Avoid warnings with R hooks when renv version don't match

[lorenzwalthert]

Attempts to solve #1835 as initially outlined in #1835 (comment). To further avoid warnings for differing R versions, we suppress warning messages on environment loading. To not write infrastructure like .Rprofile, we use renv::load() over renv::activate(). For installing additional dependencies, we also added skipping of the startup process. I think after #1878, I might have fixed the biggest bugs I created -.-

[lorenzwalthert]

@asottile can you have a look at this and tell me if it makes sense, also in the light of our discussion in #1835? I used this locally for a few weeks and it worked fine. Maybe we need to figure out licensing for the {renv} files.

[asottile]

yeah seems fine! hmm yeah licensing I didn't think about that -- perhaps we can make a ./licenses directory and add license_files = licenses/renv_LICENCE or whatever (assuming it is permissively licensed, if it isn't this might be a dead end)

[lorenzwalthert]

{renv} is MIT licensed, as pre-commit, so maybe we don’t need to include additional licenses? We can also ask in https://github.com/rstudio/renv how they suggest to handle this, because essentially every project that uses {renv} has this file in version control, and I haven’t seen any special copyright attribution in any project that uses {renv} (although our case might not be 100% the same).

[asottile]

I would rather lean on the side of caution here -- it's not difficult to abide by the license terms -- we just need a copy of their license in our source tree and our distributables

we could even put it in the resources directory so it gets copied for the empty template into renv just to be on the safe side

sorry for the slow response, things have been crazy recently

[lorenzwalthert]

sorry for the slow response, things have been crazy recently

No problem.

we just need a copy of their license in our source tree and our distributables

So licensing for R packages work usually like this: Put the type of license in the DESCRIPTION file as a field (only a few ones are permitted). For {renv}, it's MIT + file LICENCE. Then, there is a separate file called LICENSE, which can have the actual licence or just the copyright holders. {renv} has only the latter, so then I guess we should include a MIT Licence and add the copyright and year from LICENSE.

we could even put it in the resources directory so it gets copied for the empty template into renv just to be on the safe side

Agree

[asottile]

hmmmmm given the contents of that file, I'm not sure they've licensed it for redistribution -- we would probably want to clarify with them what terms that it is allowed to be copied and redistributed.

I don't see an actual license file in their repo beyond that 2-liner (which probably has no legal precedence). I think it would be wrong to take that and stuff it into a MIT license (I think that would be relicensing without approval) -- we should probably start an issue thread on their side unless there's an actual copyable license file!

[lorenzwalthert]

Sure, I can open an issue in their repo so we can figure things out and be on the Safe side. I know the authors and I think they don’t have fundamental objections to what we are doing here.

[kevinushey]

I don't see an actual license file in their repo beyond that 2-liner (which probably has no legal precedence). I think it would be wrong to take that and stuff it into a MIT license (I think that would be relicensing without approval) -- we should probably start an issue thread on their side unless there's an actual copyable license file!

FWIW this is just the canonical way that R packages (at least, those destined for CRAN) declare that they are MIT-licensed (which I agree is a bit strange).

But perhaps I can change this so the, shall we say, CRAN-compatible-declaration-of-MIT-license version of the LICENSE file is used on CRAN, but the "real" version is used on GitHub.

[asottile]

I don't see an actual license file in their repo beyond that 2-liner (which probably has no legal precedence). I think it would be wrong to take that and stuff it into a MIT license (I think that would be relicensing without approval) -- we should probably start an issue thread on their side unless there's an actual copyable license file!

FWIW this is just the canonical way that R packages (at least, those destined for CRAN) declare that they are MIT-licensed (which I agree is a bit strange).

But perhaps I can change this so the, shall we say, CRAN-compatible-declaration-of-MIT-license version of the LICENSE file is used on CRAN, but the "real" version is used on GitHub.

yeah I think it would be sufficient for my uses to have a LICENSE.CRAN and a LICENSE or whatever naming you're comfortable with (this would also make it so github auto-detects your license too)

[lorenzwalthert]

But perhaps I can change this so the, shall we say, CRAN-compatible-declaration-of-MIT-license version of the LICENSE file is used on CRAN, but the "real" version is used on GitHub.

I think at some point, LICENSE used to contain the actual license text. If that is still CRAN compatible, would it maybe also be an option?

[kevinushey]

I think at some point, LICENSE used to contain the actual license text. If that is still CRAN compatible, would it maybe also be an option?

Unfortunately this is explicitly not accepted by CRAN. 😞

[kevinushey]

I've made a change now so that:

• The canonical source code at https://github.com/rstudio/renv includes the standard MIT license file;
• When preparing a tarball for CRAN (via R CMD build), a tarball with the CRAN-compatible declaration of the MIT license is used instead.

Hopefully this is a fair compromise.

[asottile]

@kevinushey thanks a ton! that's super helpful!

[lorenzwalthert]

Great. @asottile how do we progress from here? From the {renv} authors: rstudio/renv#738

These files are (at least currently) also MIT-licensed, and so I think they're fine to use in any project that is licensed in an MIT-compatible way. That is, please feel free to use them with pre-commit.

The intent is definitely to allow users to freely use those files in their projects without needing to worry about copyright / licensing occurs, so we'll evaluate if unlicense might be a good choice.

[asottile]

for now let's take their LICENSE file and also place that into renv directories as a resource

[lorenzwalthert]

Please see c3d186d how I added the licenses:

• pre_commit/resources/empty_template_LICENSE.renv (will be copied into the users renv/ folder for local hooks, where the activate.R resides)
• testing/resources/r_hooks_repo/renv/LICENSE (where an activate.R file resides for testing)

Is that what you had in mind?

[asottile]