Add LineSuffix reserved width
#6830
Conversation
PR Check ResultsBenchmarkLinuxWindows |
cnpryer
changed the title
LineSuffixLineSuffix
cnpryer
changed the title
LineSuffixLineSuffix reserved width
cnpryer
changed the title
LineSuffix reserved widthLineSuffix reserved width
|
This is exciting. Thanks for taking the time to work on this! |
|
I started updating the PR description to reflect feedback (see todo list). As discussed on Discord we can also split this PR up if needed in order to unblock dependent work. |
cnpryer
force-pushed
the
line-suffix
branch
6 times, most recently
from
eddc725
to
686f3da
Compare
|
Was looking at the ecosystem checks. This is unstable atm. Pulled from the transformers repo: class BaseMixedInt8Test(unittest.TestCase):
# ...
EXPECTED_RELATIVE_DIFFERENCE = (
1.540025 # This was obtained on a Quadro RTX 8000 so the number might slightly change
)From django the same instability occurs with: class Field:
hidden_widget = (
HiddenInput # Default widget to use when rendering this as "hidden".
)Would this imply that FWIW I wanted to just see if this would fix it and it does. diff --git a/crates/ruff_python_formatter/src/expression/mod.rs b/crates/ruff_python_formatter/src/expression/mod.rs
index 52255ec4f..89121e38d 100644
--- a/crates/ruff_python_formatter/src/expression/mod.rs
+++ b/crates/ruff_python_formatter/src/expression/mod.rs
@@ -183,11 +183,12 @@ impl Format<PyFormatContext<'_>> for MaybeParenthesizeExpression<'_> {
} = self;
let comments = f.context().comments();
- let preserve_parentheses = parenthesize.is_optional()
- && is_expression_parenthesized((*expression).into(), f.context().source());
+ let has_parentheses = is_expression_parenthesized((*expression).into(), f.context().source());
+ let preserve_parentheses = parenthesize.is_optional() && has_parentheses;
- let has_comments =
- comments.has_leading(*expression) || comments.has_trailing_own_line(*expression);
+ let has_comments = comments.has_leading(*expression)
+ || comments.has_trailing_own_line(*expression)
+ || (has_parentheses && comments.has_trailing(*expression));Maybe that's #6771? |
It seems to me that Black removes the parentheses always. This is a file that I used to play around with the formatting. class BaseMixedInt8Test(unittest.TestCase):
# ...
EXPECTED_RELATIVE_DIFFERENCE = (
1.540025 # This was obtained on a Quadro RTX 8000 so the number might slightly change
)
EXPECTED_RELATIVE_DIFFERENCE = 1.540025 # This was obtained on a Quadro RTX 8000 so the number might slightly change
EXPECTED_RELATIVE_DIFFERENCE = (
1.540025
) # This was obtained on a Quadro RTX 8000 so the number might slightly change
# ...
EXPECTED_RELATIVE_DIFFERENCE = (
1.5 # This was obtained on a Quadro RTX 8000 so the number might slightly change
)
EXPECTED_RELATIVE_DIFFERENCE = 1.5 # This was obtained on a Quadro RTX 8000 so the number might slightly change
EXPECTED_RELATIVE_DIFFERENCE = (
1.5
) # This was obtained on a Quadro RTX 8000 so the number might slightly change
# ...
EXPECTED_RELATIVE_DIFFERENCE = (
1.540025 # This was
)
EXPECTED_RELATIVE_DIFFERENCE = 1.540025 # This was
EXPECTED_RELATIVE_DIFFERENCE = (
1.540025
) # This was
(
EXPECTED_RELATIVE_DIFFERENCE # This was obtained on a Quadro RTX 8000 so the number might slightly change
)= (
1.540025
) # This wasI haven't come to a conclusion yet what the best behavior is but we have a few options:
We'll need to play around with this a little more. |
CodSpeed Performance ReportMerging #6830 will not alter performanceComparing Summary
|
|
Ah I see. I think I was too focused on dealing with the instability. You're right. So is it specific nodes that are given the "break if line suffix exceeds line-lengths" treatment? On first glance it looks like it occurs where some kind of parentheses are present. When I say parentheses I mean Marking as ready to invite some discussion around the solution. I'll continue tinkering. |
|
I've decided to split this into two PRs. One (this one) will be for adding the new I wanted to do this because the |
Closes #5630
Summary
Currently a trailing, end-of-line comment considered a line suffix element is not measured during formatting. This change introduces a reserved width field to the
ruff_formattercrate for line suffix elements in order to allow formatter clients to opt-into this behavior.Test Plan
Currently just doctest
Outdated; See (comment)
TODO:
LineSuffixreserved width (comment)TODO(optional):
TextWidthinstead of internalLineSuffix(comment)