Function stream_extract 

pub fn stream_extract<F>(
    xml: &str,
    records_element_path: Option<&str>,
    on_record: F,
) -> Result<(), FaucetError>
where
    F: FnMut(Value),

Available on crate features source-rest and source-xml only.

Walk an XML document with quick_xml::Reader::read_event and invoke on_record once per element whose path matches the dot-separated records_element_path selector. Records are materialised as JSON values in the same shape xml_to_json would produce — attributes become @key entries, repeated children become arrays, and a single #text child is flattened to a bare string.

When records_element_path is None the entire document is emitted as a single record (matches the eager xml_to_json behaviour).

The key difference from xml_to_json is that subtree JSON values are only materialised while inside a matched element — surrounding elements are observed via the event stream but never accumulated, which bounds memory to one matched element + the path stack regardless of total document size. Combined with batched yielding in crate::stream::XmlStream’s stream_pages, this keeps client-side memory at O(batch_size * record_size) even for multi-gigabyte payloads.