Collecting Code Coverage in Go: Tools, Configuration, and Integration Testing

In the previous week we introduced basic concepts of code coverage; this article dives into how Go collects coverage data using the official go test tool.

Key go test flags include:

covermode : the counting mode (set, count, atomic)

coverprofile : file to write coverage data

coverpkg : packages to instrument (e.g., ./... for all packages)

Example command to run tests and generate coverage:

go test -covermode=count \
    -coverprofile=coverage.out \
    -coverpkg=./... ./...

The output shows test execution and coverage information.

For integration testing, compile the test binary without running it using -c , then start the binary with a custom flag to trigger the main service only during system tests:

package main
import ("testing")
var systemTestCoverage *bool
func init() { systemTestCoverage = flag.Bool("systemTestCoverage", false, "Set to true when running system tests") }
func TestMain(t *testing.T) { if *systemTestCoverage { main() } }

Compile with:

go test -c -systemTestCoverage \
    -covermode=count \
    -coverprofile=coverage.out \
    -coverpkg=./... ./...

Go's coverage collection works by instrumenting the source before compilation, inserting a GoCover variable that records execution counts for each statement. After instrumentation, the code looks like:

func Size(a int) string { GoCover.Count[0] = 1; switch { case a < 0: GoCover.Count[2] = 1; return "negative"; ... } }

The resulting coverage report (in set mode) lists file names, line ranges, and execution counts, e.g.:

mode: set
code.byted.org/tesla/notify/func.go:3.25,4.9 1 1
...

Instrumentation adds about 3% extra code and may affect performance, so it is generally not recommended to run coverage‑instrumented binaries in production.

Real‑world solutions:

Our company’s solution : Developed by a dedicated engineer, it can collect coverage without killing the tested process, improving efficiency for repeated test runs.

Qiniu’s solution : Uses go tool cover to generate raw coverage data by traversing source files and reporting it, as described in a public article.