Technically this isn't really an issue - the output is correct according to the XML spec - but it causes several tests to fail seemingly at random.
For example, see REC/test-7.1.1 (which is currently disabled pending investigation into this issue).
Making the test harness smart enough to recognise XML output and do canonicalization first would be a reasonable fix. Alternatively, sorting by prefix before writing (and fixing up the associated expected output) would be another way to fix it.
Technically this isn't really an issue - the output is correct according to the XML spec - but it causes several tests to fail seemingly at random.
For example, see REC/test-7.1.1 (which is currently disabled pending investigation into this issue).
Making the test harness smart enough to recognise XML output and do canonicalization first would be a reasonable fix. Alternatively, sorting by prefix before writing (and fixing up the associated expected output) would be another way to fix it.