Bug report
Bug description:
Since gh-44376, Node.writexml() (and so toxml() and toprettyxml()) of xml.dom.minidom writes the namespace declarations that are missing from the document. An attribute created with the DOM Level 1 API (setAttribute(), createAttribute(), or a document parsed without namespace processing) has prefix set to None even if its name is xmlns:p. _fixup_namespaces() takes the declared prefix from Attr.prefix, so it records such an attribute as a declaration of the default namespace instead of the prefix p.
from xml.dom import expatbuilder, minidom
NS = "urn:p"
XSD = "http://www.w3.org/2001/XMLSchema"
# 1. The prefix of the element is declared with setAttribute()
doc = minidom.Document()
root = doc.appendChild(doc.createElementNS(NS, "p:root"))
root.setAttribute("xmlns:p", NS)
print(doc.toxml())
# 2. A prefix is declared on an element in no namespace
doc = minidom.Document()
root = doc.appendChild(doc.createElement("root"))
root.setAttribute("xmlns:xsi", "urn:xsi")
print(doc.toxml())
# 3. A prefix is bound to the namespace of the element
doc = minidom.Document()
schema = doc.appendChild(doc.createElementNS(XSD, "schema"))
schema.setAttribute("xmlns:x", XSD)
out = doc.toxml()
print(out)
print(minidom.parseString(out).documentElement.namespaceURI)
# 4. A document parsed without namespace processing
src = '<root xmlns:p="urn:p"><p:child/><other/></root>'
dom = expatbuilder.parseString(src, namespaces=False)
print(dom.documentElement.toxml())
Output on main:
<?xml version="1.0" ?><p:root xmlns:p="urn:p" xmlns:p="urn:p"/>
<?xml version="1.0" ?><root xmlns="" xmlns:xsi="urn:xsi"/>
<?xml version="1.0" ?><schema xmlns:x="http://www.w3.org/2001/XMLSchema"/>
None
<root xmlns="" xmlns:p="urn:p"><p:child/><other/></root>
Expected output:
<?xml version="1.0" ?><p:root xmlns:p="urn:p"/>
<?xml version="1.0" ?><root xmlns:xsi="urn:xsi"/>
<?xml version="1.0" ?><schema xmlns="http://www.w3.org/2001/XMLSchema" xmlns:x="http://www.w3.org/2001/XMLSchema"/>
http://www.w3.org/2001/XMLSchema
<root xmlns:p="urn:p"><p:child/><other/></root>
The first result has a duplicate attribute and is not well-formed. The second and the fourth get a spurious xmlns="". In the third, the element is considered to have its namespace declared as the default one already, so it is written without a declaration and loses its namespace when parsed again.
For cases 1, 2 and 4, 3.13 (before gh-44376) writes the expected output, so these are regressions. Case 3 does not work there either, but it is what gh-44376 is meant to handle.
The prefix should be derived from the attribute name, as is done for the rest of the declarations in that function.
This is related to gh-158208, but has a different cause: that one is about a plain xmlns attribute on an element in no namespace, and the proposed fixes for it do not change the results above.
CPython versions tested on:
CPython main branch
Operating systems tested on:
Windows
Linked PRs
Bug report
Bug description:
Since gh-44376,
Node.writexml()(and sotoxml()andtoprettyxml()) ofxml.dom.minidomwrites the namespace declarations that are missing from the document. An attribute created with the DOM Level 1 API (setAttribute(),createAttribute(), or a document parsed without namespace processing) hasprefixset toNoneeven if its name isxmlns:p._fixup_namespaces()takes the declared prefix fromAttr.prefix, so it records such an attribute as a declaration of the default namespace instead of the prefixp.Output on main:
Expected output:
The first result has a duplicate attribute and is not well-formed. The second and the fourth get a spurious
xmlns="". In the third, the element is considered to have its namespace declared as the default one already, so it is written without a declaration and loses its namespace when parsed again.For cases 1, 2 and 4, 3.13 (before gh-44376) writes the expected output, so these are regressions. Case 3 does not work there either, but it is what gh-44376 is meant to handle.
The prefix should be derived from the attribute name, as is done for the rest of the declarations in that function.
This is related to gh-158208, but has a different cause: that one is about a plain
xmlnsattribute on an element in no namespace, and the proposed fixes for it do not change the results above.CPython versions tested on:
CPython main branch
Operating systems tested on:
Windows
Linked PRs