FR: Make read_parquet tibble of class tbl_df

[lbm364dl]

Following a comment in #104:

I wonder why the default value of class is c("tbl", "data.frame") and not c("tbl_df", "tbl", "data.frame")? It seems to me that tbl is more of an "interface" than a class, since all actual instances of tbls that I have encountered previously are actually also other classes such as either tbl_df or more complex types (such as those found in dbplyr). And the returned object is structurally a tbl_df, right, so it would be useful for any functions taking it to be able to use the most specific class information?

I arrived here because of this:

tmp <- tempfile(fileext = "parquet")

tibble::tibble(a = 1) |>
  nanoparquet::write_parquet(tmp)

df <- nanoparquet::read_parquet(tmp)

class(df)
#> [1] "tbl"        "data.frame"
class(tibble::as_tibble(df))
#> [1] "tbl_df"     "tbl"        "data.frame"

df |>
  dplyr::summarise() |>
  class()
#> [1] "data.frame"

df |>
  tibble::as_tibble() |>
  dplyr::summarise() |>
  class()
#> [1] "tbl_df"     "tbl"        "data.frame"

^{Created on 2025-07-16 with reprex v2.1.1}

Functions like dplyr::summarise() use tibble::is_tibble() internally to check whether something is a tibble, but this is checking specifically on class tbl_df, not tbl. Since this check gives FALSE, the tibble is silently converted to a raw data.frame.

Is there any reason why the object returned by read_parquet() doesn't have the more specific tbl_df class? If it did have it, situations like the one I described would be clearer and less error prone, i.e., users wouldn't have to manually add this class with tibble::as_tibble().