cmd/vet: false positive for user-defined verbs

[kortschak]

What version of Go are you using (go version)?

$ go version
go version go1.13.6 linux/amd64

Does this issue reproduce with the latest release?

Yes

What operating system and processor architecture are you using (go env)?

go env Output

$ go env
GO111MODULE="on"
GOARCH="amd64"
GOBIN="/home/user/bin"
GOCACHE="/home/user/.cache/go-build"
GOENV="/home/user/.config/go/env"
GOEXE=""
GOFLAGS=""
GOHOSTARCH="amd64"
GOHOSTOS="linux"
GONOPROXY=""
GONOSUMDB=""
GOOS="linux"
GOPATH="/home/user"
GOPRIVATE=""
GOPROXY="https://proxy.golang.org,direct"
GOROOT="/home/user/go"
GOSUMDB="sum.golang.org"
GOTMPDIR=""
GOTOOLDIR="/home/user/go/pkg/tool/linux_amd64"
GCCGO="gccgo"
AR="ar"
CC="gcc"
CXX="g++"
CGO_ENABLED="1"
GOMOD="/home/user/src/github.com/kortschak/loopy/go.mod"
CGO_CFLAGS="-g -O2"
CGO_CPPFLAGS=""
CGO_CXXFLAGS="-g -O2"
CGO_FFLAGS="-g -O2"
CGO_LDFLAGS="-g -O2"
PKG_CONFIG="pkg-config"
GOGCCFLAGS="-fPIC -m64 -pthread -fmessage-length=0 -fdebug-prefix-map=/tmp/go-build878536771=/tmp/go-build -gno-record-gcc-switches"

What did you do?

~/src/github.com/kortschak/loopy/cmd/bundle [master*]$ go test
# github.com/kortschak/loopy/cmd/bundle
./bundle.go:64:3: Fprintf format %60a has unknown verb a
~/src/github.com/kortschak/loopy/cmd/bundle [master*]$ echo $?
2

What did you expect to see?

No complaint.

What did you see instead?

Vet failure causing test failure (well, it would be a test failure if there were a test).

Additional information

The %a verb is non-standard verb, an addition to the biological sequence types provided by bíogo through their fmt.Formatter satisfaction. However, they have been in existence since 2013 and when they were included (as is still true), there was no documentation stating that only verbs listed in the fmt package for the built-in types should be used.

[robpike]

If the type being printed implements fmt.Format, vet is supposed to allow any verb. I distinctly remember writing that code and I see its implementation in the source (although much has changed since then), but it depends on type checking working.

Or maybe it never worked, but it should.

[tiriplicamihai]

Can I help with this?

[josharian]

@tiriplicamihai go for it

[tiriplicamihai]

I will try to investigate tomorrow evening (EET here).

[tiriplicamihai]

Hey @kortschak vet seems to properly work. I wrote this dummy example and the command runs without any issues:

package main

import (
    "fmt"
    "os"
)

type A struct{}

func (_ *A)  Format(f fmt.State, c rune) {}

func main() {
     out, _ := os.Create("cucu")
     defer out.Close()
     a := &A{}
     fmt.Fprintf(out, "%a", a)
}

I looked a bit over your code and it seems the dependency does not actually implement the Formatter interface: https://github.com/biogo/biogo/blob/master/seq/seq.go (at least not in the version I found on GH).

My 2c here is that you either change '%a' to an allowed verb or submit a request to have sequnece implement the Formatter interface.

@josharian I think this issue can be closed.

[josharian]

I’ll let @kortschak confirm and close if appropriate.

[kortschak]

It's not correct that the type doesn't implement fmt.Formatter. The interface doesn't, but the concrete type does.

[kortschak]

It does however explain what the problem is. Clearly vet is not considering whether the arg to fmt.Printf is an interface type when it's looking at the method set. Probably it should not try to look at interface types since the concrete value held in the interface may satisfy the fmt.Formatter interface.

You can see this kind of behaviour in vet for the common verbs here

package main

import "fmt"

func main() {
	var i interface{} = 1
	fmt.Printf("%s", i)
}

This does not cause a vet warning, but the equivalent without the interface {} type decl does. The vet check on fmt.Formatter should follow this behaviour.

In src/cmd/vendor/golang.org/x/tools/go/analysis/passes/printf/printf.go, isFormatter should optimistically return true if the type is an interface, something along the lines of (although named types may need to be unwrapped)...

func isFormatter(typ types.Type) bool {
        if _, ok := typ.Underlying().(*types.Interface); ok {
                return true
        }
        obj, _, _ := types.LookupFieldOrMethod(typ, false, nil, "Format")
        fn, ok := obj.(*types.Func)
        if !ok {
                return false
        }
        sig := fn.Type().(*types.Signature)
        return sig.Params().Len() == 2 &&
                sig.Results().Len() == 0 &&
                isNamed(sig.Params().At(0).Type(), "fmt", "State") &&
                types.Identical(sig.Params().At(1).Type(), types.Typ[types.Rune])
}

[kortschak]

If that's acceptable, I can send a CL. Presumably this goes to x/tools and will then get vendored in to the project source from there.

[gopherbot]

Change https://golang.org/cl/217180 mentions this issue: go/analysis/passes/printf: give leeway for fmt.Formatter satisfaction

[findleyr]

It looks like this is done. @kortschak can this be closed?

[kortschak]

Yes, I think that's OK. Maybe check with @robpike that that is a complete fix.