Frozen / locked conda env files for faster dev setup? (e.g. "fast-start" by OS/arch)

[mullachv]

I'm setting up a dev environment with conda env create -f conda-envs/environment-dev.yml and it's taking over an hour. The repo already has env files per OS (e.g. environment-dev.yml, windows-environment-dev.yml), but they're not "fast ready to install": package versions aren't pinned, so Conda has to resolve dependencies from scratch every time.

For example, the current files look like this (ranges, not pins):

# conda-envs/environment-dev.yml (excerpt)
dependencies:
- arviz>=0.13.0
- numpy>=1.25.0
- pytensor>=2.38.0,<2.39
- zarr>=2.5.0,<3
# ... many more with >=, <, etc.

It would help onboarding if maintainers could occasionally export a frozen (fully pinned) environment from a known-good setup and commit it. To make it obvious these are "ready to install" and to avoid using the wrong file, I’d suggest:

Naming: Something like fast-start or boost-startup in the filename (e.g. environment-dev-fast-start-.yml or environment-dev-boost-startup-.yml), so contributors know they can use it for a quick bootstrap.

Platform: One file per OS and architecture, with a clear suffix so people pick the right one — e.g. conda-style: linux-64, osx-64 (Intel), osx-arm64 (Apple Silicon), win-64. So you’d have e.g. environment-dev-fast-start-linux-64.yml, environment-dev-fast-start-osx-arm64.yml, etc.

New contributors on a matching platform could then run e.g. conda env create -f conda-envs/environment-dev-fast-start-osx-arm64.yml and get a fast, reproducible install without re-solving. Before opening a formal feature issue: would maintainers be open to adding these fast-start / boost-startup style frozen env files (or documenting a recommended way to do it)?

[mullachv]

In my case, the install didn’t just take a long time — it failed. After a long run of “Examining conflict for …”, conda exited with UnsatisfiableError and reported many package conflicts plus system constraints (__osx 10.16). So I never got a working dev env from the current spec. A frozen “fast-start” file for a given platform would at least let people on a matching platform install a known-good set without re-solving, and when the spec doesn’t resolve (e.g. on some macOS/conda setups), having a pre-solved env would be even more valuable.

[ricardoV94]

I suggest you try with mamba. conda is painfull slow

[mullachv]

Yes, and thank you. mamba worked out for me