NOTE: This vignette describes how extend the broom generics in an external package. To use the broom generics in an external you can Import broom as you would normally.

This guide describes how to extend the tidy(), glance() and augment() generics. broom recently started re-exporting these generics from the modelgenerics package. modelgenerics is not yet on CRAN, nor is the version of broom that uses modelgenerics. You’ll need to wait for these CRAN releases (timeline TBD at the moment) before release your package to CRAN if you extend broom in the manner described here.

If you are a package developer, the best place to include tidy(), glance() or augment() methods is in your own package. To do this, you’ll need to:

  1. Re-export the tidy(), glance() and augment() generics from the modelgenerics package.
  2. Implement appropriate tidying methods.
  3. Test the tidying methods

You can re-export generics from broom itself, but modelgenerics is a much lighter dependency.

The dustpan package demonstrates use of modelgenerics and modeltests, and you may find it helpful to browse.

Re-exporting tidy(), glance() and augment() generics

First you’ll need to add the modelgenerics package to Imports. We recommend using the usethis package for this:

usethis::use_package("modelgenerics", "Imports")

Next you’ll need to re-export the appropriate tidying methods. If you plan to implement a glance() method, for example, you can re-export the glance generic by adding the following somewhere in your package

#' @importFrom modelgenerics glance
#' @export

Run devtools::document() for these changes to take effect.

Note: please do not define tidy(), glance() or augment() generics in your package. This will result in namespace conflicts whenever your package is used along other packages that also export tidying methods.

Implement appropriate tidying methods

This part is mostly up to you, although there’s lots of advice in vignette("adding-tidiers").


Now that you have a tidying method, you’ll want to test it. In addition to testing the specific functionality of your tidying methods, you should also pass the standardized tests exported from modeltests.

To import these tests you’ll need to add modeltests to Suggests, which you can accomplish with:

# once modeltests makes its way onto CRAN:
# usethis::use_package("modeltests", "Suggests")

# until then:

You can then write tests for your tidiers following the advice in vignette("adding-tidiers").

If you’re run devtools::test() and devtools::check() and didn’t fail any tests or set off R CMD Check ERRORS, WARNINGS or NOTES, you’re done!


Should I Import or Suggest broom or broom.mixed?

If you are using tidiers from broom, such as tidy.lm(), then you will need to depend on broom, probably via Imports:

use_package("broom", "Imports")

Similarly, if you are using tidiers from broom.mixed, you’ll need to Import broom.mixed.

You can import broom.mixed separately from broom, or you can mix and match them however you want.

I don’t want to depend on modelgenerics. Can I re-export tidy(), glance() and augment() from broom?

I think so, although I haven’t tested this yet. This is on the TODO list at the moment.

What do I do if my tidiers don’t pass the tests?

If you are introducing new arguments or column names in output, you’ll need to make a pull request to modeltests to add the new argument/column names to the glossary of allowed names.

If you’d like to change the test specifications, or if you find a bug in the tests, open an issue in modeltests.

As a last resort, you can set strict = FALSE to run only a bare minimum of tests. This is not recommended.