Oooh, this looks awesome, thanks!

Frankly, it's a bit scary too, so I'll want to play with it a bit before merging. Unfortunately, it's the start of semester, so I may not get to this for a little while.

If

If @cjyetman has time and interest (no pressure), he could also merge this after some testing.

Frankly, it's a bit scary too, so I'll want to play with it a bit before merging

Yup, completely agree, the last thing you want is realize later that codes and names don't match. I'll see later if I can test more extensively

FYI an improvement that makes the code much simpler was suggested to me on SO, I'll modify the PR

New timing after the simplification:

#> # A tibble: 1 × 3
#>   pkg     median mem_alloc
#>   <chr> <bch:tm> <bch:byt>
#> 1 fork     1.18s     742MB

@vincentarelbundock or @cjyetman can you enable the CI for all commits in the PR so that I know if they pass?

@vincentarelbundock or @cjyetman can you enable the CI for all commits in the PR so that I know if they pass?

I cannot (enable CI on commits from forks), but I have approved the CI to run on your most recent push.

For information, if you're ok with adding a dependency on data.table (to have access to chmatch()) then it's possible to go down to 750ms and 550MB for the example above

Got it.

I think it's best to stay dependency-free, though

Looks like all the tests pass. I guess that's better than whatever impressionistic sense I could get from playing with different things. What do you think @cjyetman , should I just merge this?

I'll try to take a look at it today.

I ran revdepcheck on the 31 reverse dependencies (both Imports and Suggests) and I found no new errors

@etiennebacher can you explain what cross::run() is so that I can replicate your example?

It allows us to run the same code automatically with two (or more) versions of the same package. I specified the main branch and my fork at the beginning and then I give the code to run. cross::run() will automatically install both versions in a temp location, run the code for each version separately, and output the result in a nested table that I unnest at the end.

https://github.com/DavisVaughan/cross

This seems to work as expected, tests pass, build passes, so that seems ok. I can't exactly replicate the example because I don't know where cross::run() comes from, but loading different versions of countrycode in different sessions does seem to show a significant speed improvement with this example.

A few suggestions:

• I would strongly prefer to see minor whitespace changes on unrelated lines removed from this PR to make it easier to review.
• The ifelse() change starting on L207 I would prefer to see done in a separate PR because it's conceptually a separate idea/improvement.

Thanks a ton for the review @cjyetman, I appreciate your time.

I also agree with the comments about white space and separate PRs for different concepts.

That said, it's probably fine to merge this now in order to avoid more busywork for Etienne. So @etiennebacher, if you want to add your name as a contributor (your choice), I can merge the PR.

A 5x speedup is nothing to scoff at. Very nice! Thanks!

It allows us to run the same code automatically with two (or more) versions of the same package. I specified the main branch and my fork at the beginning and then I give the code to run. cross::run() will automatically install both versions in a temp location, run the code for each version separately, and output the result in a nested table that I unnest at the end.

https://github.com/DavisVaughan/cross

Thanks for the link, I was not finding {cross} on CRAN.

Note that if multiple destinations are used, the speed-up disappears is reduced and the memory usage is the same, but they still produce the same result:

  library(bench)
  
  out <- cross::run(
    pkgs = c("vincentarelbundock/countrycode",
             "etiennebacher/countrycode@large-speedup"),
    ~{
      library(countrycode)
      
      sourcevar <- sample(codelist$country.name.en, 1e7, TRUE)
      
      bench::mark(
        countrycode(sourcevar, "country.name", c("iso3c", "iso3n")),
        iterations = 10
      )
    }
  )
  
  tidyr::unnest(out, result) |>
    dplyr::select(pkg, median, mem_alloc) |>
    dplyr::mutate(pkg = ifelse(grepl("vincent", pkg), "main", "fork"))

#> # A tibble: 2 × 3
#>   pkg     median mem_alloc
#>   <chr> <bch:tm> <bch:byt>
#> 1 main     5.95s    2.45GB
#> 2 fork     4.56s    2.45GB

Thanks to both of you (and especially Etienne!). Merged!