<u32>(1, Some(2)), 3);
    assert_eq!(maybe_increment_by::<u64>(1, Some(2)), 3);
    assert_eq!(maybe_increment_by::<u64>(1, None), 1);
}

You could argue that the expected branch coverage should be 100% or 2/2.
Some formats, like --html or the standard report format show the expected branch coverage:

cargo +nightly llvm-cov --branch

Filename         Regions    Missed Regions     Cover   Functions  Missed Functions  Executed       Lines      Missed Lines     Cover    Branches   Missed Branches     Cover
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------
~/src/lib.rs          10                 0   100.00%           2                 0   100.00%          14                 0   100.00%           2                 0   100.00%
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------
TOTAL                 10                 0   100.00%           2                 0   100.00%          14                 0   100.00%           2                 0   100.00%

Other formats, like --cobertura or --lcov show a different branch coverage of 75% or 3/4.
The 3/4 branches are most likely reported due to Rust's monomorphization and the different
types used for the generic function maybe_increment_by.

Now I am wondering if this is a bug or intended behaviour.
If it is the latter, sorry for the noise and thanks in advance for the clarification.