Fix lambda formatting in interpolated string expressions by ntBre · Pull Request #25144 · astral-sh/ruff

ntBre · 2026-05-13T17:18:59Z

Summary

This PR fixes #24807 by making RemoveSoftLinesBuffer aware of
BestFitting and always selecting the most-flat variant, as suggested
in Micha's comment, assuming I interpreted it correctly.

As I noted on the issue, I was hoping I could get away with making a
local change to the lambda formatting, but this turned out to affect
the two other uses of RemoveSoftLinesBuffer too.

The formatting of the new regression tests:

def foo():
    subprocess.check_call(f"rm -rf {' '.join(map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove))}", shell=True)

def bar():
    lambda x=" ".join(
        map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove)
    ): ...

value = f"prefix {
    ' '.join(map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove))
} suffix"

now (nearly) matches the output before #21385:

❯ git diff <(uvx [email protected] format - < fmt.py) <(./target/debug/ruff format - < fmt.py)
 def bar():
-    lambda x=" ".join(
-        map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove)
-    ): ...
+    lambda x=" ".join(map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove)): (
+        ...
+    )

 value = f"prefix {

with the only difference being that the body of this second lambda
still wraps in the new formatting.

Test Plan

New tests based on the reported issue and possibly the ecosystem check
on this PR

Summary -- This PR fixes #24807 by making `RemoveSoftLinesBuffer` aware of `BestFitting` and always selecting the most-flat variant, as suggested in [Micha's comment], assuming I interpreted it correctly. As I noted on the issue, I was hoping I could get away with making a local change to the lambda formatting, but this turned out to affect the two other uses of `RemoveSoftLinesBuffer` too. The formatting of the new regression tests: ```py def foo(): subprocess.check_call(f"rm -rf {' '.join(map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove))}", shell=True) def bar(): lambda x=" ".join( map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove) ): ... value = f"prefix { ' '.join(map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove)) } suffix" ``` now (nearly) matches the output before #21385: ```console ❯ git diff <(uvx [email protected] format - < fmt.py) <(./target/debug/ruff format - < fmt.py) def bar(): - lambda x=" ".join( - map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove) - ): ... + lambda x=" ".join(map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove)): ( + ... + ) value = f"prefix { ``` with the only difference being that the body of this second lambda still wraps in the new formatting. Test Plan -- New tests based on the reported issue and possibly the ecosystem check on this PR [Micha's comment]: #24807 (comment)

astral-sh-bot · 2026-05-13T17:28:44Z

`ruff-ecosystem` results

Formatter (stable)

✅ ecosystem check detected no format changes.

Formatter (preview)

✅ ecosystem check detected no format changes.

ntBre · 2026-05-13T20:24:03Z

It seems that my first pass wasn't quite sufficient. Codex came up with an even more contrived example that demonstrated the missing handling for BestFitting elements inside Interned elements. I tried a couple of fixes that didn't feel very satisfactory before realizing most of this code originated in Biome and thinking to reference their current version of the code:

https://github.com/biomejs/biome/blob/05c26176573534a0abfa92d454d244f9569bc77d/crates/biome_formatter/src/buffer.rs#L597-L605

It looks like they ran into something similar, and I was able to backport enough of their code to get the recursive case working. I wonder if we need to adapt any of the other changes they've made too?

…5144) Summary -- This PR fixes astral-sh#24807 by making `RemoveSoftLinesBuffer` aware of `BestFitting` and always selecting the most-flat variant, as suggested in [Micha's comment], assuming I interpreted it correctly. As I noted on the issue, I was hoping I could get away with making a local change to the lambda formatting, but this turned out to affect the two other uses of `RemoveSoftLinesBuffer` too. The formatting of the new regression tests: ```py def foo(): subprocess.check_call(f"rm -rf {' '.join(map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove))}", shell=True) def bar(): lambda x=" ".join( map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove) ): ... value = f"prefix { ' '.join(map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove)) } suffix" ``` now (nearly) matches the output before astral-sh#21385: ```console ❯ git diff <(uvx [email protected] format - < fmt.py) <(./target/debug/ruff format - < fmt.py) def bar(): - lambda x=" ".join( - map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove) - ): ... + lambda x=" ".join(map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove)): ( + ... + ) value = f"prefix { ``` with the only difference being that the body of this second lambda still wraps in the new formatting. Test Plan -- New tests based on the reported issue and possibly the ecosystem check on this PR [Micha's comment]: astral-sh#24807 (comment)

Summary -- This PR fixes #25397 by dropping `StartBestFittingEntry` and `EndBestFittingEntry` tags in `RemoveSoftLinesBuffer`. #25144 changed `RemoveSoftLinesBuffer` to inline the most-flat variant of `BestFitting` elements, but it inlined them with their start and end tags, which caused the problem observed in #25397 when these tags became nested in another `BestFitting` element, which interpreted the first `End` tag as ending the outer `BestFitting` rather than the nested one. You can see this clearly in the IR. First, the diff between the 0.15.13 and 0.15.14 IR, which reveals the unmatched start tags: <details> <summary>0.15.13 vs main</summary> ```diff @@ -38,29 +38,13 @@ "lambda ", group(["arg"]), ": ", - best_fitting([ - [ - [ - <interned 1> [ - "str(", - group([ - indent([soft_line_break, group(["arg"])]), - soft_line_break - ]), - ")" - ] - ] - ] - [[group(expand: true, [<ref interned *1>])]] - [ - [ - "(", - indent([hard_line_break, <ref interned *1>]), - hard_line_break, - ")" - ] + [ + <interned 1> [ + "str(", + group([indent([group(["arg"])])]), + ")" ] - ]), + ], ", self.args" ]) ]) @@ -105,27 +89,9 @@ "lambda ", group(["arg"]), ": ", - best_fitting([ - [ - [ - <interned 2> [ - "str(", - group([ - indent([soft_line_break, group(["arg"])]), - soft_line_break - ]), - ")" - ] - ] - ] - [[group(expand: true, [<ref interned *2>])]] - [ - [ - "(", - indent([hard_line_break, <ref interned *2>]), - hard_line_break, - ")" - ] + [ + <interned 2> [ + "str(", + group([indent([group(["arg"])])]), + ")" ] - ]), - ", self.args" + ] ]) + <START_WITHOUT_END<Group>> ]) - ]), - ")" + <START_WITHOUT_END<Indent>> + ]) + <START_WITHOUT_END<Group>> ]) + <START_WITHOUT_END<Group>> ]) - ]), - ")})\"" - ]), - hard_line_break, - ")" + <START_WITHOUT_END<Indent>> + ]) + <START_WITHOUT_END<Group>> + ]) + <START_WITHOUT_END<Indent>> ]) - ] + <START_WITHOUT_END<Group>> + ]) + <START_WITHOUT_END<BestFittingEntry>> ] ]) ]), ``` </details> and then the diff between main and this PR, which shows the removed brackets that represent the nested tags: <details> <summary>main vs PR</summary> ```diff @@ -38,12 +38,10 @@ "lambda ", group(["arg"]), ": ", - [ - <interned 1> [ - "str(", - group([indent([group(["arg"])])]), - ")" - ] + <interned 1> [ + "str(", + group([indent([group(["arg"])])]), + ")" ], ", self.args" ]) @@ -89,24 +87,15 @@ "lambda ", group(["arg"]), ": ", - [ - <interned 2> [ - "str(", - group([indent([group(["arg"])])]), - ")" - ] - ] + <interned 2> [ + "str(", + group([indent([group(["arg"])])]), + ")" + ], + ", self.args" ]) - <START_WITHOUT_END<Group>> ]) - <START_WITHOUT_END<Indent>> - ]) - <START_WITHOUT_END<Group>> + ]), + ")" ]) - <START_WITHOUT_END<Group>> ]) - <START_WITHOUT_END<Indent>> - ]) - <START_WITHOUT_END<Group>> - ]) - <START_WITHOUT_END<Indent>> + ]), + ")})\"" + ]), + hard_line_break, + ")" ]) - <START_WITHOUT_END<Group>> - ]) - <START_WITHOUT_END<BestFittingEntry>> + ] ] ]) ]), ``` </details> I think the diff between 0.15.13 and this PR is also nice to see because it makes it a bit more clear that the `BestFitting` element has been inlined: <details> <summary>0.15.13 vs this PR</summary> ```diff @@ -38,29 +38,11 @@ "lambda ", group(["arg"]), ": ", - best_fitting([ - [ - [ - <interned 1> [ - "str(", - group([ - indent([soft_line_break, group(["arg"])]), - soft_line_break - ]), - ")" - ] - ] - ] - [[group(expand: true, [<ref interned *1>])]] - [ - [ - "(", - indent([hard_line_break, <ref interned *1>]), - hard_line_break, - ")" - ] - ] - ]), + <interned 1> [ + "str(", + group([indent([group(["arg"])])]), + ")" + ], ", self.args" ]) ]) @@ -105,29 +87,11 @@ "lambda ", group(["arg"]), ": ", - best_fitting([ - [ - [ - <interned 2> [ - "str(", - group([ - indent([soft_line_break, group(["arg"])]), - soft_line_break - ]), - ")" - ] - ] - ] - [[group(expand: true, [<ref interned *2>])]] - [ - [ - "(", - indent([hard_line_break, <ref interned *2>]), - hard_line_break, - ")" - ] - ] - ]), + <interned 2> [ + "str(", + group([indent([group(["arg"])])]), + ")" + ], ", self.args" ]) ]) @@ -155,7 +119,6 @@ def __str__(self) -> str: return ( - f"first stringlist of args long={list(map(lambda arg: str( - arg - ), self.args))})" + "first string" + f"list of args long={list(map(lambda arg: str(arg), self.args))})" ) ``` </details> And you can also see at the bottom here that the actual formatting that #25144 sought to resolve is still fixed. Test Plan -- New regression test based on the code in the issue

…5144) Summary -- This PR fixes astral-sh#24807 by making `RemoveSoftLinesBuffer` aware of `BestFitting` and always selecting the most-flat variant, as suggested in [Micha's comment], assuming I interpreted it correctly. As I noted on the issue, I was hoping I could get away with making a local change to the lambda formatting, but this turned out to affect the two other uses of `RemoveSoftLinesBuffer` too. The formatting of the new regression tests: ```py def foo(): subprocess.check_call(f"rm -rf {' '.join(map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove))}", shell=True) def bar(): lambda x=" ".join( map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove) ): ... value = f"prefix { ' '.join(map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove)) } suffix" ``` now (nearly) matches the output before astral-sh#21385: ```console ❯ git diff <(uvx [email protected] format - < fmt.py) <(./target/debug/ruff format - < fmt.py) def bar(): - lambda x=" ".join( - map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove) - ): ... + lambda x=" ".join(map(lambda object_name: os.path.join(dest_path, object_name), objects_to_remove)): ( + ... + ) value = f"prefix { ``` with the only difference being that the body of this second lambda still wraps in the new formatting. Test Plan -- New tests based on the reported issue and possibly the ecosystem check on this PR [Micha's comment]: astral-sh#24807 (comment)

…sh#25398) Summary -- This PR fixes astral-sh#25397 by dropping `StartBestFittingEntry` and `EndBestFittingEntry` tags in `RemoveSoftLinesBuffer`. astral-sh#25144 changed `RemoveSoftLinesBuffer` to inline the most-flat variant of `BestFitting` elements, but it inlined them with their start and end tags, which caused the problem observed in astral-sh#25397 when these tags became nested in another `BestFitting` element, which interpreted the first `End` tag as ending the outer `BestFitting` rather than the nested one. You can see this clearly in the IR. First, the diff between the 0.15.13 and 0.15.14 IR, which reveals the unmatched start tags: <details> <summary>0.15.13 vs main</summary> ```diff @@ -38,29 +38,13 @@ "lambda ", group(["arg"]), ": ", - best_fitting([ - [ - [ - <interned 1> [ - "str(", - group([ - indent([soft_line_break, group(["arg"])]), - soft_line_break - ]), - ")" - ] - ] - ] - [[group(expand: true, [<ref interned *1>])]] - [ - [ - "(", - indent([hard_line_break, <ref interned *1>]), - hard_line_break, - ")" - ] + [ + <interned 1> [ + "str(", + group([indent([group(["arg"])])]), + ")" ] - ]), + ], ", self.args" ]) ]) @@ -105,27 +89,9 @@ "lambda ", group(["arg"]), ": ", - best_fitting([ - [ - [ - <interned 2> [ - "str(", - group([ - indent([soft_line_break, group(["arg"])]), - soft_line_break - ]), - ")" - ] - ] - ] - [[group(expand: true, [<ref interned *2>])]] - [ - [ - "(", - indent([hard_line_break, <ref interned *2>]), - hard_line_break, - ")" - ] + [ + <interned 2> [ + "str(", + group([indent([group(["arg"])])]), + ")" ] - ]), - ", self.args" + ] ]) + <START_WITHOUT_END<Group>> ]) - ]), - ")" + <START_WITHOUT_END<Indent>> + ]) + <START_WITHOUT_END<Group>> ]) + <START_WITHOUT_END<Group>> ]) - ]), - ")})\"" - ]), - hard_line_break, - ")" + <START_WITHOUT_END<Indent>> + ]) + <START_WITHOUT_END<Group>> + ]) + <START_WITHOUT_END<Indent>> ]) - ] + <START_WITHOUT_END<Group>> + ]) + <START_WITHOUT_END<BestFittingEntry>> ] ]) ]), ``` </details> and then the diff between main and this PR, which shows the removed brackets that represent the nested tags: <details> <summary>main vs PR</summary> ```diff @@ -38,12 +38,10 @@ "lambda ", group(["arg"]), ": ", - [ - <interned 1> [ - "str(", - group([indent([group(["arg"])])]), - ")" - ] + <interned 1> [ + "str(", + group([indent([group(["arg"])])]), + ")" ], ", self.args" ]) @@ -89,24 +87,15 @@ "lambda ", group(["arg"]), ": ", - [ - <interned 2> [ - "str(", - group([indent([group(["arg"])])]), - ")" - ] - ] + <interned 2> [ + "str(", + group([indent([group(["arg"])])]), + ")" + ], + ", self.args" ]) - <START_WITHOUT_END<Group>> ]) - <START_WITHOUT_END<Indent>> - ]) - <START_WITHOUT_END<Group>> + ]), + ")" ]) - <START_WITHOUT_END<Group>> ]) - <START_WITHOUT_END<Indent>> - ]) - <START_WITHOUT_END<Group>> - ]) - <START_WITHOUT_END<Indent>> + ]), + ")})\"" + ]), + hard_line_break, + ")" ]) - <START_WITHOUT_END<Group>> - ]) - <START_WITHOUT_END<BestFittingEntry>> + ] ] ]) ]), ``` </details> I think the diff between 0.15.13 and this PR is also nice to see because it makes it a bit more clear that the `BestFitting` element has been inlined: <details> <summary>0.15.13 vs this PR</summary> ```diff @@ -38,29 +38,11 @@ "lambda ", group(["arg"]), ": ", - best_fitting([ - [ - [ - <interned 1> [ - "str(", - group([ - indent([soft_line_break, group(["arg"])]), - soft_line_break - ]), - ")" - ] - ] - ] - [[group(expand: true, [<ref interned *1>])]] - [ - [ - "(", - indent([hard_line_break, <ref interned *1>]), - hard_line_break, - ")" - ] - ] - ]), + <interned 1> [ + "str(", + group([indent([group(["arg"])])]), + ")" + ], ", self.args" ]) ]) @@ -105,29 +87,11 @@ "lambda ", group(["arg"]), ": ", - best_fitting([ - [ - [ - <interned 2> [ - "str(", - group([ - indent([soft_line_break, group(["arg"])]), - soft_line_break - ]), - ")" - ] - ] - ] - [[group(expand: true, [<ref interned *2>])]] - [ - [ - "(", - indent([hard_line_break, <ref interned *2>]), - hard_line_break, - ")" - ] - ] - ]), + <interned 2> [ + "str(", + group([indent([group(["arg"])])]), + ")" + ], ", self.args" ]) ]) @@ -155,7 +119,6 @@ def __str__(self) -> str: return ( - f"first stringlist of args long={list(map(lambda arg: str( - arg - ), self.args))})" + "first string" + f"list of args long={list(map(lambda arg: str(arg), self.args))})" ) ``` </details> And you can also see at the bottom here that the actual formatting that astral-sh#25144 sought to resolve is still fixed. Test Plan -- New regression test based on the code in the issue

ntBre added 4 commits May 13, 2026 13:18

add unstable test case

5b890a3

choose most-flat variant for BestFitting in RemoveSoftLinesBuffer

4dfb57f

add regression tests for other RemoveSoftLinesBuffer uses

7360277

ntBre added bug Something isn't working formatter Related to the formatter labels May 13, 2026

ntBre added 2 commits May 13, 2026 13:54

add an unstable case for nested, interned elements

36f1215

port biome's updated BestFitting handling back to ruff

c29550b

ntBre marked this pull request as ready for review May 13, 2026 21:03

ntBre requested a review from MichaReiser as a code owner May 13, 2026 21:03

MichaReiser approved these changes May 15, 2026

View reviewed changes

Comment thread crates/ruff_formatter/src/buffer.rs

update comment

99a2ca3

ntBre enabled auto-merge (squash) May 15, 2026 13:40

ntBre merged commit 25af028 into main May 15, 2026
44 checks passed

ntBre deleted the brent/lambda-f-string-formatting-bug branch May 15, 2026 13:44

BrewTestBot mentioned this pull request May 21, 2026

ruff 0.15.14 Homebrew/homebrew-core#284070

Merged

This was referenced May 26, 2026

Can't format valid python file: "Invalid document: Expected end tag of kind Group but found Indent." #25397

Closed

Fix invalid IR when flattening nested BestFitting elements #25398

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix lambda formatting in interpolated string expressions#25144

Fix lambda formatting in interpolated string expressions#25144
ntBre merged 7 commits into
mainfrom
brent/lambda-f-string-formatting-bug

ntBre commented May 13, 2026

Uh oh!

astral-sh-bot Bot commented May 13, 2026 •

edited

Loading

Uh oh!

ntBre commented May 13, 2026

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

ntBre commented May 13, 2026

Summary

Test Plan

Uh oh!

astral-sh-bot Bot commented May 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

ruff-ecosystem results

Formatter (stable)

Formatter (preview)

Uh oh!

ntBre commented May 13, 2026

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

astral-sh-bot Bot commented May 13, 2026 •

edited

Loading

`ruff-ecosystem` results