R Package Maintenance

I just got an automated email from the developers of the Tidyverse that they are updating the package tibble such that it will break the code in my package colorednoise, which depends on it. Part of being a package maintainer is updating your package so it won’t break when other packages do. So I thought I would document how I did this in a blog post.

First, I carefully read the email to see which tests were failing. It seemed that the problem lay in calls to tibble functions from my function matrix_model.

Then, I loaded the alpha version of tibble so I could test out some solutions to the problems indicated in the email.

library(devtools)
install_github("tidyverse/tibble")

The failing tests indicated that the following code in matrix_model was problematic with the new version of tibble:

pop %>% map(as_tibble) %>% bind_rows() %>%
      group_by(row_number()) %>% nest() %>% mutate(total = map(data, sum)) %>%
      unnest() %>% set_names(c("timestep", "total", paste0("stage", 1:stages)))

I read the documentation for tibble 2.0.0 and found that the function as_tibble behaves differently now. It doesn’t automatically repair invalid column names, which means I may have to move the call to set_names earlier in the pipeline. To investigate, I put in a breakpoint in the code before this chunk. Rstudio has breakpoints as a debugging feature. When you add a breakpoint, the code runs up until the breakpoint, then stops so that you are “inside” the function. That way I am able to play with the pop object inside the matrix_model function.

I added the .name_repair argument to my call to as_tibble, and sure enough, the column names were all in order. Now it’s time to update my package DESCRIPTION file to reflect a new dependency on tibble 2.0.0.

Imports:
  stats (>= 3.3.2),
  dplyr (>= 0.7.3),
  purrr (>= 0.2.3),
  tibble (>= 2.0.0),
  tidyr (>= 0.8.0)

I have not yet pushed these changes to GitHub or submitted to CRAN because tibble 2.0.0 is not officially live yet, but when it is, I will push to GitHub and check out the automatic package checks in Travis CI, and if they all pass, I’ll submit to CRAN.

Related