Speed of first() evaluated within grouped tibbles

[dkutner]

Using dplyr 1.1.0 and vctrs 0.5.2, I'm noticing speed issues when evaluating first within grouped tibbles.

df <- dplyr::group_by(tibble::tibble(x = 1:10000, grp = rep(1:100, 100)), grp)
bench::mark(
  first = dplyr::summarize(
    df,
    f = dplyr::first(x)
  ),
  indexed = dplyr::summarize(
    df,
    f = x[1]
  )
)
#> # A tibble: 2 x 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 first        7.44ms   7.98ms      119.    1.57MB     19.1
#> 2 indexed      1.01ms   1.06ms      901.   55.62KB     12.7

On dplyr 1.0.10 with vctrs 0.5.0, the first benchmark is about 1.7 ms.

[hadley]

Why do you think this is related to grouped tibbles?

[dkutner]

I'm not sure if it's isolated to grouped tibbles, but the effect seems to be more apparent on them. If we omit the grouping in the above reprex, both benchmarks have similar times.

[dkutner]

I'm probably just observing the effect of running first many times. Here's a simpler example without a grouped tibble. On the previous version of dplyr, the first benchmark took 8 microseconds.

x <- 1:10000
bench::mark(
  first = dplyr::first(x),
  indexed = x[1]
)
#> # A tibble: 2 x 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 first        64.7us   69.8us    13674.    10.3MB     18.9
#> 2 indexed       100ns    200ns  4016387.        0B      0

[hadley]

Thanks for clarifying!