Closed remram44 closed 5 years ago
Copying response here as well:
I do not remember any discussion about this before.
I remember only that you said that the version of R in 17.10 is good one and you can use it.
So, things you can do here instead of requiring custom Dockerfile, is that you use post-install and post-devel hooks in setup.py
to do stuff. That is what some other performers are doing. So in there you could configure R.
Please do not install R though in this way. If there is an Ubuntu package, please just use it and find a way how that can work for you. I am not sure why official package would not work correctly. This is pretty strange.
Also, do you have to install packages through R or are there Ubuntu packages corresponding to those R packages?
Finally, configuring stuff for current user (~/.Rprofile
) is bound to fail because stuff can be run as different users. You have to configure things globally. In any case, you should not have to configure a mirror at runtime or download anything at runtime.
Reposting from GitLab:
I originally discussed this with Sujen when debugging our own pipelines.
I think that we can get Ubuntu packages given that library(XXX) from R still works. The issue is that R-igraph either isn't compiled properly or not downloaded, so library(igraph) is failing at runtime.
The real fix will come when we just have no R dependencies (which began just prior to the workshop, and is set to finish before the next evaluation).
R dependency was completely removed.
The primitives appear extremely broken in the Docker image. They ask to select a CRAN mirror on the terminal when I run
d3m.primitives.jhu_primitives.SeededGraphMatching
. Not only does that freeze the process, but if stdin is closed, it will enter in a loop asking over and over, spewing hundreds of megabytes per second in the log.Downloading code at runtime is also definitely against the rules.