corrupt 'group_df'` error with dplyr 0.4.3

[eipi10]

I just installed the latest version of dplyr and began getting corrupt 'group_df' errors with code that used to work. Here's an example error message:

Error: corrupt 'grouped_df', contains 657842 rows, and 657865 rows in groups

My current workaround is to add %>% ungroup() on the end of the code that returns the corrupt data frame, but I'd like to figure out what's causing the error so I can implement a real fix.

I'm working with data that I can't share and I haven't been able to recreate the error with fake data. I will update this post if I can manage to create a reproducible example.

[hadley]

Please provide a reproducible example

[eipi10]

I was able to reproduce the error with fake data, so here is a reproducible example:

library(dplyr)

set.seed(111)
dat1 = data.frame(group=sample(LETTERS[1:3], 1000, replace=TRUE),
                  y=rnorm(1000))
dat1 = dat1 %>% group_by(group)

dat2 = data.frame(group=sample(LETTERS[1:2], 100, replace=TRUE),
                  y=rnorm(100))
dat2 = dat2 %>% group_by(group)

dat3 = rbind(dat1, dat2)

dat3 %>% 
  summarise(n=n())

Error: corrupt 'grouped_df', contains 1100 rows, and 1000 rows in groups

In my actual use case I don't just do, say, dat2 = dat2 %>% group_by(group). After the grouping, I do filtering, mutating, etc., but I wasn't ungrouping after completing those operations. As a result, I ended up rbind-ing two data frames that had been separately grouped.

This kind of thing never caused a problem with earlier versions of dplyr but is causing an error with the new version. For my own edification, is there something wrong (in the sense of good programming style) with separately grouping an rbind-ing, or is this just a bug in dplyr?

[hadley]

Use bind_rows() instead of rbind(). Need to see if we can un-break rbind() for this use.

[eipi10]

Will do. Thanks Hadley.

[romainfrancois]

yep rbind grabs the attributes from dat1. Perhaps just :

> rbind.tbl_df <- function( x, ..., deparse.level = 1 ) bind_rows(x, ...)

[hadley]

@krlmlr can you take this for tibble please?

[krlmlr]

For this particular issue, wouldn't it be enough to define rbind.grouped_df() here?

[hadley]

@krlmlr oh good point