lionel-
commented
4 years ago The advice for <foo> to <foo> coercion errors should be different than for <foo> to <bar> errors.
-
The former means the same-class fallback failed because of diverging attribtutes. It always makes sense to implement ptype2 and cast methods for a single class. We should point to a relevant FAQ.
-
In the latter case, we should be careful not to encourage implementation of coercion methods too strongly. Instead, the advice should focus on recommending users to normalise their inputs.
[x] ~FAQ about how to normalise data frame inputs (or implement common types for data frame class) to include in fallback warning implemented in #981.~ No longer relevant as we have removed the fallback warnings.
[x] FAQ about incompatible attributes.
[ ] FAQ about lossy cast errors (see also #922).
[ ] Discussion of promotion vs class hierarchy #980.
[ ] FAQ about "input must be a vector" error.