If the length of .groups is > 1, it's caught here, which raises an error and drops to the R level, which also raises an error.

However, the error message from dplyr isn't quite fit for our purposes - it mentions that rowwise is a valid value for .groups, which is true in dplyr but not in arrow. Is it worth catching this case ourselves and providing our own error message, which doesn't mention rowwise?

[Edited after reading the tests below] Or is this something worth living with, as the alternative would require significant work to implement for little actual benefit?

mtcars %>% 
   Table$create() %>%
   group_by(mpg) %>% 
   summarise(.groups = c("drop", "keep")) %>%
   collect()
# Warning: Error : .groups is not a string (a length one character vector).; pulling data into R
# Error: `.groups` can't be <chr>
# ℹ Possible values are NULL (default), "drop_last", "drop", "keep", and "rowwise"
# Run `rlang::last_error()` to see where the error occurred.

I'm open to suggestion, and I'll think some more about whether this is a problem we have anywhere else, but my thinking for leaving it as it is: (1) when querying on a dataset, it will just error and tell you to pull into R (where rowwise would work); (2) it's an experimental feature in dplyr so IDK how much we should worry about the smoothest experience when you go off the default.

Sure, that makes a lot of sense and good point about it being an experimental feature - sounds like something to leave for now but revisit if it comes up again elsewhere.