Environment build fails with S4 => S4SXP error

[anisotropi4]

When trying to install an R environment on a new machine the build for the rsgeo package fails with the following error:

   Compiling geo v0.28.0 (https://github.com/georust/geo#ff0eccbc)
error[E0425]: cannot find value `S4SXP` in this scope
   --> /tmp/RtmpBedzSp/R.INSTALL18cf22abbf157/rsgeo/src/vendor/extendr-api/src/lib.rs:574:15
    |
574 |         S4 => S4SXP,
    |               ^^^^^ not found in this scope

Any thoughts about how to fix this?

[JosiahParry]

Looks like an issue with some libRsys bindings. Instead of compiling from source you should install from a precompiled binary.

Please install from https://josiahparry.r-universe.dev/rsgeo

[Robinlovelace]

Woa, that's a pretty deeply embedded issue. Could it be hardware related, e.g. you're running on an old CPU that is not compatible with this instruction set Will? Curious to know if the work-around works..

[JosiahParry]

This is only present in the most recent version of the package. So assuming that you’re using an renv.lock file or something that keeps track of the package version this probably shouldn’t happen. I don’t think the most recent version is what is used for reproducibility. Additionally this an issue that comes from compiling the source. Ideally you would be using a precompiled binary so any issues related to compilation would be avoided.

…

On Mon, Jul 22, 2024 at 18:18 Robin Lovelace ***@***.***> wrote: Woa, that's a pretty deeply embedded issue. Could it be hardware related, e.g. you're running on an old CPU that is not compatible with this instruction set Will? Curious to know if the work-around works.. — Reply to this email directly, view it on GitHub <#112 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADHIKLDKCXZDFH3AMDIMBCDZNWAMRAVCNFSM6AAAAABLJDTPP6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENBTHEYDMNBUGI> . You are receiving this because you commented.Message ID: ***@***.***>

[Robinlovelace]

Is there a way to tell R where to install it from in the DESCRIPTION?

https://github.com/nptscot/networkmerge/blob/main/DESCRIPTION

[anisotropi4]

This is broken when I try this on a clean install.

As clean install is by running the commands from the code/install.R I would ask that this is updated to reflect the dependencies:

 $ Rscript code/install.R

This is also the case when running install.packages('rsgeo', repos = c('https://josiahparry.r-universe.dev', 'https://cloud.r-project.org')) as suggested but this appears to still gives:

> install.packages('rsgeo', repos = c('https://josiahparry.r-universe.dev', 'https://cloud.r-project.org'))
Installing package into ‘~/R/x86_64-pc-linux-gnu-library/4.1’
(as ‘lib’ is unspecified)
trying URL 'https://josiahparry.r-universe.dev/src/contrib/rsgeo_0.1.7.tar.gz'
Content type 'application/x-gzip' length 9038197 bytes (8.6 MB)
==================================================
downloaded 8.6 MB

* installing *source* package ‘rsgeo’ ...
...
   Compiling log v0.4.22
   Compiling geo v0.28.0 (https://github.com/georust/geo#ff0eccbc)
error[E0425]: cannot find value `S4SXP` in this scope
   --> /tmp/RtmpCiyt2m/R.INSTALL1b7b633c2cb16/rsgeo/src/vendor/extendr-api/src/lib.rs:574:15
    |
574 |         S4 => S4SXP,
    |               ^^^^^ not found in this scope

For more information about this error, try `rustc --explain E0425`.
error: could not compile `extendr-api` (lib) due to 1 previous error

Further suggestions welcome.

[anisotropi4]

Update: If I run install.packages('rsgeo', repos = c('https://josiahparry.r-universe.dev', 'https://cloud.r-project.org')) on an existing install this works.

The issue appears to be how to get this to work on a clean install.

[Robinlovelace]

Have changed how we install it @anisotropi4 hoping that fixes it for future users. We can re-open if not properly fixed and any further changes welcome.

[anisotropi4]

This has failed again on my new environment. I guess what I would like to see, outwith using Docker, is an environment build guide that allows the system to run with dependencies.

For example:

1. Install R on linux using these instructions
2. Run $ Rscript code/install.R to install and build R dependencies
3. Install quarto using these instructions

...and so on...

As it stands I'm stuck.

[Robinlovelace]

I think Pixi was kind of designed to solve this kind of thing.

[Robinlovelace]

Related thread, mentioning R and Python support: posit-dev/positron#3724

[Robinlovelace]

I guess what I would like to see, outwith using Docker, is an environment build guide that allows the system to run with dependencies.

For example:

1. Install [R on linux using these instructions](https://cran.r-project.org/bin/linux/ubuntu/fullREADME.html)

2. Run `$ Rscript code/install.R` to install and build R dependencies

3. Install [quarto using these instructions](https://quarto.org/docs/get-started/)

Agree that should be in the README. Do those things not currently work, even after the 'fix' in 8a20325 ?

[JosiahParry]

If you're getting prompted to install from source say "no". I cannot repro this issue. The r-universe has linux binaries provided. It looks like you're running an old version of R (4.1) which may be the root of the issue.

If you want to create an issue in rsgeo with a reproducible example that is very welcomed.

Thanks!

[anisotropi4]

Please advice what I need to do about this, noting that I have followed recommended R install instructions for a Debian/Ubuntu based linux install.

If the intent is to deviate from the standard LTS version of R, I would ask that this is made clear.

[anisotropi4]

The issue is the R version. For some reason the clean install ignored the additional repository, reverted to the based LTS Debian package, and installed R 4.1.1.

A couple of things with this,

1. While the default LTS Debian package looks like it is three years out of date, I think we need to make it clear that only version 4.4 is supported.
3. The final step of the Linux install instructions on cran don't seem to work. In particular the following fails to update /etc/apt/source.list.d/additional-repositories.list:

$ sudo` add-apt-repository "deb https://cloud.r-project.org/bin/linux/ubuntu $(lsb_release -cs)-cran40/"  

Which is why the old version of R was installed.

4. The full Linux install instructions on cran also don't seem to work. I ended up creating the following repository entry, which does:

$ cat /etc/apt/sources.list.d/additional-repositories.list 
deb https://cran.ma.imperial.ac.uk/bin/linux/ubuntu jammy-cran40/

Based on this, I can now render the paper, but without many warm fuzzies from for Linux users.