Think it would be possible to get an official R stack?

[cboettig]

e.g. see http://blog.docker.com/2014/09/docker-hub-official-repos-announcing-language-stacks/

[eddelbuettel]

I was thinking about that too. It is appealing but puts more of a burden on to actually carry on with this.

Another 'maybe review in a week or two or three...' question?

[psftw]

Hello. I don't work with R code, but it's clear to me that you guys have created some good Dockerfiles for R, especially with all of the recent hacking (sorry, Docker isn't perfect yet, and we'll get better about documenting the edges!). I would love to see an official R language stack image based on rocker/r-base if you are interested. This would require some up-front work and then a follow-on process of bumping changes via PRs to https://github.com/docker-library/official-images.

Looking over the r-base Dockerfile there are just a couple of things which come to mind:

• You should inherit from debian:jessie to be more explicit
• I'm not sure what's going on with "littler" being installed from APT and then again from a deb? Downloading over http should at least use some kind of checksum to verify the contents

[eddelbuettel]

@psftw You rock. Thanks for the note. @cboettig and have discussed attempting to be 'official', we just wanted to mature this a little more. So really appreciate you starting the dialogue.

Re your comments:

• Debian names vs flavours: I've been a Debian developer for many years and I still like 'testing' better because we do not need to roll the names. We use testing as "sid plus 10 days and other maturation", ie we are explicitly on a rolling release, rather than "we anticipate the next Debian stable". But we can adjust if you really think it matters. Once jessie is out we roll to the next.
• The littler issue :-) I am both upstream and the Debian maintainer for it, but the current release version is not yet in testing as I just released 0.2.1 a few days ago. We needed three of these updated scripts from 0.2.1, so instead of building from source I simply prepared a deb locally. Per its status page it will be in testing in five days.

[psftw]

Great. I'm not opposed to using debian:testing, though I see you've since switched to jessie. I also see the "littler" stuff was also updated as expected. I'll try to get back to you soon in the other thread (#50). 👍 on the contributing section that was added. It isn't clear to me why you are avoiding PRs to master directly -- do you have a custom build integration with github or something?

[eddelbuettel]

@psftw : Thanks for getting back to us. Reopening the dialogue was on my todo list, but I was at a workshop most of last weekend, and this thing called 'work' gets in the way too.

You raise a few distinct issues. Allow me to disentangle them a little:

1. debian:testing vs debian: jessie. I prefer the former. If you are indifferent I would like to revert. Ok?

2. littler: yes, and as expected. We were just waiting for 0.2.1 to hit testing.

3. I do not know what you are trying to say w.r.t. to Non-root users, shared volumes, & RStudio #50. Maybe I will understand once you posted there.

4. Can you reword "It isn't clear to me why you are avoiding PRs to master directly" a little, or provide context? Where shall I have said anything against PRs?

5. " do you have a custom build integration with github or something?" : We commit Dockerfiles which Docker turns into containers via hub.docker.com. I presume you know that, so again I must be missing something.

Dirk

[eddelbuettel]

@psftw Just chatted with @cboettig -- if you are confused about master vs sandbox branches for Rocker: Each commit used to trigger half a dozen builds, and we sometimes committed multiple times a day. But going to sandbox first we are able to throttle and control that. We are approaching calmer times to we may get back to direct commits (and hence PRs) to master. Nothing sinister here.

[psftw]

Go ahead and switch back to debian:testing if that's what makes sense for you. I haven't read through #50 yet, but I want to make sure I answer any outstanding questions you may have about permissions, etc, hence why I mentioned it. If your reasoning for using sandbox vs master is purely to reduce load on DockerHub build systems, then you are being overly considerate. Sending pull requests directly to master would be a couple of steps easier for contributors who may not be Git experts. It's up to you, I'm agnostic about it.

[eddelbuettel]

@psftw : Please have another look at the current state of Dockerfiles for Rocker, and see if there there are any other outstanding issues.

[eddelbuettel]

Paging @psftw : Any interest? Or comments?

[psftw]

I'm happy with the Dockerfile for r-base. Let's continue in the official-images PR.

[cboettig]

Thanks @psftw et al for your help with this. the PR for this issue, docker-library/official-images#310 has now been merged so I think we can close this issue out for now.

I guess we may consider proposing an r-devel or rstudio flavor for the official image down the road under a new issue.