Propagate unreachability in coverage

[elinorbgr]

This is an attempt at propagating upward unreachability in processed files. The rationale being that code that can be statically determined to be unreachable cannot be tested, and should not be marked as missed in coverage.

The algorithm for propagating coverage I implemented is the following:

• unreachable!() macro invocation is unreachable
• a list of statements is unreachable iff at least one of the statements is unreachable
• a loop is unreachable iff its body is unreachable
• an if block is unreachable iff both its branches are unreachable
  • in particular an if with no else branch is reachable
• a match block is unreachable iff all its branches are unreachable
  • in particular, a match with no branch is unreachable, as it must be matching on an uninhabited type like an empty enum
• a function or method with an unreachable body is unreachable itself

This is a conservative implementation with minimal assumptions, I probably missed a few heuristics that could propagate more unreachability.

[xd009642]

Cool I'll have a look over this now. One thought/concern is in instances where this leads to big blocks of code marked as not coverable whether that would be intuitive to users or whether they'd assume this is an issue in tarpaulin

[xd009642]

All looks good, I'd just like one small test to make sure a list of statements is properly ignored 👍

[xd009642]

Can there be a test for this. Just something simple like

fn test_unreachable() {
    let x: u32 = foo();
    if x > 5 { 
        bar();
    }
    unreachable!();
}

[elinorbgr]

Added. It was a good finding: adding this test allowed me to find a bug I did not anticipate, statement macro calls like unreachable!(); are considered as items, not semicolon-ended-expressions...

[xd009642]

Ah I've been bitten by that one before, always forget it's a thing.

This all looks good to me, so merging in 👍