cmd/vet: go vet incorrectly warns of "no formatting directive in Errorf call" for non core methods with same name

[piersy]

We have a log object it provides a method Errorf

Go vet prints a warning - "no formatting directive in Errorf call"

I would expect go vet to be able to determine that this Errorf call is not a call to the standard library Errorf function.

go version go1.5 linux/amd64

[bradfitz]

What is the signature of your Errorf method?

govet doesn't care about the standard library; it enforces the rule that if your function is named one of the popular -f ending functions, then it should probably be acting like them:

var printfList = map[string]int{
    "errorf":  0,
    "fatalf":  0,
    "fprintf": 1,
    "logf":    0,
    "panicf":  0,
    "printf":  0,
    "sprintf": 0,
}

This is probably working as intended, but please share more details.

[Setomidor]

I have exactly the same problem, and my Fatalf looks as follows:

func Fatalf(tag string, fullID string, format string, a ...interface{}) {

Is it the extra arguments that matter? The signature of the original method is:

func Fatalf(format string, v ...interface{})

[m3ng9i]

Similar problem. Here is the code:

// testvet.go
package main

import "fmt"

type S struct {
}

func (this *S) Printf(i int) {
    fmt.Println(i)
}

func main() {
    s := &S{}
    s.Printf(1)
}

After run go vet testvet.go, I got an error: testvet.go:16: constant 1 not a string in call to Printf.

But parameter of S.Printf() is exactly int, not string.

My go version: go1.5.1 windows/386

[ianlancetaylor]

@m3ng9i This is working as expected. Your choices are 1) rename your Printf function; 2) don't run vet.

[spenczar]

I'd like to revive this topic, motivated by gRPC's Errorf function which triggers vet.

gRPC has a public function with this signature:

func Errorf(c codes.Code, format string, a ...interface{}) error

Using this function causes vet to shriek, as noted in grpc/grpc-go#90. Given that grpc is maintained by core Go authors who are unwilling to rename their function, I think it's fair to say some more effort in go vet's side here would be worthwhile to expand its set of covered programs.

It seems to me that vet could be a little more intelligent here. It should be easily possible to identify the function's call signature and figure out whether its first parameter is a string. Vet could merely warn that it's unable to handle non-standard Printf calls, or perhaps it could try to suss out which parameter is a format string, and appropriately vet the arguments passed in.

This comment is a proposal to gauge interest; I'm happy to be the one to do the work of writing the CL but I want to make sure that this change to vet's behavior is desirable.

[gopherbot]

CL https://golang.org/cl/20163 mentions this issue.

[valyala]

The CL https://golang.org/cl/20163 breaks go vet -printfuncs=FooBarf if FooBarf is defined outside the current package. It looks like signature() returns nil in this case, which results in no formatting directive in FooBarf call warnings for each FooBarf(format, <args>) call.
I'd suggest falling back to the previous behavior if signature() returns nil.