Issue 28238: In xml.etree.ElementTree findall() can't search all elements in a namespace

In the example there are two namespaces in one document, but it is impossible to search all elements only in one namespace:

>>> import xml.etree.ElementTree as etree
>>>
>>> s = '<feed xmlns="http://def" xmlns:x="http://x"><a/><x:b/></feed>'
>>>
>>> root = etree.fromstring(s)
>>>
>>> root.findall('*')
[<Element '{http://def}a' at 0xb73961bc>, <Element '{http://x}b' at 0xb7396c34>]
>>>
>>> root.findall('{http://def}*')
[]
>>>


And same try with site package lxml works fine:

>>> import lxml.etree as etree
>>>
>>> s = '<feed xmlns="http://def" xmlns:x="http://x"><a/><x:b/></feed>'
>>>
>>> root = etree.fromstring(s)
>>>
>>> root.findall('*')
[<Element {http://def}a at 0xb70ab11c>, <Element {http://x}b at 0xb70ab144>]
>>>
>>> root.findall('{http://def}*')
[<Element {http://def}a at 0xb70ab11c>]
>>>

lxml has a couple of nice features here:

- all tags in a namespace: "{namespace}*"
- a local name 'tag' in any (or no) namespace: "{*}tag"
- a tag without namespace: "{}tag"
- all tags without namespace: "{}*"

"{*}*" is also accepted but is the same as "*". Note that "*" is actually allowed as an XML tag name by the spec, but rare enough to hijack it for this purpose. I've actually never seen it used anywhere in the wild.

lxml's implementation isn't applicable to ElementTree (searching has been subject to excessive optimisation), but it shouldn't be hard to extend the one in ET's ElementPath.py module, as well as Element.iter() in ElementTree.py, to support this kind of tag comparison.

PR welcome.

lxml's tests are here (and in the following test methods):

https://github.com/lxml/lxml/blob/359f693b972c2e6b0d83d26a329d2d20b7581c48/src/lxml/tests/test_etree.py#L2911

Note that they actually test the deprecated .getiterator() method for historical reasons. They should probably call .iter() instead these days. lxml's ElementPath implementation is under src/lxml/_elementpath.py, but the tag comparison itself is done elsewhere in Cython code (here, in case it matters:)

https://github.com/lxml/lxml/blob/359f693b972c2e6b0d83d26a329d2d20b7581c48/src/lxml/apihelpers.pxi#L921-L1048

PR submitted, feedback welcome.

BTW, I found that lxml and ET differ in their behaviour when searching for '*'. ET takes it as meaning "any tree node", whereas lxml interprets it as "any Element". Since ET's parser does not create comments and processing instructions by default, this does not make a difference in most cases, but when the tree contains comments or PIs, then they will be found by '*' in ET but not in lxml.

At least for "{*}*", they now both return only Elements. Changing either behaviour for '*' is probably not a good idea at this point.

New changeset 47541689ccea79dfcb055c6be5800b13fcb6bdd2 by Stefan Behnel in branch 'master':
bpo-28238: Implement "{*}tag" and "{ns}*" wildcard tag selection support for ElementPath, and extend the surrounding tests and docs. (GH-12997)
https://github.com/python/cpython/commit/47541689ccea79dfcb055c6be5800b13fcb6bdd2

versions: + Python 3.8, - Python 3.6