feat: Reexport non-deprecated dplyr functions

[krlmlr]

Closes #144.

[krlmlr]

@DavisVaughan: Can you please take a look? Anything else we want to reexport, any deprecated functions I've included by accident?

[krlmlr]

Please review only the first comment, I'll need to add reexport statements once we have agreed on the set of reexports.

[DavisVaughan]

Here is my list of what I think we can re-export reasonably (and I do think we should target the exact package they came from, not dplyr in all cases). The idea is to try and support most interactive usage of dplyr, and to remove anything superseded, deprecated, defunct, etc, which could make it harder on dplyr in the future if we exported them here. I think for programmatic tools, users can pull from rlang or dplyr directly as needed.

This list is just based on the "new" exports that are in this PR, but you could do a once over of all the other existing exports and try and apply these principles too if you wanted to.

From rlang:

• .data

From tibble:

• add_row
• as_tibble
• tribble

From dplyr:

• between
• bind_cols
• bind_rows
• case_match
• coalesce
• consecutive_id
• cumall
• cumany
• cume_dist
• cummean
• cur_column
• cur_group
• cur_group_id
• cur_group_rows
• dense_rank
• group_cols
• group_indices
• group_nest
• group_split
• group_trim
• groups
• if_else
• min_rank
• n_groups
• na_if
• near
• ntile
• order_by
• percent_rank
• pick
• top_n
• with_order

From tidyselect:

• all_of
• any_of
• contains
• everything
• last_col
• matches
• num_range
• one_of
• starts_with
• where

Remove (mostly superseded or deprecated):

• add_rownames
• all_vars
• any_vars
• arrange_all
• arrange_at
• arrange_if
• as_data_frame
• as_label (we'd like to remove from dplyr, comes from rlang)
• check_dbplyr (should not be relevant)
• combine
• common_by (programmatic tool, pull from dplyr as needed)
• copy_to (should not be relevant)
• cur_data
• cur_data_all
• current_vars
• distinct_all
• distinct_at
• distinct_if
• distinct_prepare (programmatic tool)
• ensym (we'd like to remove from dplyr, comes from rlang)
• ensyms (we'd like to remove from dplyr, comes from rlang)
• expr (we'd like to remove from dplyr, comes from rlang)
• enquo (we'd like to remove from dplyr, comes from rlang)
• enquos (we'd like to remove from dplyr, comes from rlang)
• filter_all
• filter_at
• filter_if
• group_by_all
• group_by_at
• group_by_if
• group_by_drop_default (programmatic tool)
• group_by_prepare (programmatic tool)
• grouped_df (programmatic tool)
• ident (sql)
• is.tbl (sql)
• last_dplyr_warnings (probably not relevant? not sure, but kind of experimental)
• lst (trying to remove from dplyr)
• make_tbl (trying to remove from dplyr)
• mutate_at
• mutate_if
• mutate_all (this is currently exported but id remove it)
• new_grouped_df (programmatic tool)
• new_rowwise_df (programmatic tool)
• progress_estimated
• quo
• quo_name
• quos
• recode
• recode_factor
• show_query
• everything sql_*
• everything src_*
• summaris/ze_if
• summaris/ze_all
• summaris/ze_all
• sym
• syms
• tbl
• tbl_df
• tbl_nongroup_vars
• tbl_ptype
• top_frac
• transmute_all
• transmute_at
• transmute_if
• type_sum
• validate_grouped_df
• validate_rowwise_df
• wrap_dbplyr_obj

[krlmlr]

Thanks, Davis, for the extensive review.

I went through the existing reexports and identified the following obvious candidates for removal:

• auto_copy()
• do()
• dplyr_col_modify()
• dplyr_reconstruct()
• dplyr_row_slice()
• funs()
• same_src()
• vars()

The following are experimental, not sure:

• group_map()
• group_modify()
• group_walk()

In addition, there is a slew of functions related to grouping. At this stage, duckplyr does not support group_by() . Should it? Removal of the reexports might depend on if we think this should be implemented. Is there a way to do rowwise operations with the new .by argument? Should this be mentioned in ?dplyr::dplyr_by ?

• group_by()
• group_data()
• group_indices()
• group_keys()
• group_rows()
• group_size()
• group_vars()
• groups()
• is_grouped_df()
• n_groups()
• rowwise()

[DavisVaughan]

Removing those "obvious candidates" and anything related to groups_*() seems fine to me if duckplyr supports and prefers .by. Otherwise it is kind of confusing that group_by() "falls back" but .by doesn't.

For rowwise(), we don't yet support that through .by, but it would likely be a separate mutually exclusive argument like .by_row = FALSE/TRUE so we can keep .by "pure" tidyselect, see tidyverse/dplyr#6660

[DavisVaughan]

all_equal deprecated

[DavisVaughan]

Do you need all these here for testing?

[krlmlr]

Yes, for dplyr's tests converted to duckplyr.