Improve recycling for reused atomic values in tibble()

[krlmlr]

Original issue tidyverse/dplyr#2360 by @rpruim.

The following result for data_frame() is unexpected (to me) given the result of mutate() below it.

data_frame(a = 1:3, b = 1, c = b/sum(b))
## # A tibble: 3 × 3
##       a     b     c
##   <int> <dbl> <dbl>
## 1     1     1     1
## 2     2     1     1
## 3     3     1     1
data_frame(a = 1:3) %>% mutate(b = 1, c = b/sum(b))
## # A tibble: 3 × 3
##       a     b         c
##   <int> <dbl>     <dbl>
## 1     1     1 0.3333333
## 2     2     1 0.3333333
## 3     3     1 0.3333333

[rpruim]

Glad to see this is on the list for fixing. Thanks.

[NikNakk]

tibble is currently implemented using as_tibble(lst(...)). The eval_tidy is done in lst, but the length == 1L columns aren't expanded to length == nrow(tbl) until the as_tibble is done. It looks as though either lst would need to be altered to take a parameter to indicate that scalars should be duplicated to fill the number of rows, or that some of the code will need to be duplicated from lst into tibble.

The bigger issue is that we don't know the number of rows until the first non-scalar column is evaluated. For example:

tibble(w = 1, x = 2, y = sum(w) / mean(x), z = 3:5)

is valid syntactically but when y is evaluated, the number of rows is unknown. Another example is:

f <- function(a) {a + c(4, 9, 12)}
tibble(w = 1, x = 2, y = sum(w) / mean(x), z = f(x))

It's even less evident looking at the call to know that there will eventually be three rows until all of the arguments have been evaluated.

[krlmlr]

Thanks @NikNakk for the analysis. I think tibble(a = ..., b = ...) generally should return the same as tibble(a = ...) %>% mutate(b = ...), so your last two examples aren't a big problem. Still, I don't see an easy way here, let's shelve this for now.

[github-actions]

This old thread has been automatically locked. If you think you have found something related to this, please open a new issue and link to this old issue if necessary.