eval_tidy failures: induced by differences in dplyr::mutate and tidytable::dt_mutate

[xiaodaigh]

I think there is a difference between how dplyr treats quoted code and how tidytable treats them.

This is a blocking issue for integrating tidytable with disk.frame, see DiskFrame/disk.frame#271

fn = function(a,b) {
  a + b
}

# dplyr seems to be able to handle this correctly
mwe_mutate = function(...) {
  quo_dotdotdot = rlang::enquos(...)
  data = data.frame(num = 1:100)
  code = rlang::quo(mutate(data, !!!quo_dotdotdot))  
  rlang::eval_tidy(code)
}

mwe_mutate(b = fn(num, num))

# tidytable seems to fail 
mwe_dt_mutate = function(...) {
  quo_dotdotdot = rlang::enquos(...)
  data = data.frame(num = 1:100)
  code = rlang::quo(dt_mutate(data, !!!quo_dotdotdot))  
  rlang::eval_tidy(code)
}

mwe_dt_mutate(b = fn(num, num)) # this gives an error

[markfairbanks]

This issue comes from the fact that quosures don't work inside of data.table.

For this reason you can't use rlang::enquos(), and must instead use rlang::enexprs():

mwe_dt_mutate = function(...) {
  expr_dotdotdot = rlang::enexprs(...) # Use enexprs() instead of enquos()
  data = data.frame(num = 1:100)
  code = rlang::quo(dt_mutate(data, !!!expr_dotdotdot))
  rlang::eval_tidy(code)
}

mwe_dt_mutate(b = fn(num, num))

[markfairbanks]

Here's a good example of a very basic mutate translation:

quo_dt_mutate <- function(.data, ...) {
  dots <- enquos(...)
  
  eval_tidy(expr(
    .data[, ':='(!!!dots)][]
  ))
}

expr_dt_mutate <- function(.data, ...) {
  dots <- enexprs(...)
  
  eval_tidy(expr(
    .data[, ':='(!!!dots)][]
  ))
}

test_df <- data.table(x = 1:3)

quo_dt_mutate(test_df, double_x = x * 2) # Fails

expr_dt_mutate(test_df, double_x = x * 2) # Works

[markfairbanks]

This issue is now fixed in the new version of tidytable up on CRAN (v0.5.0).

Note there's been a small syntax change where all functions now use verb.() syntax instead of dt_verb(). For example dt_mutate() has been renamed to mutate.().

The dt_ versions still exist in tidytable, but it seems like a lot of packages are using that prefix - so I decided to change things during version v0.4.0.

library(tidytable, warn.conflicts = FALSE)

fn = function(a,b) {
  a + b
}

mwe_dt_mutate = function(...) {
  quo_dotdotdot = rlang::enquos(...)
  data = data.frame(num = 1:100)
  code = rlang::quo(mutate.(data, !!!quo_dotdotdot))  
  rlang::eval_tidy(code)
}

mwe_dt_mutate(b = fn(num, num)) %>% head()
#>    num  b
#> 1:   1  2
#> 2:   2  4
#> 3:   3  6
#> 4:   4  8
#> 5:   5 10
#> 6:   6 12

[xiaodaigh]

I will take a look. Busy with work and some Julia open source work atm.

[markfairbanks]

Sounds good, if there's anything I can help with let me know