feat: Initial ALTREP support for LIST logical type

[romainfrancois]

This uses the same approach as with alters strings, i.e. Elt() materialise all with VECTOR_ELT(AltrepVectorWrapper::Get(x)->Vector(), i);

there is probably a better way involving not forcing materialisation, i.e. an Elt() method that would just call duckdb_r_transform(chunk.data[column_index], dest, dest_offset, chunk.size(), false);

[krlmlr]

Thanks, nice! Can you please add a test?

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 85.93%. Comparing base (10b3416) to head (68460ac).

❗ Current head 68460ac differs from pull request most recent head 167880e. Consider uploading reports for the commit 167880e to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #77      +/-   ##
==========================================
+ Coverage   85.89%   85.93%   +0.03%     
==========================================
  Files         106      106              
  Lines        3602     3612      +10     
==========================================
+ Hits         3094     3104      +10     
  Misses        508      508              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[romainfrancois]

Will do. I just need to make the use case I started from independent of duckplyr, in test-list.R ?

library(duckdb)
library(duckplyr)

df <- tibble(a = list(1, c(1,2))) |>
  as_duckplyr_df()

filter(df, 0 == 0)
#> materializing:
#> ---------------------
#> --- Relation Tree ---
#> ---------------------
#> Filter [==(0.0, 0.0)]
#>   r_dataframe_scan(0x12b4a5e30)
#> 
#> ---------------------
#> -- Result Columns  --
#> ---------------------
#> - a (DOUBLE[])
#> 
#> # A tibble: 2 × 1
#>   a        
#>   <list>   
#> 1 <dbl [1]>
#> 2 <dbl [2]>

^{Created on 2024-02-22 with reprex v2.1.0}

[krlmlr]

Try this:

duckdb <- asNamespace("duckdb")
con <- DBI::dbConnect(duckdb::duckdb(config = list(allow_unsigned_extensions = "true")))
experimental <- FALSE
invisible(
  DBI::dbExecute(con, "INSTALL 'rfuns' FROM 'http://duckdb-rfuns.s3.us-east-1.amazonaws.com'")
)
invisible(DBI::dbExecute(con, "LOAD 'rfuns'"))
invisible(DBI::dbExecute(con, "CREATE MACRO \"==\"(x, y) AS \"r_base::==\"(x, y)"))
df1 <- tibble::tibble(a = list(1, c(1,2)), b = 1)

rel1 <- duckdb$rel_from_df(con, df1, experimental = experimental)
rel2 <- duckdb$rel_filter(
  rel1,
  list(
    duckdb$expr_function(
      "==",
      list(
        duckdb$expr_reference("b"),
        if ("experimental" %in% names(formals(duckdb$expr_constant))) {
          duckdb$expr_constant(1, experimental = experimental)
        } else {
          duckdb$expr_constant(1)
        }
      )
    )
  )
)
rel2
duckdb$rel_to_altrep(rel2)

I have infrastructure to generate those.

[krlmlr]

Even simpler:

duckdb <- asNamespace("duckdb")
con <- DBI::dbConnect(duckdb::duckdb())
experimental <- FALSE
df1 <- tibble::tibble(a = list(1, c(1,2)), b = 1)

rel1 <- duckdb$rel_from_df(con, df1, experimental = experimental)
rel2 <- duckdb$rel_filter(
  rel1,
  list(
    if ("experimental" %in% names(formals(duckdb$expr_constant))) {
      duckdb$expr_constant(TRUE, experimental = experimental)
    } else {
      duckdb$expr_constant(TRUE)
    }
  )
)
rel2
duckdb$rel_to_altrep(rel2)

[krlmlr]

In a duckdb test, we don't need duckdb$, of course.

[romainfrancois]

I added a test in test_list.R but I get the feeling it might belong to test-rel_api.R instead ?

[krlmlr]

Tests now fail if ALTREP is not supported. Can you please add conditional compilation?

[krlmlr]

Thanks, sorry for the confusion. I'm not sure we even need the filter here for the roundtrip, but it doesn't hurt.

[krlmlr]

Looks good, thanks!

[krlmlr]

Thanks!