Should collect 'fully' materialize a duckplyr_df?

[TimTaylor]

Calling str() on a collected duckplyr_df is very slow on first use.

I assumed collect fully materialized the object. Is this a misunderstanding on my part?

packageVersion("duckplyr")
#> [1] '1.1.0.9000'
base_url <- "https://blobs.duckdb.org/flight-data-partitioned/Year=2024/data_0.parquet"
flights_parquet <- read_parquet_duckdb(base_url)
x <- collect(flights_parquet)
system.time(str(x))
#>    user  system elapsed 
#>   4.898   0.623   5.539
system.time(str(x))
#>    user  system elapsed 
#>   0.014   0.000   0.014

[TimTaylor]

I'm assuming it is loading some columns as ALTREP vectors. If so, do you think is it worth forcing their materialisation too?

[krlmlr]

Thanks, interesting. This could be strings that are allocated (in the R world) on demand.

Why is it important to "fully" materialize?

[TimTaylor]

Why is it important to "fully" materialize?

A good question to which I don't have a good answer! Just a feeling that something feels a little 'not quite right'. Perhaps a little additional commentary in the documentation is sufficient.