cmd/compile: -G=3 mode correctly reports all "declared but not used" errors, but old compiler does not

[danscales]

For current master, if you enable -G=3 in the compiler by adding 'Flag.G = 3' at the end of ParseFlags in cmd/compile/internal/base/flag.go, then this command fails

go test -count=1 -race -run=TestOutput runtime/race

This is one of the commands that is run under the "##### Testing race detector" section of the all.bash tests.

The reason is that types2 reports an error "main.go:4:2: x declared but not used" (which does seem most correct) for:

package main
func main() {
        done := make(chan bool)
        x := 0
        go func() {
                x = 42
                done <- true
        }()
        x = 43
        <-done
}

whereas the current compiler/typechecker (-G=0) does not report any error. (Note that 'go vet' does return the error "vet: ./test.go:4:2: x declared but not used".)

So, we have to decide if we can make this change in the behavior in the typechecker or not. If not, then we need to "fix" types2 to not report this error.

If it's OK to have this change in behavior, then we can make the test work for both by changing the test program (occurs several times in runtime/race/output_test.go) to end with:

        <-done
        _ = x
}

[griesemer]

See also: #8560 (comment)

[danscales]

This test is clearly working now, since we have enabled -G=3 by default. I don't see any code in TestOutput() that disables the test. On the other hand, there is still the difference in behavior between -G=3 and -G=0 mode for the program described above. Haven't tried to narrow this down further as to when/how this was fixed.

[griesemer]

Summary:

package main
func main() {
        p := true
        func() {
                p = true
        }()
}

The old compiler (-G=0) does not report an error.
The new compiler (-G=3, default) reports: p declared but not used. So does go vet.

Since go vet is reporting this error there are probably few programs in the wild that rely on gc's old behavior. It should be reasonably safe to accept the new behavior. I am leaning towards the correct behavior. Still, NeedsDecision.

@rsc any thoughts?

[cuonglm]

FYI, @mdempsky has a CL to make old typechecker also reports the error: https://go-review.googlesource.com/c/go/+/238317

[griesemer]

Filed proposal #49214 to decide whether we can make this change.

[griesemer]

Per #49214 we will accept the correct behavior and document it in the release notes. If this ends up breaking a lot of code, we can re-open.