Previously, @xiaoruiDong and I observed a strange glitch related to building the RMG website. While updating the main RMG website, we observed that creating the rmg_env specified by RMG-Py and RMG-website (I will call this the "canonical" environment) led to a package compatibility error. Our short-term solution was to clone the dev environment to the main environment. However, this solution cannot be extended indefinitely because it does not solve the core issue (being unable to update environments). It was also confusing to us why we experienced a package compatibility error if the websites were working perfectly fine.
To troubleshoot, here are the steps I took:
Roll back RMG-website to earliest commit with a working environment.yml (if this didn't work, we would have also tried rolling back RMG-Py
Noting that the recent commit involving adding chemprop_solvation to the environment led to the inconsistencies, my first thought was that the package might be conflicting with the chemprop installed in RMG-Py. I tried redoing the installation process by removing chemprop from the RMG-Py environment before installing RMG-website but still ran into the same error. This indicated that the error was instead with another package!
The next logical step is that there is possibly a dependency in chemprop_solvation that conflicts with the other packages. I checked using conda search chemprop_solvation -c fhvermei --info and compared to the canonical environment minus chemprop_solvation. I observed that the canonical environment included all of chemprop_solvation's dependencies except for pytorch.
Simultaneously, I compared the "working" website environment with chemprop_solvation to the canonical environment minus chemprop_solvation. I observed that there were several package differences, but most importantly, the working website had pytorch in the environment.
To confirm pytorch was the source of the error, I tried installing pytorch using the default conda channels in the canonical environment minus chemprop_solvation and ran into... the package incompatibility error. Finally, properly installing pytorch fixed the issue, so I tried a fresh install by modifying environment.yml and the error was fixed!
Short description
Adds missing dependencies that previously caused the conda environment to fail to properly build due to recent
chemprop_solvationinstall. more info on the incompatibility can be found in this thread.Long description
Previously, @xiaoruiDong and I observed a strange glitch related to building the RMG website. While updating the main RMG website, we observed that creating the
rmg_envspecified byRMG-PyandRMG-website(I will call this the "canonical" environment) led to a package compatibility error. Our short-term solution was to clone the dev environment to the main environment. However, this solution cannot be extended indefinitely because it does not solve the core issue (being unable to update environments). It was also confusing to us why we experienced a package compatibility error if the websites were working perfectly fine.To troubleshoot, here are the steps I took:
RMG-websiteto earliest commit with a workingenvironment.yml(if this didn't work, we would have also tried rolling backRMG-Pychemprop_solvationto the environment led to the inconsistencies, my first thought was that the package might be conflicting with thechempropinstalled inRMG-Py. I tried redoing the installation process by removingchempropfrom the RMG-Py environment before installing RMG-website but still ran into the same error. This indicated that the error was instead with another package!chemprop_solvationthat conflicts with the other packages. I checked usingconda search chemprop_solvation -c fhvermei --infoand compared to the canonical environment minuschemprop_solvation. I observed that the canonical environment included all ofchemprop_solvation's dependencies except forpytorch.chemprop_solvationto the canonical environment minuschemprop_solvation. I observed that there were several package differences, but most importantly, the working website hadpytorchin the environment.pytorchwas the source of the error, I tried installingpytorchusing the default conda channels in the canonical environment minuschemprop_solvationand ran into... the package incompatibility error. Finally, properly installingpytorchfixed the issue, so I tried a fresh install by modifyingenvironment.ymland the error was fixed!