Docker: use R installation manager

[jeroen]

Switch the host R installation to use rig from @gaborcsardi.

This makes it much easier to pin the host installation to match the webr version (4.3.0 currently). Also it automatically sets up pak and PPM binaries, so I think this will much simplify management.

[georgestagg]

Nice! This looks like a great idea to me, thanks so much.

I'm just doing some local testing in Docker before I approve and merge.

[jeroen]

OK cool! If rig does a little bit too much for your taste you can also grab the installer directly from:

https://cdn.posit.co/r/ubuntu-2204/pkgs/r-4.3.0_1_amd64.deb

But then you still have to manually set the p3m repositories.

[georgestagg]

If rig does a little bit too much for your taste you can also grab the installer directly from [...]

Hmmm. I'll take a look at how much rig adds to the image size vs. pulling the .deb directly and make a decision once I know for sure.

Edit: Image size before this PR: 2.94GB. After: 4.69GB. However, using the .deb directly doesn't seem to reduce the image size much, so I don't think it's rig adding the extra size.

[gaborcsardi]

Edit: Image size before this PR: 2.94GB. After: 4.69GB. However, using the .deb directly doesn't seem to reduce the image size much, so I don't think it's rig adding the extra size.

The rstudio/r-builds R builds have a lot of dependencies that are not strictly required, e.g. compilers. The rig binary itself is less than 10MB AFAIR, and has no dependencies.

You probably also want to delete the downloaded deb packages after rig add.

[georgestagg]

The rig binary itself is less than 10MB AFAIR, and has no dependencies.

Brill, thanks for confirming. 10MB is a bargain for the management tools rig itself provides.

So, the cost is in using the rstudio R builds. I guess it's not so bad, and probably worth the extra GB (if you're already downloading 3GB in any case) to ensure we have a confirmed working native R of the right version.

[jeroen]

Hmm are you sure those numbers are correct? I would be surprised if the posit build of R has many more dependencies than the default ubuntu builds.

I have the version from the above PR in my own ghcr.io, and the sizes* are uncompressed:

# docker image ls ghcr.io/r-wasm/webr:main
REPOSITORY            TAG       IMAGE ID       CREATED      SIZE
ghcr.io/r-wasm/webr   main      3a13fe3601ba   5 days ago   2.94GB

# docker image ls ghcr.io/jeroen/webr:main
REPOSITORY            TAG       IMAGE ID       CREATED       SIZE
ghcr.io/jeroen/webr   main      8456fd927511   3 hours ago   3.17GB

And compressed:

dockersize() { docker manifest inspect -v "$1" | jq -c 'if type == "array" then .[] else . end' |  jq -r '[ ( .Descriptor.platform | [ .os, .architecture, .variant, ."os.version" ] | del(..|nulls) | join("/") ), ( [ .SchemaV2Manifest.layers[].size ] | add ) ] | join(" ")' | numfmt --to iec --format '%.2f' --field 2 | column -t ; }

## jeroen@dev:~$ dockersize ghcr.io/r-wasm/webr:main
linux/amd64  1.07G

##jeroen@dev:~$ dockersize ghcr.io/jeroen/webr:main
linux/amd64  1.18G

*scripts from here

[georgestagg]

Interesting! Here is how things look on my end:

Where the untagged images are with and without rig. Perhaps the difference is because I built those locally on arm macOS, rather than through GHA/Linux? I don't really see why that should be the case, though.

In any case, I'm happy to continue with using rig and RStudio's R builds. Other than a few minor suggestions, let's merge and see where things lie after CI catches up.

[jeroen]

Thanks. I'm reverting the last commit that added apt-cache clean because it did not reduce the image size at all.

[jeroen]

Awesome. Going to trigger 40k rebuilds with this R-4.3.0 so that we don't get the warning: this package was built with R-4.3.2 when loading the packages in webr.

[georgestagg]

Wow, that's a lot of packages! Good luck!

[gaborcsardi]

Just out of curiosity, why are you not using R 4.3.2?

[jeroen]

Afaik, webr has not been ported yet to 4.3.2. Op wo 8 nov. 2023 15:50 schreef Gábor Csárdi ***@***.***>:

…

Just out of curiosity, why are you not using R 4.3.2? — Reply to this email directly, view it on GitHub <#312 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AABUZ7Z3XZXF6K5TCY3H2HLYDOL2FAVCNFSM6AAAAAA65TB6LCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQMBSGA2DAOBYGI> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[georgestagg]

The current release of webR uses R 4.3.0 as the base source. I think using R 4.3.2 to build Wasm packages does work, but it shows a warning message when loading the resulting package in webR.

---

As for the reason webR still uses R 4.3.0, it's because webR patches the source code to reimplement some of R's features using web browser APIs, and also makes some other changes for compatibility with the WebAssembly and Emscripten environment. From experience, when I update the version of the R source used the patches don't apply cleanly and need to be reworked. This has been a non-trivial amount of work in the past, so I don't focus too much on following every R patch release.

Perhaps in the future, some of the patches can be upstreamed to R core to make the Wasm build process simpler.