text/template: Inconsistent handling off nil values in if and with

[bep]

package main

import (
	"bytes"
	"fmt"
	"log"
	"text/template"
)

type Niller interface {
	Nil() Niller
}

type Nill struct {
	nilv Niller
}

func (n Nill) Nil() Niller {
	return n.nilv
}

func main() {

	tmpl, err := template.New("").Parse(`{{ .Nil }}: {{ with .Nil }}Failed, got {{ . }}{{ else }}OK{{ end }}`)
	if err != nil {
		log.Fatal(err)
	}

	var (
		nil1 Niller = (*Nill)(nil)
		nil2 Niller
		nil3 *Nill
		nil4 Niller = nil
	)

	for i, niller := range []Niller{nil1, nil2, nil3, nil4} {
		var buff bytes.Buffer
		err = tmpl.Execute(&buff, &Nill{nilv: niller})
		if err != nil {
			log.Fatal(err)
		}
		fmt.Println(i+1, buff.String())
	}
}

The above prints:

1 <nil>: Failed, got <nil>
2 <nil>: OK
3 <nil>: Failed, got <nil>
4 <nil>: OK

This issue makes more sense if you also glance at these:

• text/template: panics on method on nil interface value #30143 panic on nil interface value; it was when implementing workarounds for that issue that I got other curious nullpointers which I had hard time understanding. Unfortunately I found no workarounds that both avoid panics and work sensible with if and where.
• text/template: allow callers to override IsTrue #28391 about the subjective truth in template.IsTrue

The documentation for both if and with is similar:

"If the value of the pipeline is empty, no output is generated"

I'm not sure what the definition of "empty" in the above, but as all of the above cases outputs <nil>, I assume it is the truthines defined in template.IsTrue that is used.

[bcmills]

CC @robpike @mvdan for text/template

[bcmills]

Note that nil2 and nil4 in your example are identical values, and nil1 and nil3 are identical when stored in a []Niller.

[bcmills]

“empty” in the context of “If the value of the pipeline is empty” means “equal to the zero value of its type”. At the core, this is another instance of https://golang.org/doc/faq#nil_error, and text/template is behaving as documented.

[bep]

@bcmills I can sympathize with your urge to be effective and close this issue. But this is a real issue that need some kind of action, even if just a documentation update.

In general, I think every person who uses the Go template API expect this to not panic with a nilpointer, if at all possible:

{{ with .MaybeNil }}{{ .Hello }}{{ end }}

Even after carefully reading the documentation, I think most people would expect the above to not crash. And note that this is less obvious and much harder to debug in a dynamic template context.

All of my 4 examples (even if there are duplicates) would be considered empty by most people.

There is nothing here that talks about "empty" (or even zero values), so that is, I guess, a vague sentence that could be interprited to mean just about anything.

/cc @mvdan @ianlancetaylor

[bep]

I'm sure I'm talking to myself here, but if you expand the program above:

package main

import (
	"bytes"
	"fmt"
	"log"
	"text/template"
)

type Niller interface {
	Nil() Niller
	Foo() string
}

type Nill struct {
	nilv Niller
}

func (n Nill) Nil() Niller {
	return n.nilv
}

func (n Nill) Foo() string {
	return "asdf"
}

func main() {

	tmpl, err := template.New("").Parse(`{{ with .Nil }}Failed, got {{ . }}{{ else }}OK{{ end }}`)
	if err != nil {
		log.Fatal(err)
	}

	var (
		nil1 Niller = (*Nill)(nil)
		nil2 Niller
		nil3 *Nill
		nil4 Niller = nil
	)

	for i, niller := range []Niller{nil1, nil2, nil3, nil4} {

		var buff bytes.Buffer
		n := &Nill{nilv: niller}
		a, b := template.IsTrue(n.Nil())
		err = tmpl.Execute(&buff, n)
		if err != nil {
			fmt.Println("error:", err)
			continue
		}
		fmt.Println(i+1, a, b, buff.String())
	}
}

You get:

1 false true Failed, got <nil>
2 false true OK
3 false true Failed, got <nil>
4 false true OK

I have looked at the code, and I fail to understand it. template.IsTrue, which is used by with and if, all returns correctly false for all 4 cases, yet with (and if) behaves as if it was truthy.

[mvdan]

I'm sure I'm talking to myself here

I did follow this thread yesterday, but I had work and a meetup to speak at, so I didn't have a chance to respond properly :)

I see you've opened a new issue, so I'll just focus on that one.