data.frame cast fallback

[hadley]

I think we may need a some data.frame fallbacks so that we get something like this:

vec_rbind(tibble(x = 1), data.table(x = 1))
# Warning: combining <tibble> and <data.table>; falling back to <data.frame>

(This is not a general regret that we're moving away from inheritance; it's a specific pragmatic fallback + warning for people working with multiple data frame subclasses)

[lionel-]

I like the idea, the warning will encourage people to normalise their inputs (we should link to a FAQ), and so their code won't be affected when the class authors end up implementing methods.

Implementation will probably require a sentinel data frame class to avoid repeat warnings.

[lionel-]

Somehow it makes more sense for data frames than for other types. For example it wouldn't make sense to coerce factors to integers with a warning.

Maybe the general rule is that if a class explicitly inherits from one of the base classes (including data.frame), then a fallback + warning is in order? (Not that we should rush implementing a general fallback.)