Fix rendering vctrs::list_of columns

[zeehio]

DT needs to convert data frames/tibbles to json.

The R to json conversion has an ambiguity on scalar vector conversions: R can't tell the difference between an "atomic vector of length 1" and "a scalar", while in json a scalar is "just the value" and a vector of length 1 is represented in a list ["just the value"].

When our data frame has a list column, we need to resolve this ambiguity. For list-columns, in DT we need each item of the column to be itself a json list. This means that if the item already is a vector of length > 1, we have no trouble, but if the item is a scalar, we need to coerce it to a list to guarantee that the conversion behaves as expected. See the example:

Example:

Imagine in a two-row dataframe we have a list column with the following content: list(c("a", "b"), "c"). Each element of this list column needs to become a list when converted to json. Naturally, c("a", "b") becomes ["a", "b"], but what about "c"? It would become "c" and we need it to become ["c"] instead.

In R, both items c("a", "b") and "c" are character vectors. We need to coerce the second to a list so x becomes list(c("a", "b"), list("c")) and the conversion to json works as we require.

The problem:

This coercion fails if the list column is not of type list, but a vctrs::list_of with a ptype=character(0) vector. In that scenario, when we try to turn "c" (a character vector) into list("c") (a list), vctrs steps in and tries to coerce the list back to a character, because vctrs::list_of enforces the character type to all elements of the list. This raises a coercion error as seen on #1180.

The easiest solution in this case is to drop the list_of type if we need to, and leave the column as a regular list, so items may have mixed types in R -but will have homogeneous types in json!- and the assignment works without coercions from vctrs.

Reprex in the associated issue #1180 .

I added a unit test to prevent regressions and updated the NEWS file.

Closes #1180

[CLAassistant]

All committers have signed the CLA.

[yihui]

Thanks! This looks good to me. Cc @gadenbuie