Conversation
|
That seems like a reasonable change! I only have two concerns:
Is this still a good idea despite that? It would be nice if this cleared out any unneeded dependencies and helped reduced the install size. The other thing I was hoping to do was to upgrade the r-arrow dependency to v22.0 and see if there are any newer versions of terra available for conda + windows yet. Would you have time to do that? No worries if not. |
|
@shreeramsenthi Good points! I think you're right that installation will likely slow given minimally explicit dependencies - will have to check with Katie about the concerns on implicit dependency versions. However, yes, I do have time to update the r-arrow dependency, and will push for you to review. Just a note that last I checked (December), there was still no newer version of terra available for Windows, but will check that as well. |
|
Thanks a bunch! And yeah, I guess that's to be expected on the terra front. It's a longer term goal, but we could start looking at alternatives to Any thoughts on that @BadgerOnABike ? |
|
Just following up - I couldn't upgrade the r-arrow dependency to 22.0.0, as it is incompatible with the current available version of r-terra on Windows (so r-rsyncrosim and r-base are also stuck in older versions). |
|
Good to know! That's a little too bad since there appears to be some noticeable memory improvements in 22.0, but performance is already noticeably better with the newer terra anyway. Just one more reason to try to move away from terra or suggest users not use conda, I guess. |
|
@shreeramsenthi , agreed that the new conda env creation will probably be slower, but as it is right now the conda env can't be built on linux because the explicit dependencies are specific to Windows. The conda env creation is hopefully something that people just have to do once, so I'm thinking that the slow-down in the initial creation is worth the compatibility with linux. We're currently doing this for all the SyncroSim packages that use conda environments as we move more towards running our models on linux. For BurnP3+, it probably makes sense for people to avoid conda anyway, but at least with this minimal yml file they have the option to build on linux if they want. We're also making plans to remove the terra dependency from rsyncrosim because it's causing so many issues. I know BurnP3+ uses terra independently of rsyncrosim though, so that still doesn't really fix the problem with the BurnP3+ conda env unless we decided to also remove terra from BurnP3+... |
|
That makes sense to me! I guess the only other consideration might be reproducibility in case a new version of a package includes a breaking change, but I don't think that's too likely. |
increment conda environment version and create minimalist yml file using versions indicated in script. built in Linux and Windows, both work