testing: allow creation of subtests

[cflewis]

Problem: When writing tests that take >3 minutes each, the tests should be parallelized. If your tests are also amenable to being written in a table-driven format, you have to resort to goroutines to make things parallel (rather than t.Parallel()). If you do that, then your logs get interleaved, and go test run .* can't help you out to isolate particular failures, and you lose the benefits of t.Parallel(), like setting number of parallel lines at CLI, ensuring non-parallel ones don't run with it...

Solution: A proposed solution discussed involves the ability for tests to create their own subtests.

[ianlancetaylor]

Marcel said he is working on something along these lines.

[neild]

It's also easy enough to convert a table-driven test into a set of individual tests, along the lines of:

type testFoo struct {...}
func (test testFoo) Test(t *testing.T) { ... }
func TestFooA(t *testing.T) {
testFoo{...}.Test(t)
}

[adg]

Closing in favor of #12166