cmd/vet: false positive about recursive call to String method

[marat-rkh]

Please answer these questions before submitting your issue. Thanks!

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

go1.9.3 darwin/amd64

Does this issue reproduce with the latest release?

Yes

What did you do?

Run go vet and go run on the following file:

package main

import "fmt"

type X string
func (x *X) String() string { return fmt.Sprint(">> ", &x) }

func main() {
	var x X = ""
	fmt.Println(x.String())
}

What did you expect to see?

go vet reports no errors as program runs without problems, prints pointer address and exits normally.

What did you see instead?

go vet reports arg &x in Sprint call causes recursive call to String method.

[mvdan]

This might be because, until recently, vet did not always have type information. Should be fairly easy to make the check precise.

[marat-rkh]

As far as I understand, check for print functions will be executed by default before tests in go 1.10. Check results will be treated as errors. Can this be a reason to fix this before 1.10 or, at least, disable the check for this particular case?

[mvdan]

That's a good point. Not a regression in vet, but could qualify as a regression in test. /cc @rsc

[mvdan]

Actually, I'm not 100% sure if fmt is right here. I know that the program does run, but I don't follow the logic as to why it doesn't crash with an infinite recursion. /cc @robpike

[jimmyfrasche]

@mvdan the address is taken twice (once implicitly then once explicitly) so it's a **X which has no method set

[robpike]

Vet should not complain here. The code is legal and should (and does) print the address of x, of type **X. The argument to Sprint has no method set.

[mvdan]

Thank you both. I do understand the double pointer - what still confuses me at times is what rules apply when a method of *T can be called on T (and vice versa). Thanks for confirming that the code is right before I looked into a fix.

[mvdan]

Indeed - the problem is that vet tracked only the types that had a String() string method declared on them, but it did not track whether the receiver in the method had a pointer type. With that extra information, we can now fix this false positive.

[gopherbot]

Change https://golang.org/cl/90417 mentions this issue: cmd/vet: **T is not Stringer if *T has a String method