10:15 <Hodgestar> mattip: Woot?

11:17 <mattip> any ideas why the hpy-0.0.4 extra_tests/hpy_tests/_vendored/conftest.py is not being picked up?

11:17 <mattip> I think the conftest.py one directory up is being pulled in, but not the _vendored one

13:14 <mattip> ahh, it is a limitation in pytest_addoption which must be in the toplevel conftest.py

13:35 <komasa> I vaguely recall pypy being slower than CPython when run under a debugger (specifically in PyCharm), but I can't rememember what the rough factor of this was. In a related question: Is there some easy way to keep a pypy and a cpython virtual env in sync in terms of installed packages and versions?

15:12 <fijal> tumbleweed: ping?

15:47 <tumbleweed> fijal: ohi (from CPT)

17:47 <jean-paul[m]> komasa: nix would be a good way to keep two such environments in sync

17:48 <jean-paul[m]> let mkenv = python: python.withPackages(ps: [ ... ]); in { cpy = mkenv pkgs.python3; pypy = mkenv pkgs.pypy3; }

17:49 <jean-paul[m]> (wrapped up in the new flake format for convenient cli access)

18:16 <komasa> jean-paul[m]: I have various packages in that env that are installed via --editable, and that would just not work with the nix philosophy at all

18:17 <jean-paul[m]> I don't know if the tension between "well-specified, carefully-controlled environments" and "ad-hoc mutable environments" is _absolutely_ irreconcilable or merely extremely tricky to navigate.

18:20 <jean-paul[m]> I have Nix-managed environments with`--editable`-like setup for single Python packages, though, so I suppose that could probably generalize to multiple Python packages if you tried hard enough, so I guess it's probably not irreconcilable.

18:29 <komasa> I'm currently sitting next to a few Nix Maintainers in the hackspace, and they didn't sound convinced that this was a good idea

18:33 <jean-paul[m]> The trouble is we're awash in a sea of bad ideas. Is it a worse idea than some tool that carefully juggles a bunch of venvs? I dunno. Maybe, maybe not.

18:33 <jean-paul[m]> Just use tox, right? That always works.

18:34 <jean-paul[m]> I don't have any better suggestions, anyway.

18:37 <komasa> My hope was that maybe there is some trick like a `.pth` file. I think the issue with just that is native code though that has to be built against the interpreter