xml.dom.minidom.Node.normalize() merges adjacent text nodes with node.data = node.data + child.data, once per absorbed node. A run of n text nodes copies the growing string n times, so it's quadratic.

from xml.dom.minidom import Document
doc = Document()
root = doc.appendChild(doc.createElement('r'))
for _ in range(32_000):
    root.appendChild(doc.createTextNode('x' * 48))
root.normalize()  # ~450 ms; 8k nodes ~25 ms, 16k ~95 ms

The parsers already merge character data, so you only hit this with documents built through the DOM API (appendChild/createTextNode, etc.).

Fix: collect the pieces for each run and ''.join them once at the end. I'll open a PR.

Linked PRs

• gh-158860: Make minidom Node.normalize() linear in adjacent text nodes #158861