junit-xml output attribute can be too big to handle with -vv

Suppose you have a test that may fail with a very large diff; say:

def test_xyz() -> None:
    assert ["x"] * 10_000 == ["x\n"] * 10_000

With normal verbosity (0), the junit-xml output would look like this:

<?xml version="1.0" encoding="utf-8"?>
<testsuites>
 <testsuite errors="0" failures="1" hostname="..." name="pytest" skipped="0" tests="1" time="1.172" timestamp="...">
  <testcase classname="test_file" file="test_file.py" line="0" name="test_xyz" time="0.767">
   <failure message="AssertionError: assert ['x', 'x', 'x...'x', 'x', ...] == ['x\n', 'x\n'...', 'x\n', ...]
  At index 0 diff: 'x' != 'x\n'
  Use -v to get more diff">
    def test_xyz() -&gt; None:
&gt;       assert ["x"] * 10_000 == ["x\n"] * 10_000
E       AssertionError: assert ['x', 'x', 'x...'x', 'x', ...] == ['x\n', 'x\n'...', 'x\n', ...]
E         At index 0 diff: 'x' != 'x\n'
E         Use -v to get more diff

test_file.py:2: AssertionError
   </failure>
  </testcase>
 </testsuite>
</testsuites>

noting the <failure message='...'> tag's message and body both have a Use -v to get more diff.

If you do supply a -vv, then both the body of the <failure> tag, AND the content of the message attribute can be very large. The latter can be problematic for parsers, and projects like BeautifulSoup and lxml will fail to parse this correctly (truncating or crashing, respectively).

Maybe this is a limitation of the junit schema, but is there anything acceptable that can be done to reduce only the content of the message attribute? I think it's more expected and "fine" for the body of a tag to be really long, but an attribute less so.

applies to pytest 8.1.1

Start by reproducing the issue with the test example under normal verbosity and -vv, then trace the pytest junit-xml output path; the payload names no files or tests. Done means the failure body can remain detailed while the message attribute is safely bounded and the resulting XML remains parseable by common parsers.