libxml2 2.9.12+dfsg-3 causes regression

[Corey Bryant]

root@i1:~# cat example.xml
<domain type="qemu">
  <name>fake-name</name>
  <os>
    <type/>
  </os>
  <devices>
    <interface type="bridge">
      <mac address="22:52:25:62:e2:aa"/>
      <model type="virtio"/>
      <source bridge="br0"/>
      <mtu size="9000"/>
      <target dev="nicdc065497-3c"/>
      <bandwidth>
        <inbound average="100" peak="200" burst="300"/>
        <outbound average="10" peak="20" burst="30"/>
      </bandwidth>
    </interface>
  </devices>
</domain>

root@i1:~# cat xml.py
from lxml import etree

def parseXML(xmlFile):
    # Parse the xml
    with open(xmlFile) as fobj:
        xml = fobj.read()

    doc = etree.fromstring(xml)
    ret = doc.findall('./devices/interface')
    node_xml = etree.tostring(ret[0]).decode()
    print("node_xml={}".format(node_xml))

if __name__ == "__main__":
    parseXML("example.xml")

== Expected result ==

root@i1:~# python3 xml.py
node_xml=<interface type="bridge">
      <mac address="22:52:25:62:e2:aa"/>
      <model type="virtio"/>
      <source bridge="br0"/>
      <mtu size="9000"/>
      <target dev="nicdc065497-3c"/>
      <bandwidth>
        <inbound average="100" peak="200" burst="300"/>
        <outbound average="10" peak="20" burst="30"/>
      </bandwidth>
    </interface>

== Actual Result ==

root@i1:~# python3 xml.py
node_xml=<interface type="bridge">
      <mac address="22:52:25:62:e2:aa"/>
      <model type="virtio"/>
      <source bridge="br0"/>
      <mtu size="9000"/>
      <target dev="nicdc065497-3c"/>
      <bandwidth>
        <inbound average="100" peak="200" burst="300"/>
        <outbound average="10" peak="20" burst="30"/>
      </bandwidth>
    </interface>
  </devices>
</domain>

This is broken in 2.9.12+dfsg-3 on impish.
It is not broken in 2.9.10+dfsg-6.3ubuntu0.1 on hirsute.

There are some fixes in master since the 2.9.12 release was cut.
https://gitlab.gnome.org/GNOME/libxml2/-/commits/master

I'm planning to pick all 3 of the following commits from upstream master since they are all intertwined. With this patches the regression is fixed.

85b1792e37b131e7a51af98a37f92472e8de5f3f:
Work around lxml API abuse. Make xmlNodeDumpOutput and htmlNodeDumpFormatOutput work with corrupted parent pointers.

13ad8736d294536da4cbcd70a96b0a2fbf47070c:
Add patch from upstream to fix regression in xmlNodeDumpOutputInternal. Commit 85b1792e could cause additional whitespace if xmlNodeDump was called with a non-zero starting level.

92d9ab4c28842a09ca2b76d3ff2f933e01b6cd6f:
Add patch from upstream to fix whitespace when serializing empty HTML documents.

[Launchpad Janitor (janitor)]

This bug was fixed in the package libxml2 - 2.9.12+dfsg-4

---------------
libxml2 (2.9.12+dfsg-4) unstable; urgency=medium

  * Team upload.
  * Add a few patches from upstream:
    + Work around lxml API abuse.
    + Fix regression in xmlNodeDumpOutputInternal. LP: #1943277
    + Fix whitespace when serializing empty HTML documents.
    + Forbid epsilon-reduction of final states.
    + Fix buffering in xmlOutputBufferWrite.

 -- Mattia Rizzolo <email address hidden> Fri, 10 Sep 2021 22:13:09 +0200