Skip to content

Slower fsetdiff vs dplyr::setdiff #7783

Open
Open
Slower fsetdiff vs dplyr::setdiff#7783
Labels
atimeRequests related to adding/improving/monitoring performance regression tests via atime.

Description

@cgiachalis
cgiachalis
opened on Jun 7, 2026

This came up after code profiling where a loop was calling fsetdiff in each pass among other things.

library(data.table)

olddt <- data.table::data.table(a = paste0(1:10000, "a"))
newdt <- data.table::data.table(a = paste0(1:10010, "a"))


bench::mark(DT = data.table::fsetdiff(newdt, olddt),
            DPLYR = dplyr::setdiff(newdt, olddt),
            min_iterations = 30,
            relative = TRUE)[, 1:5]
# A tibble: 2 × 5
  expression     min  median `itr/sec` mem_alloc
  <bch:expr>   <dbl>   <dbl>     <dbl>     <dbl>
1 DT         10.7859 11.0259    1        1.81242
2 DPLYR       1       1        10.0019   1  

bench::mark(DT = data.table::fsetdiff(newdt, olddt),
            DPLYR = dplyr::setdiff(newdt, olddt),
            min_iterations = 30,
            relative = FALSE)[, 1:5]

# A tibble: 2 × 5
  expression      min   median `itr/sec` mem_alloc
  <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>
1 DT           4.77ms   5.66ms   171.193     763KB
2 DPLYR         432µs  505.4µs  1799.31      421KB

Session info

 
sessioninfo::session_info()
#> ─ Session info ───────────────────────────────────────────────────────────────
#>  setting  value
#>  version  R version 4.5.3 (2026-03-11 ucrt)
#>  os       Windows 11 x64 (build 26200)
#>  system   x86_64, mingw32
#>  ui       RTerm
#>  language (EN)
#>  collate  English_United Kingdom.utf8
#>  ctype    English_United Kingdom.utf8
#>  date     2026-06-07
#> 
#> ─ Packages ───────────────────────────────────────────────────────────────────
#>  package     * version date (UTC) lib source
#>  bench         1.1.4   2025-01-16 [1] CRAN (R 4.5.2)
#>  cli           3.6.6   2026-04-09 [1] CRAN (R 4.5.3)
#>  data.table  * 1.18.4  2026-05-06 [1] CRAN (R 4.5.3)
#>  digest        0.6.39  2025-11-19 [1] CRAN (R 4.5.2)
#>  dplyr         1.2.1   2026-04-03 [1] CRAN (R 4.5.3)
#>  evaluate      1.0.5   2025-08-27 [1] CRAN (R 4.5.2)
#>  fastmap       1.2.0   2024-05-15 [1] CRAN (R 4.5.2)
#>  fs            2.1.0   2026-04-18 [1] CRAN (R 4.5.3)
#>  generics      0.1.4   2025-05-09 [1] CRAN (R 4.5.2)
#>  glue          1.8.1   2026-04-17 [1] CRAN (R 4.5.3)
#>  htmltools     0.5.9   2025-12-04 [1] CRAN (R 4.5.2)
#>  knitr         1.51    2025-12-20 [1] CRAN (R 4.5.2)
#>  lifecycle     1.0.5   2026-01-08 [1] CRAN (R 4.5.2)
#>  magrittr      2.0.5   2026-04-04 [1] CRAN (R 4.5.3)
#>  otel          0.2.0   2025-08-29 [1] CRAN (R 4.5.2)
#>  pillar        1.11.1  2025-09-17 [1] CRAN (R 4.5.2)
#>  pkgconfig     2.0.3   2019-09-22 [1] CRAN (R 4.5.2)
#>  profmem       0.7.0   2025-05-02 [1] CRAN (R 4.5.2)
#>  R6            2.6.1   2025-02-15 [1] CRAN (R 4.5.2)
#>  reprex        2.1.1   2024-07-06 [1] CRAN (R 4.5.2)
#>  rlang         1.2.0   2026-04-06 [1] CRAN (R 4.5.3)
#>  rmarkdown     2.31    2026-03-26 [1] CRAN (R 4.5.3)
#>  rstudioapi    0.18.0  2026-01-16 [1] CRAN (R 4.5.2)
#>  sessioninfo   1.2.4   2026-06-04 [1] CRAN (R 4.5.3)
#>  tibble        3.3.1   2026-01-11 [1] CRAN (R 4.5.2)
#>  tidyselect    1.2.1   2024-03-11 [1] CRAN (R 4.5.2)
#>  utf8          1.2.6   2025-06-08 [1] CRAN (R 4.5.2)
#>  vctrs         0.7.3   2026-04-11 [1] CRAN (R 4.5.3)
#>  withr         3.0.2   2024-10-28 [1] CRAN (R 4.5.2)
#>  xfun          0.57    2026-03-20 [1] CRAN (R 4.5.3)
#>  yaml          2.3.12  2025-12-10 [1] CRAN (R 4.5.2)
#> 
#>  [1] C:/Program Files/R/library
#>  [3] C:/Program Files/R/R-4.5.3/library
#>  * ── Packages attached to the search path.
#> 
#> ──────────────────────────────────────────────────────────────────────────────

Activity

  1. ben-schwen commented on Jun 9, 2026

    @ben-schwen
    ben-schwen
    on Jun 9, 2026
    Member

    I would say that this something implementation dependent.

    data.table uses a join (based on sorting) while dplyr uses a hashtable. Enabling hashjoins is something I have on my roadmap for the future but has no PRIO atm.

    See for example if we use chmatch (which uses also a hashmap)

    olddt <- data.table(a = paste0(1:10000, "a"))
newdt <- data.table(a = paste0(1:10010, "a"))

bench::mark(
    DT_fsetdiff = fsetdiff(newdt, olddt),
    DT_chmatch  = unique(newdt[chmatch(a, olddt$a, nomatch = 0L) == 0L]),
    DPLYR       = dplyr::setdiff(newdt, olddt),
    min_iterations = 30,
    relative = TRUE, check=FALSE,
)[, 1:5]

# A tibble: 3 × 5
#   expression    min median `itr/sec` mem_alloc
#   <bch:expr>  <dbl>  <dbl>     <dbl>     <dbl>
# 1 DT_fsetdiff  2.79   2.49      1         4.30
# 2 DT_chmatch   1      1         2.49      1
# 3 DPLYR        1.29   1.30      1.87      2.81

  2. cgiachalis commented on Jun 9, 2026

    @cgiachalis
    cgiachalis
    on Jun 9, 2026
    Author

    Indeed. I've used chmatch approach before, for instance:

    https://github.com/cgiachalis/storr.tiledb/blob/8b6433e07f846cad5c1feb1ab46355e3281ea366/R/TileDBDriver.R#L666-L667

     # Find unused hashes (equiv. setdiff(x, y))
 xh[data.table::chmatch(xh, yh, 0L) == 0L]
  3. tdhock
    added
    atimeRequests related to adding/improving/monitoring performance regression tests via atime.
    on Jun 9, 2026

  4. tdhock commented on Jun 9, 2026

    @tdhock
    tdhock
    on Jun 9, 2026
    Member

    we should convert this benchmark to atime

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    atimeRequests related to adding/improving/monitoring performance regression tests via atime.

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions