.keep = "transmute" is implemented in mutate.

[apalacio9502]

Hi @hadley and @DavisVaughan,

This merge request aims to align the mutate functionality with that of transmute. Presently, the documentation states that the equivalent of transmute is mutate(.keep="none"). However, it's important to note that this function retains the column order, unlike transmute.

After reviewing the bug reported here: #6861, I concur that there should be a fifth variant of keep, named "transmute."

This proposal should also be implemented in the following tidyverse libraries:

• dtplyr
• dbplyr: mutate() now supports the argument .keep = "transmute" dbplyr#1508

And in the following third party libraries:

• duckplyr: mutate() now supports the argument .keep = "transmute" duckplyr#176
• arrow
• tidypolars

This change opens the possibility of un-superseded transmute. For example, you could implement that when a user calls transmute, it internally executes mutate(.keep="transmute"). This would greatly benefit veteran users who have extensively used transmute in their projects. Additionally, it improves code readability, as a simple transmute call is clearer than mutate(.keep="transmute").

Below I leave an example comparing transmute with mutate(.keep = "transmute")

example <- tibble(
  group = c("a", "a", "b", "b"),
  old = c(1, 2, 3, 4)
)

# transmute --------------------------------------------------------------------

example %>% transmute(
  group,
  new = old * 2,
  old
)

# A tibble: 4 × 3
  group   new   old
  <chr> <dbl> <dbl>
1 a         2     1
2 a         4     2
3 b         6     3
4 b         8     4

# mutate(.keep = "transmute") --------------------------------------------------

example %>% mutate(
  .keep = "transmute",
  group,
  new = old * 2,
  old
)

# A tibble: 4 × 3
  group   new   old
  <chr> <dbl> <dbl>
1 a         2     1
2 a         4     2
3 b         6     3
4 b         8     4

# group_by + transmute ---------------------------------------------------------

example %>% 
  group_by(group) %>% 
  transmute(
    new = sum(old),
    old
) %>% 
 ungroup()

# A tibble: 4 × 3
  group   new   old
  <chr> <dbl> <dbl>
1 a         3     1
2 a         3     2
3 b         7     3
4 b         7     4

# group_by + mutate(.keep = "transmute") ---------------------------------------

example %>% 
  group_by(group) %>% 
  mutate(
    .keep = "transmute",
    new = sum(old),
    old
) %>% 
 ungroup()

# A tibble: 4 × 3
  group   new   old
  <chr> <dbl> <dbl>
1 a         3     1
2 a         3     2
3 b         7     3
4 b         7     4

# mutate(.keep = "transmute", by = ...) ----------------------------------------

example %>%  
  mutate(
    .keep = "transmute",
    .by = "group",
    new = sum(old),
    old
  )

# A tibble: 4 × 3
  group   new   old
  <chr> <dbl> <dbl>
1 a         3     1
2 a         3     2
3 b         7     3
4 b         7     4

Regards,

[DavisVaughan]

We may simply un-supercede transmute() since many people seem to like it and have pushed back on that, in which case we probably would not add this because it does have very different properties than the theoretically consistent set of existing .keep options. I am not sure yet though, we will need to have an internal discussion about it and we aren't currently doing a large amount of dplyr dev at the moment.

[apalacio9502]

Hi @DavisVaughan,

Thank you for your response and I appreciate your perspective.

From my standpoint as a user, I think it would be beneficial to consider the following alternative.

• Incorporating the parameter mutate(.keep = "transmute"): I am convinced that adding this parameter would enrich the process, providing greater flexibility. This would allow for quick adaptation of the outcome through the option .keep = c("all", "used", "unused", "none", "transmute"). If we only opt for "un-superseded transmute" and a user wishes to transform a "mutate" into a "transmute", it would entail changing the function name and removing parameters that are not compatible with "transmute".

• Un-superseded transmute: This would be valuable for those users who prefer using transmute due to its clarity in the code, or for those who encounter difficulties when migrating to mutate(.keep = "transmute"), as it involves more than just a direct replacement (for example, if one wishes to add .keep at the end of the code or adopt a particular style). Additionally, I believe it would be interesting to include the parameter .by in transmute, which would add value for users.

I look forward to your comments.

Regards,

[DavisVaughan]

Additionally, I believe it would be interesting to include the parameter .by in transmute, which would add value for users.

If we unsupercede, definitely!