cbind on tbl_df should return a tbl_df?

[daattali]

Not sure if this is a feature or a bug - if I cbind two separate tbl_df objects, I expected the result to also be a tbl_df rather than having to cast it myself.

df <- tbl_df(data.frame(a = letters))
DF <- tbl_df(data.frame(A = LETTERS))
bound <- cbind(df, DF)
class(bound)

[romainfrancois]

We don't have control over what cbind does. dplyr had bind_all and cbind_list but that also gives a data.frame no matter what. See https://github.com/hadley/dplyr/blob/master/src/bind.cpp#L171

Perhaps we should replace this line :

out.attr( "class" ) = "data.frame" ;

by this line to make a tbl_df instead, as this is what rbind_* are doing.

out.attr( "class" ) = classes_not_grouped() ;

[hadley]

@romainfrancois can you please make that change? Ideally we'd preserve the input class, but it's hard to know what to do if you have a mix of tbl df's and data frames

[daattali]

Thanks for responding so promptly.
IMO the safest behaviour would be to revert to the "most recent common ancestor" -- I would expect to get a data.frame back if I pass in a data.frame and a tbl_df. But that's clearly a design choice up to you that I would not argue about.

[lionel-]

We don't have control over what cbind does.

Actually cbind accepts S3 methods. So it's possible to produce a tbl_df whenever the first argument is a tbl_df. However, we can't have this kind of control if the first argument is a data.frame.

So, for consistency, cbind.tbl_df and friends could produce a tbl_df when all arguments are tbl_dfs, and a data.frame otherwise?

[hadley]

@lionelgit have you read the dispatch section in help? It's not standard S3 dispatch, and I'm pretty sure it won't work when you have a mix of tbl dfs and data frames. But I'm happy to be proved wrong

[lionel-]

Indeed, I didn't know this subtlety.

Still, it may make sense to have a cbind.tbl_df method to produce a tbl_df when all arguments are tbl_df.

[steromano]

I'll just comment here instead of opening a new issue, as this is likely a small thing.
Contrary to what the documentation says, cbind_cols seems to always return a data.frame:

x <- as.tbl(iris)
bind_cols(x, x) %>% class  # data.frame

[jennybc]

👍 to this thread and proposed enhancement. What @steromano said: when I supply two tbl_dfs to bind_cols(), I get a data.frame back. It's unexpected and contradicts the documentation.

[romainfrancois]

Now it's always making a tbl_df which is harmless anyway, even if both are just "data.frame".