Recently, the result back to learnr changed to a graded_feedback which incorporates graded. We need to make sure we satisfy this new structure.

From https://github.com/rstudio-education/gradethis/blob/6baf2d9ca584a526761ae76b0510dc274bde0561/R/graded.R#L48

structure(
    list(
      message = grade$message, 
      correct = grade$correct, 
      type = type, 
      location = match.arg(location)
    ),
    class = "grader_feedback"
)

hook up and wrap the ast checker too

Tests for this package will simply be testing that the wrapper py_grade_learnr works. Python testing will eventually be added to a Python package. Things to consider:

• data type translations are good
• returns the right graded results

We were thinking it might be good to separate some issues for R and some for the pygradethis library. We might want to still clarify this since certain issues might only make sense in one package or the other.

This will be especially important to remain consistent with gradethis and avoid possible function name clashing.

https://rstudio.github.io/reticulate/articles/python_dependencies.html#using-configreticulate

To start outlining the functions to expose and map to gradethis, it would be useful to start an API reference page to start.

https://pkgdown.r-lib.org/articles/pkgdown.html

There are some data structures that "require" an associated class. The R wrapper py_grade_learnr will have to handle this because reticulate will not likely be able to properly translate a Python class.

As a priority, let's make sure the final returned structure to learnr at least reflects newest gradethis for consistency. It isn't clear whether the class attribute is actually used beyond testing purposes.

Tasks:

• grader_graded here

The class here might only be used in gradethis internally, so it may not be necessary.

structure(
    list(message = message %||% "", correct = correct),
    class = "grader_graded"
)

• grader_feedback here

This class is definitely more important for learnr.

structure(
    list(
      message = grade$message, 
      correct = grade$correct, 
      type = type, 
      location = match.arg(location)
    ),
    class = "grader_feedback"
)

In the scenario that you would want only to grade with failing cases via python_fail_if cases, allow grading to "fall-through" or be marked as correct, re: this gradethis issue: rstudio/gradethis#118

The PR solution (rstudio/gradethis#124) adds two default arguments to grade_result, so the same must happen for python_grade_result

Now that the python package is on pypi here, so let's try:

• using installed python package
• import learnr/gradethis specific functions