Compatibility with air + housekeeping#1291
Conversation
lorenzwalthert
changed the title
|
This is how benchmark results would change (along with a 95% confidence interval in relative change) if 3becceb is merged into main:
Further explanation regarding interpretation and methodology can be found in the documentation. |
|
@randy3k recently ran a similar comparison internally -- care to share if anything further was found there? |
|
This is how benchmark results would change (along with a 95% confidence interval in relative change) if 2351f15 is merged into main:
Further explanation regarding interpretation and methodology can be found in the documentation. |
|
This is how benchmark results would change (along with a 95% confidence interval in relative change) if aa8803f is merged into main:
Further explanation regarding interpretation and methodology can be found in the documentation. |
|
For map(x, f,
arg = 1,
arg = 2
)
# ->
map(
x,
f,
arg = 1,
arg = 2
)We really try to limit the forms a function call can be in. There are really only 2: # "Flat" (all args on same line)
map(x, f, arg = 1, arg = 2)
# "Expanded" (all args broken across lines)
map(
x,
f,
arg = 1,
arg = 2
)With one exception for trailing anonymous functions, i.e. common # Allowed
map(x, function(x) {
# do the thing
})This has generally proved to be pretty useful in keeping complexity down, especially across intermingled Air features This also comes into play with # air will expand this
switch(x,
a = 2,
y = 3
)posit-dev/air#199 would fix part of other_trigger_on_same_line <- (pd$token[
remaining_row_idx_between_trigger_and_line_break
] %in%
other_trigger_tokens)posit-dev/air#114 would be opt in but would probably help with style_guide_name <- "styler::tidyverse_style@https://github.com/r-lib"
create_style_guide(
# transformer functions
initialize = default_style_guide_attributes,
line_break = line_break_manipulators,
space = space_manipulators,
token = token_manipulators,
indention = indention_manipulators,
...
) |
|
Thanks for following up on this @DavisVaughan
|
|
Some more interesting cases: mask <- c(
1L, 2L, if (roxygen[3L] == "\n") 3L, last(which(roxygen == "}"))
) %>% sort()
call(
1, 2, 3, 4
)Both are expanded, despite not maxing out the character width limit. In my opinion this consumes too much vertical space and is not optimal: mask <- c(
1L,
2L,
if (roxygen[3L] == "\n") 3L,
last(which(roxygen == "}"))
) %>%
sort()
call(
1,
2,
3,
4
)In particular in the second case, one could argue this matches @DavisVaughan's description |
|
Another difference is that ignoring happens per line in {styler}, air ignores by expression. Since air lacks ignoring ranges, this comes in handy if the range you want to ignore coincides with a multi-line expression like a function call. Also, to ignore the subsequent expression, air places the marker above the expression to ignore, {styler} on the same line, and to exclude a range in {styler}, you put the marker above the first line to ignore, while air interprets that marker position as a marker to just ignore one expression. I was initially inclined to set the marker text options to include both {styler}'s and airs marker text, but given that they work differently, it would probably just create more confusion. |
|
This is how benchmark results would change (along with a 95% confidence interval in relative change) if cb2404d is merged into main:
Further explanation regarding interpretation and methodology can be found in the documentation. |
3 participants
There are surprisingly few conflicts.Now, people can format this repo with air and {styler}, with the same output. {styler} is far less invasive than air, so formatting with air once means a lot of code is not changed by {styler} on subsequent styling.Bug in air: Air breaks code with a variable called else_* posit-dev/air#499 caused a re-naming for convenience here.
multi-line assignment is handled differently by {styler} than air, this can be circumvented by making sure the first line break in the statement does not occur directly after
<-.{styler} forced no line break after first argument in
switch(). This was relaxed for compatibility with air.We used
fmt: skipinswitch()statements to make the pre-commit hook pass.air does not respect the tidyverse rules which allow to keep unnamed argument on the first lineThis rule seems removed. it turns:Into
Which I think is not optimal.
Another case is
becomes (with air)
Where I also think {styler} styles superior.
It could not preserve alignment in the blow snippet
it broke a lot of long lines, which I think is desirable. This increased the line count of the source code.
Only single lines can be ignored, no ranges: Support skip ranges posit-dev/air#456.