Very cool! Are you planning for there to be a corresponding R package that exposes the high level commands? The popularity of the usethis package really showed the power of keeping people within the R interpreter rather than going back and forth with the terminal. This is so important for a language like R that has so many users without much CS training
You might want to consider writing a plugin for R with Mise en Place https://mise.jdx.dev/core-tools.html This would extend your reach and might take some of the heavy lifting out of the project. (At least for the runtime portion. I don't think it will help with package management.)
Mise is pretty new, and it's userbase is tiny (afaik), so the overlap with it and anything is tiny. But I've enjoyed it as a replacement for ruby/node/java/python version managers, and I think it's a solid, thoughtful piece of kit. I think it targets curious, multilingual hackers who I imagine would be the kind of people to try out R to "kick the tires" just for fun (I imagine Elixer, Erlang, and Zig are in there for the same reason...surprised not to see Julia). It's also the case that mise is already doing all the heavy lifting of documentation, website, installation, etc so might as well not reinvent the wheel (such projects always have far more scope than you think, in my experience!) It could free you to making the package manager that much better (a very hard problem in itself).
I'm not affiliated, btw, just a happy user. Shout out to DHH for introducing it (to me) as part of Omakub.
We actually do use a bit of their code for the linking phase, which they seem to have taken from Cargo. For the rest, Python and R are way too different in how they handle packages to allow sharing code.
I'm curious — does rv support or plan to support per-project isolation of system-level dependencies (e.g., gfortran, libxml2, etc.) like what renv sometimes indirectly requires users to manage outside R? If not, do you have recommendations for managing these in a reproducible way alongside rv?
It's not planned for rv, this is whole other can of worms. Something like nix/docker should work but I'm not working on that part myself so I can't comment.
My biggest issue with R package management is version pinning. If I specify an older version of a package, R will fetch the latest versions of all its dependencies, regardless if they're compatible or not, which leads to manually chasing down and re-installing specific versions of dependencies and sub-dependencies one-by-one.
Microsoft's CRAN time machine helped solved this, but I think they've recently shut it down and I don't really trust Posit to not have a version behind a paywall.
Posit offers something similar to Microsoft’s CRAN Time Machine, but it works not only for CRAN, but also for Bioconductor and PyPI. You can add a date to the Public Posit Package Manager URL to access a snapshot of all packages from that day.
You're hitting one of my (and many people's) main issue with the R package distribution system. In CRAN, only one package version is available at a time, which makes things like version pinning quite difficult. Now the benefit of that is that CRAN guarantees all packages will work together at any moment in time, but then trying to reach back into the Archive breaks that guarantee.
What the CRAN time machine (and now Posit Package Manager) does is take that compatibility guarantee, and freeze it so you have access to all the same, compatible, packages at any moment in time.
While I personally do use PPM fairly extensively, I do understand the paywall concern for long-term reproducibility so `rv` can help you here, with a bit of manual massaging. I'd recommend setting the repositories section of the config file to be a snapshot date in PPM that contains the package version(s) you're interested in and then installing using that repository (taking the benefit of that CRAN guarantee), then in both the config file and `rv.lock`, replace all the references to the PPM repo with your preferred CRAN mirror. This will allow you to resolve to compatible package versions, but then for your POSIT concern, will still be able to reproduce using the CRAN archive.
By default, rv will create a library folder in the same folder as your rproject.toml and there's rv activate/deactivate to add it to your loaded libs. Pretty much the same stuff as a virtualenv in Python.
Ultimately, you're right that `rv` and `renv` get you to the same spot, both create reproducible, isolated projects. `renv` has a few issues that we often hit that lead to `rv`.
`renv` is an iterative process of installing some packages, then snapshotting your project state, and then trying to reproduce. The time between the installation and snapshot can often lose information (think `install.packages("my_pkg", repos = "https://my-repo.com")`, your repo source is lost by time the snapshot occurs). You can also install incompatible versions over-time.
rv solves both of these problems because it will lock the source at the time of installation. Additionally, because it is declarative, we are able to resolve the full dependency tree before installing packages to ensure everything will be compatible.
While I am a big proponent of using rv, if renv fits your needs, then switching to rv may not be worth it. For our organization, we did have multiple issues with renv, so created a replacement for it that we hope others in the community will find useful to address their needs.
I had similar issues in the past. Setting up renv, everything seems good, but after working in a project for a few weeks and installing packages, renv constantly complains about the library being out-of-sync and resolving these complaints took way longer than new ones came around. I think renv has good intentions, but there are just too many edge cases (Bioconductor, installing an experimental package from Github, weird package dependencies etc.), that it always failed me in a real-world scenario.
Very cool! Are you planning for there to be a corresponding R package that exposes the high level commands? The popularity of the usethis package really showed the power of keeping people within the R interpreter rather than going back and forth with the terminal. This is so important for a language like R that has so many users without much CS training
You might want to consider writing a plugin for R with Mise en Place https://mise.jdx.dev/core-tools.html This would extend your reach and might take some of the heavy lifting out of the project. (At least for the runtime portion. I don't think it will help with package management.)
I could be wrong but I feel like the overlap between mise and R users is likely very tiny
Mise is pretty new, and it's userbase is tiny (afaik), so the overlap with it and anything is tiny. But I've enjoyed it as a replacement for ruby/node/java/python version managers, and I think it's a solid, thoughtful piece of kit. I think it targets curious, multilingual hackers who I imagine would be the kind of people to try out R to "kick the tires" just for fun (I imagine Elixer, Erlang, and Zig are in there for the same reason...surprised not to see Julia). It's also the case that mise is already doing all the heavy lifting of documentation, website, installation, etc so might as well not reinvent the wheel (such projects always have far more scope than you think, in my experience!) It could free you to making the package manager that much better (a very hard problem in itself).
I'm not affiliated, btw, just a happy user. Shout out to DHH for introducing it (to me) as part of Omakub.
Tend to agree. Majority of users are leaving that icky computer stuff to RStudio and have no idea what happens behind the scenes.
Maybe some code could be shared with the `uv`[1] to avoid re-implementing same things.
[1] https://github.com/astral-sh/uv
We actually do use a bit of their code for the linking phase, which they seem to have taken from Cargo. For the rest, Python and R are way too different in how they handle packages to allow sharing code.
I'm curious — does rv support or plan to support per-project isolation of system-level dependencies (e.g., gfortran, libxml2, etc.) like what renv sometimes indirectly requires users to manage outside R? If not, do you have recommendations for managing these in a reproducible way alongside rv?
It's not planned for rv, this is whole other can of worms. Something like nix/docker should work but I'm not working on that part myself so I can't comment.
I'm curious how your team ended up doing this. We settled on Nix with flakes after some pain with Docker / RStudio Server.
My biggest issue with R package management is version pinning. If I specify an older version of a package, R will fetch the latest versions of all its dependencies, regardless if they're compatible or not, which leads to manually chasing down and re-installing specific versions of dependencies and sub-dependencies one-by-one.
Microsoft's CRAN time machine helped solved this, but I think they've recently shut it down and I don't really trust Posit to not have a version behind a paywall.
Posit offers something similar to Microsoft’s CRAN Time Machine, but it works not only for CRAN, but also for Bioconductor and PyPI. You can add a date to the Public Posit Package Manager URL to access a snapshot of all packages from that day.
For example: https://packagemanager.posit.co/cran/2025-03-02
You can browse available snapshot dates here: https://packagemanager.posit.co/client/#/repos/cran/setup?sn...
This also works for PyPI and Python packages: https://packagemanager.posit.co/pypi/2025-03-04/simple
You're hitting one of my (and many people's) main issue with the R package distribution system. In CRAN, only one package version is available at a time, which makes things like version pinning quite difficult. Now the benefit of that is that CRAN guarantees all packages will work together at any moment in time, but then trying to reach back into the Archive breaks that guarantee.
What the CRAN time machine (and now Posit Package Manager) does is take that compatibility guarantee, and freeze it so you have access to all the same, compatible, packages at any moment in time.
While I personally do use PPM fairly extensively, I do understand the paywall concern for long-term reproducibility so `rv` can help you here, with a bit of manual massaging. I'd recommend setting the repositories section of the config file to be a snapshot date in PPM that contains the package version(s) you're interested in and then installing using that repository (taking the benefit of that CRAN guarantee), then in both the config file and `rv.lock`, replace all the references to the PPM repo with your preferred CRAN mirror. This will allow you to resolve to compatible package versions, but then for your POSIT concern, will still be able to reproduce using the CRAN archive.
Might be worth looking at groundhog, if you want a 'time machine' less likely to have a commercial motive
is using Rust to rewrite existing package managers a new trendy thing
feels we eagerly need cv -> C/C++ package manager
Is it possible to override transitive dependencies?
Yes, if you list it in the rproject.toml from a specific repo/url/git etc it will use that
Can it be used for installing from bioconductor?
Not yet but it is on the radar.
Can this be used to effectively create R environments? I’m desperate for such a solution.
{renv} is pretty solid, I've been using it in production for years now and have no complaints.
I used pixi for that. It uses Conda ecosystem but you get proper lockfiles and great native binary package support.
By default, rv will create a library folder in the same folder as your rproject.toml and there's rv activate/deactivate to add it to your loaded libs. Pretty much the same stuff as a virtualenv in Python.
Doesn’t renv do that? What need does renv not meet for you?
Ultimately, you're right that `rv` and `renv` get you to the same spot, both create reproducible, isolated projects. `renv` has a few issues that we often hit that lead to `rv`.
`renv` is an iterative process of installing some packages, then snapshotting your project state, and then trying to reproduce. The time between the installation and snapshot can often lose information (think `install.packages("my_pkg", repos = "https://my-repo.com")`, your repo source is lost by time the snapshot occurs). You can also install incompatible versions over-time.
rv solves both of these problems because it will lock the source at the time of installation. Additionally, because it is declarative, we are able to resolve the full dependency tree before installing packages to ensure everything will be compatible.
While I am a big proponent of using rv, if renv fits your needs, then switching to rv may not be worth it. For our organization, we did have multiple issues with renv, so created a replacement for it that we hope others in the community will find useful to address their needs.
I had similar issues in the past. Setting up renv, everything seems good, but after working in a project for a few weeks and installing packages, renv constantly complains about the library being out-of-sync and resolving these complaints took way longer than new ones came around. I think renv has good intentions, but there are just too many edge cases (Bioconductor, installing an experimental package from Github, weird package dependencies etc.), that it always failed me in a real-world scenario.
You can do that with Nix or Guix.