Introduce line-breaks in multiline arguments

[IndrajeetPatil]

cf. tidyverse/style#209

styler::style_text("
list(
  x, y,
  c(
    a
  ), c(
    b
  )
)
")
#> list(
#>   x, y,
#>   c(
#>     a
#>   ),
#>   c(
#>     b
#>   )
#> )

^{Created on 2024-12-20 with reprex v2.1.1}

[IndrajeetPatil]

Wow, that's quite the slowdown. Need to figure out what's causing it once the scope is clear.

[IndrajeetPatil]

@lorenzwalthert Any thoughts?

I am blocked here without any further feedback from the team.

[lorenzwalthert]

Maybe we should talk about the abstract rule first. What is it exactly that we try to achieve?
As for the code, I just wanted to say that you can easily figure out if there are line-breaks in a nest by looking at the respective column. To figure out if there are line breaks hidden in one of the children, consier the is_multi_line column. That will save you a lot of computations I think.

Also important: in the example you gave, all line breaks should simply be removed if we had #247.

[IndrajeetPatil]

What is it exactly that we try to achieve?

This is where I need some clarity.

I am interpreting the discussion in tidyverse/style#209 to mean that the style guide recommends "one line per argument" rule.

Elsewhere, the style guide carves out an exception to put multiple arguments per line, but the intent of clubbing similar arguments is not something {styler} can understand:

Therefore, I opted to just always follow the stated rule and if users wish to retain multiple arguments per like, they will just have to add ignore directives.

[lorenzwalthert]

as it is evident from the test cases, enforcing one argument per line is quite a change to the current behaviour I'd discourage this, at least at this point. Also, sometimes you want more than one argument per line when named arguments are involved:

map(x, f
  y = 2
)

So I suggest to just focus on a small interesting scope first and then maybe extend it gradually. And in my view, a small improvement involves the focus of multi-line arguments and it is to enforce a line-break between two multi-line arguments using the is_multi_line argument.

c(
 c(
  'b'
 ), fun(
  f = 2
 )
)
# becomes 
c(
 c(
  'b'
 ), 
 fun(
  f = 2
 )
)

Or even slightly more rigid, each multi-line expression has to be on its own line:

c(
 a, c(
  'b'
 ), fun(
  f = 2
 )
)
# becomes 
c(
 a, 
 c(
  'b'
 ), 
 fun(
  f = 2
 )
)

Comments should probably be ignored in that logic, i.e. token after or before comma should be defined as in skip the comments. Do you want to adapt your PR to that effect? I can also tell you how I'd solve the puzzle in abstract terms first, but I don't want to take away the fun of figuring it out, so you can go ahead if you want.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 7525889 is merged into main:

• 🚀cache_applying: 162ms -> 159ms [-4.02%, -0.08%]
• ❗🐌cache_recording: 534ms -> 1.09s [+103.17%, +105.7%]
• ❗🐌without_cache: 1s -> 2.39s [+137.01%, +139.91%]

Further explanation regarding interpretation and methodology can be found in the documentation.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if c624bfc is merged into main:

• ✔️cache_applying: 189ms -> 186ms [-6.59%, +3.31%]
• ❗🐌cache_recording: 575ms -> 1.25s [+112.43%, +121.28%]
• ❗🐌without_cache: 1.07s -> 2.79s [+156.09%, +163.95%]

Further explanation regarding interpretation and methodology can be found in the documentation.