b, err := lang.MarshalData(p, dataType, data)

func exampleCommand(p *lang.Process) error {
    data := map[string]string {
        "foo": "hello foo",
        "bar": "hello bar",
    }

    dataType := "json"

    b, err := lang.MarshalData(p, dataType, data)
    if err != nil {
        return err
    }

    _, err := p.Stdout.Write(b)
    return err
}

package lang

import (
    "errors"
    "fmt"
)

type MarshallerT func(*Process, any) ([]byte, error)

var (
    // _marshallers defines the Go functions for converting a Go interface into a murex data type
    _marshallers = make(map[string]MarshallerT)
)

func RegisterMarshaller(dataType string, marshaller MarshallerT) {
    if _marshallers[dataType] != nil {
        panic(fmt.Sprintf("marshaller already exists for %s", dataType))
    }
    _marshallers[dataType] = marshaller
}

// MarshalData is a global marshaller which should be called from within murex
// builtin commands (etc).
// See docs/apis/marshaldata.html for more details
func MarshalData(p *Process, dataType string, data any) ([]byte, error) {
    // This is one of the very few maps in Murex which isn't hidden behind a sync
    // lock of one description or other. The rational is that even mutexes can
    // add a noticeable overhead on the performance of tight loops and I expect
    // this function to be called _a lot_ while also only needing to be written
    // to via code residing in within builtin types init() function (ie while
    // murex is effectively single threaded). So there shouldn't be any data-
    // races -- PROVIDING developers strictly follow the pattern of only writing
    // to this map within init() func's.
    if _marshallers[dataType] == nil {
        return nil, errors.New("I don't know how to marshal `" + dataType + "`.")
    }

    b, err := _marshallers[dataType](p, data)
    if err != nil {
        return nil, errors.New("[" + dataType + " marshaller] " + err.Error())
    }

    return b, nil
}