speedup <DataMask>$set(name, chunks)

[romainfrancois]

related to #5017

with the code:

library(dplyr, warn.conflicts = FALSE)

m0 <- matrix(0, 50, 2000)
groups <- sample(1:10, 50, replace = TRUE)

m1 <- apply(m0, c(1,2), function(x) sample(c(0,1),1)) %>%
  as.data.frame() %>%
  mutate(groups = groups)

m2 <- m1 %>%
  group_by(groups)

profvis::profvis({
  summarise_all(m2, list(sum))
})

we're down from:

to

[romainfrancois]

[romainfrancois]

Seems there is some more room:

library(dplyr, warn.conflicts = FALSE)
library(purrr)
library(vctrs)
#> Warning: package 'vctrs' was built under R version 4.0.2

m0 <- matrix(0, 50, 2000)
groups <- sample(1:10, 50, replace = TRUE)

m1 <- apply(m0, c(1,2), function(x) sample(c(0,1),1)) %>%
  as.data.frame() %>%
  mutate(groups = groups)

m2 <- m1 %>%
  group_by(groups)

bench::workout({
  summarise_all(m2, sum)
})
#> # A tibble: 1 x 3
#>   exprs                   process     real
#>   <bch:expr>             <bch:tm> <bch:tm>
#> 1 summarise_all(m2, sum)    2.05s    2.05s

lump <- function(x) vec_c(!!!x)
bench::workout({
  indices <- group_rows(m2)
  chops <- map(m2[1:2000], vec_chop, indices)
  chunks <- map(chops, function(chunks) {
    lump(map(chunks, sum))
  })
})
#> # A tibble: 3 x 3
#>   exprs                                                             process
#>   <bch:expr>                                                        <bch:t>
#> 1 indices <- group_rows(m2)                                            35µs
#> 2 chops <- map(m2[1:2000], vec_chop, indices)                        31.4ms
#> 3 chunks <- map(chops, function(chunks) { lump(map(chunks, sum)) }) 304.4ms
#> # … with 1 more variable: real <bch:tm>

^{Created on 2020-08-11 by the reprex package (v0.3.0.9001)}

[romainfrancois]

I think these are harmless changes, just moving some code down to compiled code.