In ComputationalProteomicsUnit/cputools: Various helper functions

suppressPackageStartupMessages(library("BiocStyle"))

This is work in progress.

Introduction

This vignette summarises some good practice for R programming and package development. I tend to follow these myself and advise my collaborators to use them when we work on projects together.

library("cputools")

See also the CPU wiki for a non-exhaustive list.

GitHub user

Several function in the cputools package require a GitHub user name to generate correct links to the software's page using makeGithubUrl, for example. The user name, when not provided explicitly, is fetched from the session options with options("GitHubUserName"), which can be set in your .Rprofile using

options(GitHubUserName = "ComputationalProteomicsUnit")

If set, one can just call this function above with

makeGithubUrl("cputools")

Otherwise, the user can be specified explicitly with

makeGithubUrl("cputools", user = "ComputationalProteomicsUnit")

Credibility

Demonstrate good programming skills by writing clean and maintainable code, provide good documentation and follow good community practice. I would recommend to follow the Bioconductor guidelines for package development and submission, in particular good documentation, vignettes and unit testing. I would also add continous integration (see below) to this.

An essential aspect is to officially release the package. This can be done by submitting it to a public package repository such as CRAN or Biocondcutor. For any package related to high throughput biology, acceptance to Bioconductor is becoming, in my opinion, a requirement. It doesn't mean that there aren't any good biology-related packages that are not in Bioconductor, or that all packages in Bioconductor are first class packages, but any serious package developer should ideally develope a package that matches Bioconductor standards and review. A package that is only available on a personal web page might not be reagarded positively in a first instance; a package that has passed a well defined and formal review process (and has been looked at by an experience R developer, which is probably more that most of the bioinformatics tools available out there), shows that the authors cared enough about his work to polish it to Bioconductor standards.

If the package is not to be submitted to Bioconductor (in particular if it doesn't fall into the project's remit), then the developer should implement their own continous integration system (see below) to demonstrate that the package successfully builds/checks on a third-party infrastructure, ideally on different systems.

DESCRIPTION

Use Authors@R to define authors and their respective roles.

This is a requirement to generate package webpages with r Githubpkg("pkgdown") later on.

README files

README files are important for any software project, as they provide an obvious first entry point to read more about the package. On github, using markdown makes them visually appealing. Using R markdown enables to include dynamically generated content. Finally, as we will see below, they can be used to easily produce a compelling and useful webpage for the R package.

Create a README.Rmd file.

The default pipeline will this be to knit the REAMDE.Rmd file to READEM.md using the knitr::knit function. To avoid the two file to come out of sync when forgetting to knit. To avoid this, one can set up a git hook, more specifically a pre-commit hook, that will check that the Rmd file is not more recent than the md file.

The simplest is to use devtools::use_readme_rmd() to create the README.Rmd template (see Important files) in the Advanced R book. This pre-commit hook ./.git/hooks/pre-commit will contain

#!/bin/bash
README=($(git diff --cached --name-only | grep -Ei '^README\.[R]?md$'))
MSG="use 'git commit --no-verify' to override this check"

if [[ ${#README[@]} == 0 ]]; then
  exit 0
fi

if [[ README.Rmd -nt README.md ]]; then
  echo -e "README.md is out of date; please re-knit README.Rmd\n$MSG"
  exit 1
elif [[ ${#README[@]} -lt 2 ]]; then
  echo -e "README.Rmd and README.md should be both staged\n$MSG"
  exit 1
fi

To create the hook if you already have a README.Rmd file, use devtools::use_git_hook:

devtools::use_git_hook("pre-commit", devtools:::render_template("readme-rmd-pre-commit.sh"))

See ?use_git_hook for details.

Add a pre-commit hook

Code versioning, dissemination and issues

Continous integration

• Suggest travis. Explain how to set up.

• If want to use another system, please document it here.

• Biocondcutor packages are testing on all platforms. Not necessary to have additional third-pary CI.

Adding status

Status shields provide quick visual information regarding the build/check status of your package and it's testing coverage.

Bioconductor shields

If the package is on Biocondcutor, then the build/check status and testing coverage will be provided by them and the shields can be directly generated with the makeBiocBuildShield and makeBiocCovrShield. Below is an example for the hpar package.

makeBiocBuildShield("hpar")
makeBiocCovrShield("hpar")

By default, the shields for the devel branch are produced. Use branch = "release" to use the release branch.

If your package is on Bioconductor, display the build and coverage shields.

Travis shield

If the package is not on Bioconductor, you'll have to set up a continuous integration server that automatically builds and checks your package. I recommend travis-ci for linux (and OSX, although I haven't successfully used that myself) and appveyor for windows to build/check the package and r CRANpkg("covr") and codecov for test coverage.

makeTravisShield("cputools", user = "ComputationalProteomicsUnit")

Add a travis build/check shield, unless there's already a Bioconductor shield.

Note: I haven't looked up how to automatically generate the code for a codecov shield yet, so that one will have to be cut and pasted manually.

Bioc and/or travis

Asking for help

It is important to provide a venue for users to ask for help and report bugs. I tend to use two channels for that, namely GitHub issues and, for Bioconductor package, the Bioconductor support site. The pkgqsts function will create a default section (level 2 below, default is 1):

pkgqsts("cputools", level = 2L, user = "ComputationalProteomicsUnit")

As this package is not on Bioconductor, the Bioconductor support site is disables with bioc = FALSE and rendered as shown below.

pkgqsts("cputools", bioc=FALSE, level = 2L, user = "ComputationalProteomicsUnit")

Add a Questions and support section.

Note that this section can be added to the README.Rmd file and in the package vignette.

Asking a question is not a straightforward thing. Here are some resources to guide users:

• Think carefully where to ask you question, where you'll reach the people most qualified to answer it: Bioconductor support forum, Stackoverflow, Stackoverflow with r tag, Bioc-devel mailing list, specific Github repositories, ...
• Don't cross-post.
• Always provide your sessionInfo()
• Jenny Bryans's reprex package.
• How to write a reproducible example

To update or discuss this, please use this issue^[https://github.com/ComputationalProteomicsUnit/cputools/issues/3] or send directly a pull request.

Testing coverage

Package webpage

Session information

sessionInfo()

ComputationalProteomicsUnit/cputools documentation built on May 6, 2019, 12:50 p.m.