Use mutate(.keep = "none") in transmute()

[markfairbanks]

Closes #470

Note: I removed most of the dedicated transmute() tests as they're now covered by mutate() tests

[eutwt]

The code change looks good. There's a tradeoff to the change, because if you're doing an immediate transmute after converting to a lazy frame, the data table is copied but currently in main it would have only used .()/list() in j. This makes the keep = "none" translation take more memory (example below). I'm not sure the relative frequency of using transmute ~immediately vs transmute later in a pipeline where your data would have already been copied.

library(bench)
library(data.table)
library(magrittr)
n <- 1e6

dt <- replicate(10, sample(n), simplify = FALSE) %>% 
  setNames(., tail(letters, length(.))) %>% 
  as.data.table()

mark(
  add_then_remove = copy(dt)[, `:=`(double_x = x * 2)][, `:=`(names(dt), NULL)],
  list_in_j = dt[, .(double_x = x * 2)],
)
#> # A tibble: 2 × 6
#>   expression           min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>      <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 add_then_remove    6.8ms   8.96ms      91.7    47.6MB     504.
#> 2 list_in_j         2.19ms   2.86ms     315.     15.3MB     137.

^{Created on 2024-12-25 with reprex v2.0.2}

One thought I had is that transmute could be a reframe() which uses vec_recycle_common(..., .size = .N) instead of list(...) in j, but that is a bad option because it would prevent GForce optimizations on the ... expressions

[markfairbanks]

I think we need a re-order to match dplyr output for tibble(x = 1, y = 2) %>% transmute(y, x)

Ah yep I'll make the reorder change.

One thought I had is that transmute could be a reframe() which uses vec_recycle_common(..., .size = .N) instead of list(...) in j, but that is a bad option because it would prevent GForce optimizations on the ... expressions

We could also do something like this to preserve GForce but I think it makes the translation look odd.

library(dtplyr)
library(dplyr)

df <- tibble(x = 1:2, y = 3:4) %>%
  lazy_dt()

df %>%
  reframe(y, mean_x = mean(x), .row_preserve = row_number()) %>%
  select(-.row_preserve)
#> Source: local data table [2 x 2]
#> Call:   `_DT1`[, .(y = y, mean_x = mean(x), .row_preserve = seq_len(.N))][, 
#>     `:=`(".row_preserve", NULL)]
#> 
#>       y mean_x
#>   <int>  <dbl>
#> 1     3    1.5
#> 2     4    1.5
#> 
#> # Use as.data.table()/as.data.frame()/as_tibble() to access results

[markfairbanks]

Actually sorry I completely did the column order thing wrong 😅

Let me take another pass at it.

[markfairbanks]

Ok good to review now

[eutwt]

Looks good, thanks! This will be good to have. Apologies for the delayed review