Docker-stacks: R reshape2 lib fails to load (stringi.so / libicui18n.so.64 not found)

[emthomas]

Hi! Thanks for using the Jupyter Docker Stacks.

Please review the following guidance about how to ask questions, contribute changes, or report bugs in the Docker images maintained here.

1. If you have a general question about using these Docker images or with Jupyter + Docker in general, please post your question in the Jupyter Discourse Q&A section so that it can be seen, discussed, and hopefully answered by the community.

2. If you are looking to contribute to the images, please see the Contributor's Guide in the documentation for our preferred processes.

3. If you are reporting an issue with one of the existing images, please answer the questions below to help us troubleshoot the problem. Please be as thorough as possible.

What docker image you are using?

jupyter/all-spark-notebook:2ce7c06a61a1

What complete docker command do you run to launch the container (omitting sensitive values)?

docker run -it jupyter/all-spark-notebook:2ce7c06a61a1 /bin/bash

What steps do you take once the container is running to reproduce the issue?

jovyan@484398e1fe17:~$ R
> library(reshape2)

What do you expect to happen?

Package to load without error

What actually happens?

Throws the following error:

Error: package or namespace load failed for ‘reshape2’ in dyn.load(file, DLLpath = DLLpath, ...):
 unable to load shared object '/opt/conda/lib/R/library/stringi/libs/stringi.so':
  libicui18n.so.64: cannot open shared object file: No such file or directory

[emthomas]

Similar to #497. Hope it will be a quick fix.

[emthomas]

Workaround is to install stringi. conda install -c r r-stringi

[playermanny2]

Also having this issue, @emthomas workaround solves it...any ideas here on a longterm fix@parente @minrk

[parente]

If this is still an issue, the upstream conda-forge r-reshape2 package definition should be updated to require r-stringi if it is indeed a dependency. Then the stacks here should pick up that new r-reshape2 package version for the fix.

[matthewbegun]

Not sure if I should put this here or start a new issue - I get the same error trying to load tidyverse. Happy to open a new issue or raise upstream, but it does seem unlikely they would have their dependencies wrong?

edit - Also in my case the package is listed as installed from the conda-forge channel, re-installing the same version number from the r channel fixes the problem.

[terzieff]

Also happens for me, probably linked to a release/update from the last ~7 days, since it was okay last Friday. Using jupyter/r-notebook:latest image.

Error: package or namespace load failed for ‘tidyverse’ in dyn.load(file, DLLpath = DLLpath, ...):
 unable to load shared object '/opt/conda/lib/R/library/stringi/libs/stringi.so':
  libicui18n.so.64: cannot open shared object file: No such file or directory

[parente]

In the conda-forge recipes:

• r-reshape2 lists r-stringr as a dependency: https://github.com/conda-forge/r-reshape2-feedstock/blob/master/recipe/meta.yaml#L45
• r-stringr lists r-stringi as a dependency: https://github.com/conda-forge/r-stringr-feedstock/blob/master/recipe/meta.yaml#L39
• r-stringi package is currently broken, missing the same .so listed above: https://github.com/conda-forge/r-stringi-feedstock/issues/20

Fixing the issue in the third bullet should resolve the problem in the images here. Note that the issue title says OSX, but others have reported the same problem on Ubuntu in the comments.

[dmpe]

Alternatively, much more prefered, would be to include stringi in all images. That would also help people like me behind firewall :)

[romainx]

Hi,

The R stack has been updated recently in the PR #1000 and the library can be loaded.
My guess is that there was an incompatibility between the latest system lib and the previous R version 3.6.1, now the 3.6.2 is deployed.

> library(reshape2)
> packageVersion("reshape2")
[1] ‘1.4.3’

So could you please give it a try?

However regarding the reshape2 case it is now deprecated and you should switch to tidyr (that is also available in the image).

reshape2 is retired: only changes necessary to keep it on CRAN will be made. We recommend using tidyr instead.
-- https://github.com/hadley/reshape

The most recent version available 1.4.3 last from 2017.

Side note

We cannot remove the dependency since it is brought by ggplot2.
However, we should explicitly install the tidyverse package since I think the tidyr dependency is brought by sparklyr.

[romainx]

Hello,

I've no feedback on this issue. Moreover now tests are in place ensuring that all the installed packages can be imported. For example this is an extract of the build logs showing the result of the import test for
reshape2.

2020-03-17 19:49:47 [    INFO] Grabing the list of specifications ... (helpers.py:76)
-------------------------------- live log call ---------------------------------
2020-03-17 19:49:48 [    INFO] Testing the import of packages ... (test_packages.py:133)
2020-03-17 19:49:48 [    INFO] Trying to import reshape2 (test_packages.py:135)
[...]
PASSED                                                                   [100%]

So I'm closing this issue, feel free to open a new one in case of problem.

Best.