Don't error when grouping zero-row data frames. Fixes #486

[wch]

Fixes the error that previously occurred with:

dplyr:::grouped_df_impl(data.frame(a = numeric(0), g = character(0)), list(quote(g)), drop = TRUE)

I'm new to the C++ part of this codebase so please feel free to change, critique, etc.

[hadley]

This seems reasonable (cc @romainfrancois), but did you check what happens when you try to filter, summarise etc a data frame with 0-length index?

[romainfrancois]

+1. Looks correct, but needs some dedicated test cases

[hadley]

@wch could you please add a few test cases?

[wch]

Writing tests, but I'm not sure that all results are correct:

dfg <- group_by(data.frame(a = numeric(0), g = character(0)), g)
group_size(dfg)
# integer(0)

Should that be 0 instead of integer(0)?

[hadley]

No, because group_size() should have the same length as the number of groups.

[wch]

So it shouldn't group on g? Because it does that with this patch:

dfg
# Source: local data frame [0 x 2]
# Groups: g

groups(dfg)
# [[1]]
# g

[wch]

It looks like I misunderstood what group_size is supposed to return. The test results are correct, and I've added the tests.

[wch]

I just added some more tests.

@romainfrancois the call to mutate fails, with this error:

> x <- mutate(dfg, c = b + 1)
Error: not compatible with requested type 
4 stop(structure(list(message = "not compatible with requested type", 
    call = NULL, cppstack = NULL), .Names = c("message", "call", 
"cppstack"), class = c("Rcpp::not_compatible", "C++Error", "error", 
"condition"))) 
3 mutate_impl(.data, named_dots(...), environment()) 
2 mutate.tbl_df(dfg, c = b + 1) at manip.r#81
1 mutate(dfg, c = b + 1) 

[romainfrancois]

Thanks @wch; I'll pick it up.

[romainfrancois]

I only slightly modified two test cases. I think the rest was fixed as part of dealing with #486

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/