Fast `select()`

[markfairbanks]

Currently select() always takes a deep copy. We can instead drop columns by reference if an implicit or explicit copy has already occurred in the pipe chain.

Waiting on #366

library(dplyr, warn.conflicts = FALSE)
library(dtplyr)

df <- lazy_dt(tibble(x = 1, y = 2))

# Old
df %>%
  mutate(z = 3) %>%
  select(x, z)
#> Source: local data table [1 x 2]
#> Call:   copy(`_DT1`)[, `:=`(z = 3)][, .(x, z)]
#> 
#>       x     z
#>   <dbl> <dbl>
#> 1     1     3
#> 
#> # Use as.data.table()/as.data.frame()/as_tibble() to access results

# New
remove_vars <- dtplyr:::remove_vars

df %>%
  mutate(z = 3) %>%
  remove_vars("y")
#> Source: local data table [1 x 2]
#> Call:   copy(`_DT1`)[, `:=`(z = 3)][, `:=`("y", NULL)]
#> 
#>       x     z
#>   <dbl> <dbl>
#> 1     1     3
#> 
#> # Use as.data.table()/as.data.frame()/as_tibble() to access results

[markfairbanks]

As I think about this there would also have to be some consideration for select() being able to reorder columns. The remove_vars case might also need a call to setcolorder().

Ex: df %>% mutate() %>% select(z, x)

Also would have to account for when columns are renamed.

And some benchmarks:

pacman::p_load(dplyr, dtplyr, stringi, data.table)

data_size <- 10000000
df <- tibble(a = sample(stri_rand_strings(100, 4), data_size, TRUE),
             b = sample(stri_rand_strings(100, 4), data_size, TRUE),
             c = sample(1:100, data_size, TRUE)) %>%
  lazy_dt()

remove_vars <- dtplyr:::remove_vars

bench::mark(
  old = df %>%
    mutate(d = 1) %>%
    select(a, b, d) %>%
    as.data.table(),
  new = df %>%
    mutate(d = 1) %>%
    remove_vars("c") %>%
    as.data.table(),
  check = FALSE, iterations = 30
)
#> Warning: Some expressions had a GC in every iteration; so filtering is disabled.
#> # A tibble: 2 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 old         406.1ms  461.7ms      2.12     728MB     2.19
#> 2 new          51.3ms   58.9ms     10.2      267MB     4.09