Neonira
by Neonira

Categories

  • R
  • programmation
  • offensive-programming

Tags

  • R
  • programmation
  • offensive-programming

I submitted my first R packages a few weeks ago, and I believe it’s worth sharing some return on experience, as the CRAN submission and acceptance processes are much more tricky than they appear at first glance.

Package preparation and bundles were achieved using RStudio as IDE to implement code and documentation, and to generate package bundles. I gave attention to get to no error, no warning and no note status prior any CRAN submission.

Package submission process

Package submission process is quite straightforward. It is a fully digital process, achieved online, that is mainly a sequence of three steps.

  1. Upload your R package source bundle
  2. Confirm R package submission
  3. Wait for feedback

R package source bundle uploading

This is done online through submission URL. Fill in your name, email address and select your R package source bundle file to be uploaded. Then submit the form.

R package submission process will send an email to the provided email address.

This first activity of the submission process us quite simple, works fine, and is an easy step to go through.

Confirm R package submission

In this received email, click on the confirmation link to confirm your package submission.

This second activity of the submission process is simple and easy. I met no issue there.

Wait for feedback

This part implies much more hidden work, from the CRAN team. It includes both automated checks and human introspection of your package content.

Feedback might requires some changes in your package content. So, you will have to proceed to the required changes, and replay the submission process. Unfortunately, the number of replays is nearly unpredictable, even when starting from a perfectly correct R package bundle (no error, no warning, no note).

This third activity of the submission process is really difficult. First difficulty is understanding the required changes and their root cause. Second difficulty is proceeding to changes while keeping a high level of integrity in your package pieces. Sometimes a small required change will have a humongous impact on your code or documentation and you may have to transform your approach whenever required.

From my experience, here are the main pitfalls I faced, and if you plan to submit a package to CRAN, get acknowledged to them

R Package structure

Main pitfalls I faced came from two sources: The DESCRIPTION file, and the folder structure of the R package.

DESCRIPTION FILE

This file has been the source of the greatest numbers of changes in my various submissions.

First, version field does not allow for zero led version numbers. Version 0.1.1 is legal, 1.4.23 is also legal, but version 1.04.023 is not. That’s a fact, I was ignorant about.

Second, do not write lastnames entirely using upper cased letters. Capitalizing last name is sufficient and the expected way by CRAN. I stil do not know why, but I had to re-submit a package for that reason.

Third, each time you submit a new release of your R package, you must increase the version number. The CRAN submission process does not allow you to submit several times the same version, and will flag this as an error. Even if you change a single comma, you must change the version number. You may change it, in anyway you like, up the fix number, the minor number, the major number of any combination of them. When taken in the flow of re-submissions, do not forget, to increase the version number, or get ready for a new submission process run.

Fourth, description field requires understand-ability. Readers should understand the purpose and the value of your package by reading it. I had the bad trend to keep it short. I was wrong. Being explicit and being long is right for this field. Use plain English language and be sure to fix all the typos here. Stick to known and regular vocabulary. Do not try to be creative here, as it will very probably bring you in the aftermath of re-submission.

Fifth, I encountered some issues with references. If you need to state a reference to an ISBN or DOI for example, syntax to use is described CRAN policy. Main issue here, is that this documentation lacks inspiring examples. I ended up guessing the syntax, and generally I guessed wrongly. To avoid failure, consider following template, ‘Refer to chapter {X} of {bookName}, {Firstname} {LASTNAME} ({YEAR}, ISBN:{isbn_number})’, and replace everything between braces (including braces themselves}, by the data for your case.

R package folder structure

Main issues came from the use of package folder structure to achieve various goals, that are not related to R package delivery. In particular, I was used to store architecture notes, design notes, implementation notes and maintenance notes in a dedicated folder under inst folder. That’s wrong for at least two reasons. First it exposes information that should not. Second, it increases weight of the R package without adding any value for the package users.

I get back to a KISS strategy and simply removed those folders from the delivery.

R code

Here I faced two issues, that it is worth to know

First, I used assign to global environment to increase package usability, and increase user’s productivity. Wrong, as this is not allowed by the CRAN. Apparently, assigning to global environment is simply forbidden by their acceptance procedure.

Second issue is a very common one, apparently. Writing files in package structure instead of in a temporary folder. Each time, your code, for any reason, requires to write a file, you must use tempdir for the package to pass the submission process. It is not allowed to write in the package structure. This means, that I had to change the function signatures to take into account a target folder parameter, which defaults to tempdir(), and can be changed by user whenever and wherever required.

Note that CRAN acceptance does not mean your package code is valid, nor useful, nor correct. This is the package author’s responsibility to ensure these qualities, not the CRAN team ones.

R documentation

I am very familiar with Writing R extensions and about R documentation scheme. So, I expected to have a smooth submission process, from the documentation point of view.

I faced two main issues, one with .Rdoc files, the other with vignette files.

R documentation issues

My main issue was about the usage I did of \dontrun tag. Human introspection of the code I submitted, brought change requests on this topic.

This tag should be used when the processing time implied by the considered part, exceeds 5 seconds. Change, rince and re-submit.

R vignettes issues

Issue what tied to dynamically generated vignettes, which implied previously presented issue about R code writing into un-allowed places. As this was done, to achieve incremental trace-ability, with no value for end-users, I simply dropped the erroneous vignettes, and kept the useful ones.

Process weirdness

CRAN vs RStudio

You have to know that CRAN does not know about commercial tool RStudio.

Main consequence of this fact is that CRAN checks are not the same as the checks achieved in RStudio, even when passing –as.cran parameter, thus leading each of us, to understand two different verification processes: the one from the CRAN, that involves human inspection, and the one from RStudio tools, that is fully automated.

It took me some time to figure out why I faced no issue using RStudio while CRAN team, in a fair and meaningful way, was requiring changes to my packages. All issues I encountered, except the assignment one, have not been detected by the RStudio tools.

Exchanges with CRAN team members brought some light. Human inspection of the package content aims to ensure understand-ability, to achieve a kind of integrity accros packages, and to filter out most common R code and R documentation issues.

As in any process, they are flaws. Here are the ones I met using CRAN delivery information.

Information email is not always working correctly. From the analysis I can do, considering the 3 packages submission I did, many differences exists. Main issue is emailing, as I did not received the same emails for each of the packages and some emails are simply missing (either not sent, or not delivered).

Compilation process of a package on many platforms brings also some weirdness. I received only confirmation for windows that is the platform I am the less interested in. Got no information for other platforms (DEBIAN, OSX, UBUNTU, …). Finally, I discovered online that some of them were able to process correctly the package, while some others were not. Introspection of log files has shown that issues where operating system context related and under CRAN responsibility (PDF compilations requires some extraneous packages on Solaris, bug on OSX). This has been fixed by CRAN team.

Last and probably the most disturbing was that I often faced some delays in the process because I did not received or did not know what was expected from me by the CRAN as improvements. Distinguishing a CRAN change request from a CRAN comment is not so easy, in received emails. Some issues (especially notes) arise but are not real issues. For example, your last name might not be recognize by the English dictionnary, thus generating a note. Identifying all the changes from a CRAN request is also not as easy as it should. I think that’s rather a matter of presentation - due to email - than a matter of content itself.

A final word

CRAN team does a great and uneasy job. They often face same issues from R package submitters, and so they have to deal with very repetitive tasks. Nevertheless, they keep things simple and oriented toward delivery.

As a result of my return on experience, I have two improvement suggestions that could alleviate several pains.

Primo, I really believe that being able to run the same automated verifications as CRAN, will reduce the forth and back on a package submission. This will contribute to reduce CRAN burden for automated checks while leaving more time for human inspection.

Secundo, for any submitted package, a submission process summary html page will be helpful to share submission process status, progress and next steps, without any ambiguities. Having such a page as submission process summary will avoid loss of time and make clear the leader for the next action. This will also avoid deadlocks where each part awaits the other one, as each believe the other part is currently owning the lead. That will streamline and accelerate the whole process, especially in case of many re-submissions.