Let’s fix extras in Core Metadata 3.0

Slides for a discussion in the PyCon US 2022 Packaging Summit. This probably won’t make much sense unless you read the minute notes. You can also find there other topics discussed during the summit.

https://hackmd.io/Rv_FmGp6TniDWM4EXyZmMw

Transcript

1. Let’s fix extras in Core Metadata 3.0 1
2. Extras Name normalisation and comparison PEP 685 Make sense of extra comparison in markers == “works” but is wrong Default extras 2
3. Name normalisation Covered by Brett’s PEP 685 but has consequences Extra name must be normalised in markers Incompatibility Easy for tools to get wrong 3
4. Extra in marker is a mess extra == "foo" is wrong pip install package[docs,tests] any(marker.evaluate({"extra": x}) for x in extras) No other operators make sense “Negative extra” is impossible Exclude a dependency when an extra is specified 4
5. extras instead of extra pytest>=6.0; 'tests' in extras Use not in for negation Special semantics for operators Normalise operands on comparison Should drop extra to avoid confusion Incompatibility 5
6. Default extra group I think Pradyun showed interests Probably can use help from interested folks A “default extra” specification in metadata Requires-Dist: package asks for that extra group Semantic incompatibility What happens when pip install package[another] ? package[] requests to not install the defaul group? 6
7. Proposals (1/2) Metadata 3.0 or not? PEP 685 already (slightly) broke compatibility Or we keep extra and stay on 2.x But if we bump to 3.0… Default extra group I should get in touch with Pradyun 7
8. Proposal (2/2) Drop old fields made problematic by PEP 621 Home-page The Author and Maintainer divide I think there are more There must be more we can do 8