Look at {tidytable}

filter() by group.

.before= and .after= for mutate().

pivot_longer() with multiple columns.

fquantile() and frange() operate on vectors only, but it would be useful to get a function that returns only one quantile in the fast stat functions. Also fcumsum(), flag(), ... => make a category for these additional functions.

collect_dtx(): do not use collect, but as_tibble() and the like that automatically ungroup (currently, we collect and ungroup then). Should we???

Deprecate the sXXX() speedy functions in favor of the XXX_() functions..

Make the XXX_() functions aware of dbplyr, dtplyr, tsibble, DuckDB and polars (and sf?).

Implement {dplyr} verbs: auto_copy(), case_match(), case_when(), coalesce(), cross_join(), c_across(), nest_join(), relocate(), rowwise(), slice_sample(), slice_max(), slice_min(). Also across(), all_vars(), any_vars(), between(), consecutive_id() == data.table::rleid(), cumall(), cumany(), cummean(), cume_dist(), dense_rank(), first(), group_cols(), group_map(), group_modify(), group_walk(), group_trim(), ident(), if_any(), if_all(), if_else(), intersect(), lag(), last(), lead(), min_rank(), n_distinct(), na_if(), near(), nth(), ntile(), order_by(), percent_rank(), pick(), recode(), recode_factor(), row_number(), rows_append(), rows_delete(), rows_insert(), rows_patch(), rows_update(), rows_upsert(), setdiff(), setequal(), sql(), symdiff(), union(), union_all(), vars(). ?context: n(), cur_group(), cur_group_id(), cur_group_rows(), cur_column(). ?scoped _if, _at, _all (superseded by across() and pick()). ?select (everything(), last_col(), group_cols(), starts_with(), ends_with(), contains() matches(), num_range(), all_of(), any_of(), where()). No? compute(), collect() and collapse() + copy_to(), explain(), show_query(), glimpse() == pillar::glimpse(), tbl(), is.tbl()? See also ?dplyr_extending, including dplyr_reconstruct(), dplyr_row_slice() and dplyr_col_modify().

Implement {tidyr} verbs that are not done yet: extract(), separate_rows(), nest_join(), chop(), unchop(), nest(), unnest(), unnest_longer(), unnest_wider(), hoist(), pack(), unpack().

Check package dependencies and make sure that the speedy and fast functions can be used in a package with minimal dependencies ({data.table} and {collapse} are OK, but {dplyr} or {tidyr} are not).

Should rownames() and rownames<-() get and assign to .rownames for data/table and tbl_df objects? Should we do something similar for keys in data.table or tbl_df?

Display more information in the print() methods of data.frame, data.table and tbl_df: help page, comment, labels units, ... See: vignette("extending", package = "pillar"). Also indicate key for data.table object.