Code Coverage

[meme]

Taking a run at #352.

There are a couple of outstanding questions here:

1. Do we want to invent our own coverage format and visualizer, or do we want to emit data that is compatible with pre-existing tools? If so, what tool(s) are we integrating with?
2. How can we dynamically allocate memory in compiler_rt? This is needed for storing encountered program counters while tracking code coverage.
3. How can we get a mapping from program counter to function names or line numbers? Can we guarantee that Zig tests will be compiled with symbols or DWARF symbols and parse our own binary at runtime during coverage output?

You can find an implementation of ASan's SanitizerCoverage runtime here: https://github.com/llvm/llvm-project/blob/master/compiler-rt/lib/sanitizer_common/sanitizer_coverage_libcdep_new.cpp. If Zig decides to act similarly to the SanitizerCoverage runtime here, we also need to support disk access in the compiler_rt.

---

Thoughts on 1)

The LLVM coverage report server (tools/sancov/coverage-report-server.py) generates highlighted HTML from SanitizerCoverage files, but I was imagining more what Go does, e.g.:

$ go test -cover fmt
ok  	fmt	0.060s	coverage: 91.4% of statements

Alternatively we could write out the coverage file:

$ go test . -coverprofile cover.out
ok      example        0.001s  coverage: 100.0% of statements

And then provide a tool for collecting more information:

$ go tool cover -func cover.out
example.go:4:              DoSomething     100.0%
total:                     (statements)    100.0%

The advantage of writing out a coverage file is other members of the Zig community could create external software that takes advantage of Zig coverage files and e.g. produce more graphical representations like stylized HTML output.