Bug Description
When creating a new conda environment from environment.yml using conda env create, the solver hangs indefinitely (observed >47 minutes with no resolution).
conda env create -f environment.yml -n rmg_env18 -v
Channels:
- conda-forge
- rmg
Platform: osx-64
Collecting package metadata (repodata.json): done
Solving environment: |
The trigger for needing a fresh environment was the recent addition of pysidt-rmg to the dependencies.
How To Reproduce
- On macOS 26 (Tahoe) with an Intel (
osx-64) machine, clone RMG-Py at or near commit 3.3.0-605-g17990e8ee
- Run
conda env create -f environment.yml -n rmg_env_new
- Observe that the solver never completes
Expected Behavior
The conda environment should resolve and install within a few minutes, as it does on CI (macOS runners on GitHub Actions).
Workaround
Rather than creating a fresh environment, I activated my previous environment (rmg_env17, Python 3.11.14) and installed the missing package directly:
conda install -c rmg pysidt-rmg
This succeeded, for me, and my environment is working.
But it leaves our installation method for others with a similar setup broken.
Installation Information
- OS: macOS 26.3.1 alpha (
ProductVersionExtra: (a), Darwin 25.3.0) — Intel x86_64 (osx-64 platform)
- Installation method: Source install from
environment.yml via Miniconda3
- conda version: 26.1.1
- solver: libmamba (default)
- conda-libmamba-solver: 25.4.0
- libmamba / libmambapy: 2.1.1
- RMG-Py:
3.3.0-605-g17990e8ee (from git describe --tags on main)
- RMG-database:
3.3.0-22-g2c90c6985
Key packages in the working rmg_env17 environment (Python 3.11.14, osx-64):
| Package |
Version |
Channel |
| python |
3.11.14 |
conda-forge |
| cython |
3.0.12 |
conda-forge |
| numpy |
1.26.4 |
conda-forge |
| scipy |
1.16.0 |
conda-forge |
| rdkit |
2025.09.5 |
conda-forge |
| openbabel |
3.1.1 |
conda-forge |
| matplotlib |
3.10.8 |
conda-forge |
| mopac |
23.1.2 |
conda-forge |
| pydas |
1.0.4 |
rmg |
| pydqed |
1.0.4 |
rmg |
| pysidt-rmg |
1.2.0 |
rmg |
Additional Context
Hypothesis (from Claude Code, not verified): The hang is likely caused by the __osx virtual package being set to 26.3.1 (macOS Tahoe alpha). The libmamba solver may be spending a very long time trying to satisfy package constraints in conda-forge and the rmg channel that were authored before macOS 26 existed — for example, packages that specify an upper bound on __osx or that pin osx_is_app_ext. This would explain why CI on macOS (GitHub Actions runners on Sequoia/15.x) passed without issue. Other Intel Mac users on stable macOS releases may not hit this at all.
I note that updating conda itself didn't help (conda update -n base -c defaults conda upgraded from 26.1.0 to 26.1.1 but 26.3.2 appears to be the newest, which may require a newer Python in the base environment, which would mean reinstalling Miniconda entirely).
Bug Description
When creating a new conda environment from
environment.ymlusingconda env create, the solver hangs indefinitely (observed >47 minutes with no resolution).The trigger for needing a fresh environment was the recent addition of
pysidt-rmgto the dependencies.How To Reproduce
osx-64) machine, clone RMG-Py at or near commit3.3.0-605-g17990e8eeconda env create -f environment.yml -n rmg_env_newExpected Behavior
The conda environment should resolve and install within a few minutes, as it does on CI (macOS runners on GitHub Actions).
Workaround
Rather than creating a fresh environment, I activated my previous environment (
rmg_env17, Python 3.11.14) and installed the missing package directly:This succeeded, for me, and my environment is working.
But it leaves our installation method for others with a similar setup broken.
Installation Information
ProductVersionExtra: (a), Darwin 25.3.0) — Intel x86_64 (osx-64platform)environment.ymlvia Miniconda33.3.0-605-g17990e8ee(fromgit describe --tagsonmain)3.3.0-22-g2c90c6985Key packages in the working
rmg_env17environment (Python 3.11.14,osx-64):Additional Context
Hypothesis (from Claude Code, not verified): The hang is likely caused by the
__osxvirtual package being set to26.3.1(macOS Tahoe alpha). The libmamba solver may be spending a very long time trying to satisfy package constraints in conda-forge and thermgchannel that were authored before macOS 26 existed — for example, packages that specify an upper bound on__osxor that pinosx_is_app_ext. This would explain why CI on macOS (GitHub Actions runners on Sequoia/15.x) passed without issue. Other Intel Mac users on stable macOS releases may not hit this at all.I note that updating conda itself didn't help (
conda update -n base -c defaults condaupgraded from 26.1.0 to 26.1.1 but 26.3.2 appears to be the newest, which may require a newer Python in the base environment, which would mean reinstalling Miniconda entirely).