Test coverage results don't seem to be right

[heisen-li]

I tested the randproject and found that it had zero branch coverage.

Is this normal?

[marco-c]

Are you using source-based coverage or gcov-based coverage?

[heisen-li]

This is based on the coverage of the source.Will the results of the two be different?

[heisen-li]

Thanks, I tested it. The results of the two are still quite different.

[heisen-li]

The reason I reopened is that I want to know what is the difference between these two test methods, and why is there such a big gap in the results of branch coverage?

[victor-aembit]

+1.

I tried to run it against my small project (with code based coverage) and I can see quite a lot of unexpected things:

• Zero branch coverage info (even though I used "--branch" flag) (This is my biggest concern)
• Some of macros for tests marked as hit. Some tests function are included (green) and some excluded (white)
• Function coverage percentage is strangely off. As an example, there are two functions (marked as green). However, it shows 66.67% (which implies that there is a third function, but nothing is marked as red)

[heisen-li]

• Zero branch coverage info (even though I used "--branch" flag) (This is my biggest concern)

This is because the source-based approach does not currently support branch coverage.
refer to:rust-lang/rust#79649

I agree to other situations. In addition, you can continue the discussion here.rust-lang/rust#91661

[marco-c]

The reason I reopened is that I want to know what is the difference between these two test methods, and why is there such a big gap in the results of branch coverage?

gcov-based coverage is using DebugInfo.
Source-based coverage instead is using the AST directly.

There is some information on https://clang.llvm.org/docs/SourceBasedCodeCoverage.html. It's about Clang, but should apply similarly to Rust.