go: encoding/xml: cannot marshal self-closing tag

Any news on this?

Any news?

Any news?

This is a good idea since I’m running into the same issue. Is this idea still active or is it “dead”?

Do this

import (
	"regexp"
)

func SelfClosing(xml []byte) ([]byte, error) {
	regex, err := regexp.Compile(`></[A-Za-z0-9_]+>`)
	return regex.ReplaceAll(xml, []byte("/>")), err
}

Any chances of this being addressed? Stumbled upon this issue while working with the XMPP protocol. Some clients will only accept <required/>, but not <required></required>.

Change https://go.dev/cl/469495 mentions this issue: encoding/xml: Add EmptyElement token.

We still need the ability to output non-selfclosing pairs of tags. The idea is to give the user the option to make certain fields self closing, not to do that for all fields.

just created a mod with patch in https://go.dev/cl/469495 applied.

and fixed a indent issue. in case someone need a workaround, here is the repo:

https://github.com/ttys3/go-xml#usage

is anyone aware of a custom marshaller example that can achieve self-closing aka empty tags?

@iWdGo how does this solve the original question?

The desire is to produce either <tag /> or <tag>notempty</tag>. Not <tag></tag>.

For the same reason this bug exists in the first place; not all XML parsers handle this correctly and I wanted maximum flexibility.

I’m fine with doing it the other way though.

Change https://golang.org/cl/59830 mentions this issue: encoding/xml: add 'allowempty' tag for self-closing XML tags

I like this idea; though i suspect the community will have to agree on it (and also the naming of the tag)

However, I whipped up a quick implementation at https://go-review.googlesource.com/59830 .

I’ll polish it up and add tests if this proposal actually goes through.

It’s also not correct - it would, for example, also apply the replacement in comments.

@jybp we use this

func forceSelfClosingTags(b []byte) []byte {

	emptyTagIdxs := regexp.MustCompile(`<(\w+)></\w+>`).FindAllSubmatchIndex(b, -1)

	var nb []byte

	for _, idx := range emptyTagIdxs {
		// get everything in b up till the first of the submatch indexes (this is the start of an
		// "empty" <thing></thing> tag), then get the name of the tag and put it in a self-closing
		// tag.
		nb = append(b[0:idx[0]], fmt.Sprintf("<%s/>", b[idx[2]:idx[3]])...) //nolint: gocritic

		// finally, append everything *after* the submatch indexes
		nb = append(nb, b[len(b)-(len(b)-idx[1]):]...)
	}

	return nb
}

I’m not sure that there’s any other way

There is, we simply support this for marshalling structs – the primary use case – and not for using EncodeToken. (which is what I’ve been suggesting)

What we can also do is allow Token to have a SelfClosingElement variant so that logic for choosing when to emit empty tags lives in the caller

Thanks, I’ll mark this as a release blocker for 1.13 so that we decide on that CL.

Go is an open source project. Want to send a patch?

We hope to do a general overhaul of the JSON and XML patches and feature requests in the 1.13 timeframe, but we hoped to do it for 1.12 also and didn’t make it.

Can we expect this to be implemented in Go 1.10?

Currently we have a Go library, that needs to communicate with a remote API endpoint, which can only understand empty XML elements with self-closing tags.

With the current implementation of XML package we are not able to use Go in order to communicate with that API endpoint. I realize that this is an issue of the remote side by not supporting both ways of specifying empty elements, but having the option to define how empty elements are defined in Go structs would be very useful and would allow us to overcome such edge cases.

Thanks, Marin

not all XML parsers handle this correctly and I wanted maximum flexibility

If a parser did not handle this you couldn’t mix the two anyways, you’d just have to turn it on or off entirely to make sure that parser would work.

In case there’s confusion, I’m not suggesting that we always do self-closing tags or non-self-closing tags, but that an individual *Encoder should be configurable to do one or the other.

Oh, I see, you’re suggesting this be global. Hm. I don’t really like the idea of making this a global change, but we could also make it an option on the marshaler.