bpo-30485: Change the prefix for defining the default namespace in ElementPath from None to ''

[scoder]

There is existing code that uses that and it's more convenient to have an all-string-keys dict (e.g. when sorting items etc.).

https://bugs.python.org/issue30485

[tirkarthi]

lxml now supports both None and '' with commit lxml/lxml@aefded0. CPython with this PR only supports '' though initial PR for issue supported only None and ValueError for '' was incompatible. Is there a plan for lxml to drop support for None in future to unify with CPython or for CPython to support None and '' with lxml?

[scoder]

I was considering allowing both in ET as well (and implemented it along the way), but decided against it because a) the implementation is quite ugly, b) there are probably not that many users who go from lxml to ET, and c) ET's API generally has almost no exposure of prefixes whatsoever, so it doesn't matter there. Using prefixes in ElementPath expressions is not even the recommended way, because self-contained string expressions with explicit "{ns}tag" notation are much nicer to handle overall.

lxml, on the other hand, returns a dict with None as prefix when you let it collect the namespaces defined on an Element and its parents (nsmap), so not supporting both there would be unnatural, and changing this is essentially impossible. At least for now, I think it's ok to support only the simpler (and more convenient) empty string in ElementTree, and I also don't see lxml deprecate or remove support for None as prefix. Adapting the docs should fix it over time.

[serhiy-storchaka]

This looks like a breaking change, so we should emit a deprecation warning before dropping the support of None and support both prefixes in meantime.

[scoder]

@serhiy-storchaka, the feature itself was only added to ElementTree last weekend and never released even in an alpha. The only question is whether anyone sees a problem with '' as a prefix key (which was actually in the original contribution) versus the None which lxml used for the last two years (which I do not consider ideal any more, specifically not for the context of ElementTree). It's not ideal to have an ambiguity about the "one way to do it", but they both have their long standing user bases, and since this is a new feature for ET, I think it should have at least its own clean implementation.

For me, the plan forward is to support '' in both, and leave None only as the alternative in lxml where some use cases suggest it (because lxml's tree model is prefix aware, while ET is not).

[serhiy-storchaka]

Ah, then all right!

[bedevere-bot]

@scoder: Please replace # with GH- in the commit message next time. Thanks!