Did something change related to pypi or pip today? Today we hit a bunch of build failures about requires-python. We learned that many internal packages had requires-python: >=2.7,<3 in them (by accident), despite being written for Python 3. Silly mistake on our part that somehow did not surface for over a year.
What’s not clear to me is why this was working until just a few hours ago (April 26, first failure seen at ~1PM PT, definitely was working around 5AM PT). Just trying to figure out how we can get ourselves running again while we go update all package versions and dependencies to remove the incorrect requires-python specification.
Any ideas? I know this is strange but just confused if I missed some notification somewhere.
Requires-Python has been on PyPI for 5+ years iirc. Most likely you were on an old pip version that did not respect it previously? Debian-derived distributions ship with a pip without Requires-Python support until very recently, for example.
@uranusjr that’s a really excellent suggestion - however I know we’ve been using a newer version of pip for a while. (We checked previous and current AMIs and they’ve all seem to have same version of pip).
I was hoping there was some magical announcement I missed about PyPI API responses that would explain the delta
It’s all very surreal to us, because this seemed to stop working totally divorced from when we upgraded pip.
Oh actually I just had another thought. You mentioned you were seeing the issue for your private packages, so you’re not hosting them on pypi.org? What are you using to host them instead? I think I heard there’s a server implementation (forgot which) that did not implement Requires-Python as well…
I’m not sure if it relative to my issue. We also recent had to stop up grade pip and freeze back to version 19.?? Instead of 20.* to avoid currently unresolved issue with internal packages
