`xml` - XML encode / decode

Enable with use xml;. Parses and serializes XML over an opaque xml.Value, designed exactly like json and toml: decode returns a handle you walk with accessors, and a small build surface constructs a tree to encode. Hand-rolled (no encoding/xml, which is reflect-bound), so it runs on both binaries.

Unlike JSON's map/list/scalar tree, XML is a tree of elements - each with a tag name, ordered attributes, and ordered children that mix elements and text. The accessors reflect that shape, and navigation uses a small XPath-style path dialect in place of JSON Pointer.

jennifer

use io;
use xml;

def root as xml.Value init xml.decode(
    "<library><book id=\"1\"><title>Go</title></book></library>");
io.printf("%s\n", xml.tag($root));                         # library
io.printf("%s\n", xml.attr(xml.get($root, "book"), "id")); # 1
io.printf("%s\n", xml.text(xml.get($root, "book/title"))); # Go

Decode / encode

Call	Returns	Notes
`xml.decode(s)`	`xml.Value`	Parse a document into the root element. Errors (with line / column) on malformed XML.
`xml.encode(v)`	`string`	Serialize compactly.
`xml.encodePretty(v)`	`string`	Serialize with 2-space indentation.

decode handles elements, attributes, text, <![CDATA[...]]> (as text), comments, processing instructions, an XML declaration, and a DOCTYPE (the last four are skipped, not kept). The five predefined entities (< > & ' ") and numeric character references (A, A) are decoded; an unknown entity is an error. Namespace prefixes are kept verbatim in the name (ns:tag), and xmlns declarations are ordinary attributes.

encodePretty indents element-only content, but an element that contains any text child is emitted inline so its character data stays byte-exact.

Reading

Every accessor takes an xml.Value. xml.get / xml.findAll / xml.has take a path as a second argument (see below).

Call	Returns	Notes
`xml.typeOf(node)`	`string`	`"element"` or `"text"`.
`xml.tag(node)`	`string`	The element's tag name (with prefix, if any).
`xml.text(node)`	`string`	The concatenated character data of the element's direct text children (or a text node's own string).
`xml.attr(node, name)`	`string`	An attribute's value; errors if the attribute is absent.
`xml.hasAttr(node, name)`	`bool`	Whether the attribute is present.
`xml.attrs(node)`	`list of string`	The attribute names, in document order.
`xml.children(node)`	`list of xml.Value`	The element children (text children are excluded).
`xml.get(node, path)`	`xml.Value`	The first element matching the path; errors if none.
`xml.findAll(node, path)`	`list of xml.Value`	Every element matching the path (empty if none).
`xml.has(node, path)`	`bool`	Whether the path matches at least one element.

Path dialect

A path is /-separated steps, evaluated against the element children of the node (a leading / is optional; "" addresses the node itself):

Step	Matches
`name`	every element child with that tag.
`name[k]`	the 1-based `k`-th element child with that tag.
`*`	every element child, whatever its tag.

xml.findAll($root, "book") returns all book children; xml.get($root, "book[2]/title") drills to the second book's title. Attributes and text are read with xml.attr / xml.text after navigating, not addressed inside the path. A malformed path is an error; a well-formed path that matches nothing is an empty findAll (and a false has, and an error from get).

Building

The build surface is non-mutating - each call returns a fresh xml.Value, the same $v = xml.set(...) idiom as json / toml:

Call	Returns	Notes
`xml.element(name)`	`xml.Value`	A new empty element.
`xml.setAttr(node, name, value)`	`xml.Value`	The element with the attribute added or updated.
`xml.setText(node, s)`	`xml.Value`	The element with its children replaced by one text node.
`xml.append(parent, child)`	`xml.Value`	The parent with `child` (an `xml.Value` element) appended.

jennifer

use xml;

def note as xml.Value init xml.element("note");
$note = xml.setAttr($note, "type", "info");
$note = xml.setText($note, "hi & bye");
def page as xml.Value init xml.append(xml.element("page"), $note);
# <page><note type="info">hi &amp; bye</note></page>

Notes and limits

convert.typeOf reports "object"; convert.objectType reports "xml.Value". A bare $v at the REPL (or %v) shows the compact XML.
Comments, PIs, the prolog, and the DOCTYPE are dropped on decode, so a decode / encode round-trip is faithful for the element tree but not for those. The one root element is required (XML has exactly one).
Namespaces are lexical: prefixes are kept as written, not resolved to their xmlns URIs. Full namespace resolution and richer XPath (predicates, @attr / text() path steps) are future extensions.
Element nesting is capped on both decode and encode: a deeper document (or a tree built that deep with xml.append) raises a catchable error rather than overflowing the stack. The limit is shared by every parser in the toolchain and is set per binary: 1000 levels on the default jennifer (Go's growable stack), 64 on jennifer-tiny (its fixed 4 MB stack). No entity is expanded beyond the five predefined names and numeric references, so there is no entity-expansion ("billion laughs") blow-up, and nothing is fetched externally - an unknown or external entity is simply an error.

#xml - XML encode / decode

#Decode / encode

#Reading

#Path dialect

#Building

#Notes and limits

#See also