Skip to content

avoid quadratic node removal when converting fmt: off/on blocks - #5232

Merged
cobaltt7 merged 1 commit into
psf:mainfrom
sahvx655-wq:fmt-off-remove-quadratic
Jul 13, 2026
Merged

avoid quadratic node removal when converting fmt: off/on blocks#5232
cobaltt7 merged 1 commit into
psf:mainfrom
sahvx655-wq:fmt-off-remove-quadratic

Conversation

@sahvx655-wq

@sahvx655-wq sahvx655-wq commented Jul 12, 2026

Copy link
Copy Markdown
Contributor

Description

Profiling a file with many # fmt: off/# fmt: on regions interspersed with ordinary code showed Base.remove in blib2to3/pytree.py dominating and growing quadratically: the number of children scanned across all removals was 502k at 500 blocks, 2.0M at 1000 and 8.0M at 2000, quadrupling every time the block count doubled. The cause is that normalize_fmt_off converts each block to a STANDALONE_COMMENT through _handle_regular_fmt_block, which calls ignored.remove() on every node of the block, and remove locates the node by scanning its parent's children from index 0. Every top-level block shares one parent (file_input), which grows to O(N) children, so converting N blocks is O(N^2) work dominated by that scan.

Because the blocks are converted left to right, the position of each removal within a given parent only moves forward. I added an optional search_start hint to remove and thread a small per-parent position map through normalize_fmt_off/convert_one_fmt_off_pair/_handle_regular_fmt_block, so each removal resumes the scan from the previous conversion's index instead of restarting. The full children list is still searched (the scan wraps around modulo its length), so a stale hint only costs a little extra work and never changes which node is removed; children scanned drops to linear (3k/6k/12k for the same inputs). This is the inner counterpart to #5169, which removed the outer re-traversal of this same pass but left the per-node removal alone, and to #5213, which fixed the equivalent remove behaviour on the separate --line-ranges path. Output is byte-identical across the test suite (226 format cases), the fmtonoff data and the Black source tree in both stable and preview; left unfixed a small # fmt: off-heavy input burns quadratic CPU, and this pass runs before any formatting so it is reachable through unauthenticated blackd.

Checklist - did you ...

  • Implement any code style changes under the --preview style, following the stability policy?
  • Add an entry in CHANGES.md if necessary?
  • Add / update tests if necessary?
  • Add new / update outdated documentation?

@github-actions

github-actions Bot commented Jul 12, 2026

Copy link
Copy Markdown
Contributor

diff-shades results comparing this PR (1841fc1) to main (d7587ce):

--preview style: no changes

--stable style: no changes

What is this? | Workflow run | diff-shades documentation

@cobaltt7
cobaltt7 merged commit 513058f into psf:main Jul 13, 2026
57 checks passed
luketainton pushed a commit to luketainton/luke_instant-msg-api that referenced this pull request Oct 5, 2026
renovate[bot] @luketainton
chore(deps): update dependency black to <26.10.1,>=26.10.0 (#290)
5cf1d34 
This PR contains the following updates:

| Package | Change | [Age](https://docs.renovatebot.com/merge-confidence/) | [Confidence](https://docs.renovatebot.com/merge-confidence/) |
|---|---|---|---|
| [black](https://github.com/psf/black) ([changelog](https://github.com/psf/black/blob/main/CHANGES.md)) | `<26.5.2,>=26.5.1` → `<26.10.1,>=26.10.0` | ![age](https://developer.mend.io/api/mc/badges/age/pypi/black/26.10.0?slim=true) | ![confidence](https://developer.mend.io/api/mc/badges/confidence/pypi/black/26.5.1/26.10.0?slim=true) |

---

### Release Notes

<details>
<summary>psf/black (black)</summary>

### [`v26.10.0`](https://github.com/psf/black/blob/HEAD/CHANGES.md#Version-26100)

[Compare Source](psf/black@26.5.1...26.10.0)

##### Stable style

- `--line-ranges` no longer inserts an empty line after a docstring when the range
  covers only the docstring itself ([#&#8203;5312](psf/black#5312))
- Fix `# fmt: skip` on a bracketed ternary turning the surrounding tuple into a call
  ([#&#8203;5464](psf/black#5464))
- Fix crash when `# fmt: skip` is placed on a one-line function or class with PEP 695
  type parameters ([#&#8203;5429](psf/black#5429))
- Fix an inline comment after the closing bracket of optional parentheses being moved
  inside the parentheses when the parenthesized expression contains own-line comments
  ([#&#8203;5395](psf/black#5395))
- Fix unparseable output when `# fmt: skip` is placed on a bracket of an `if`, `while`,
  `for`, or `with` header ([#&#8203;5401](psf/black#5401), [#&#8203;5405](psf/black#5405))
- Fix crash when formatting parenthesized expressions with multiple inline comments and
  `# fmt: skip` ([#&#8203;5414](psf/black#5414))
- Fix parsing Jupyter notebook assignment magics when non-ASCII characters appear
  earlier on the line ([#&#8203;5381](psf/black#5381))
- Preserve blank lines that come immediately before a `# fmt: on` comment ([#&#8203;5300](psf/black#5300))
- Keep the parentheses around the target of an annotated assignment (e.g.
  `(x): int = 5`), which prevent CPython from including the name in `__annotations__`
  ([#&#8203;5321](psf/black#5321))
- Stop treating a t-string in docstring position as a docstring (e.g. `t"  spam  "` as
  the first statement of a module, class or function) ([#&#8203;5287](psf/black#5287))
- Fix unparseable output for a t-string whose replacement field contains a quote (e.g.
  `t'\'{a["b"]}\''`) ([#&#8203;5265](psf/black#5265))
- Fix unparseable output for a triple-quoted string whose body ends in an
  already-escaped double quote (e.g. `'''\'''\"'''`) ([#&#8203;5262](psf/black#5262))
- Fix `--skip-magic-trailing-comma` dropping the trailing comma from a split
  single-element tuple used as a lambda parameter default ([#&#8203;5246](psf/black#5246))
- Fix unstable formatting when an inline comment sits on optional parentheses (e.g. a
  parenthesized assert message) ([#&#8203;5241](psf/black#5241))
- Fix `--skip-magic-trailing-comma` dropping the trailing comma of a one-element
  subscript (`a[x,]`) when the line is long enough to be split and contains a power
  operator ([#&#8203;5272](psf/black#5272))
- Fix crash when a standalone comment sits between tokens of a comprehension or lambda
  ([#&#8203;5144](psf/black#5144))
- Fix inline comments on a bracket inside a comprehension being dropped ([#&#8203;5330](psf/black#5330))
- Respect the magic trailing comma in a PEP 695 type parameter list containing a
  `*TypeVarTuple` or `**ParamSpec` ([#&#8203;5244](psf/black#5244))
- Fix crash when a comment-only `# fmt: off`/`# fmt: on` block is followed by a `with`
  statement after another standalone comment ([#&#8203;5189](psf/black#5189))
- Fix a crash when splitting `case case if ...` match patterns at very small line
  lengths ([#&#8203;5147](psf/black#5147))
- Fix multiline docstring indentation when leading tabs are used inside indented
  docstrings ([#&#8203;5148](psf/black#5148))
- Respect `# fmt: skip` on a line that opens a bracket (e.g.
  `from x import (  # fmt: skip`) when a standalone comment is among the bracket's
  contents ([#&#8203;5161](psf/black#5161))
- Fix an AST safety error when separate `# type: ignore` comments in a parenthesized
  attribute chain were merged onto one physical line ([#&#8203;5297](psf/black#5297))
- Preserve comments and blank lines outside requested ranges when formatting with
  `--line-ranges` ([#&#8203;5175](psf/black#5175))
- Fix crash when `# fmt: skip` is used on one-line `async def`, `async with`, and
  `async for` statements containing a semicolon ([#&#8203;5311](psf/black#5311))
- Stop converting form feeds or other similar characters in docstrings into newlines
  ([#&#8203;5461](psf/black#5461))
- Fix `--skip-source-first-line` turning the skipped line's CRLF ending into `\r\r\n`
  when reformatting a file with CRLF line endings ([#&#8203;5438](psf/black#5438))

##### Preview style

##### New preview features

- Add missing blank lines after classes whose last method has an ellipsis body ([#&#8203;5439](psf/black#5439))
- Split only the brackets holding a magic trailing comma when more trailers follow them
  (e.g. stop splitting inside the `[2]` of `[1,][2](3)`) ([#&#8203;5448](psf/black#5448))
- Keep dictionary keys containing operators together on one line when the value can be
  wrapped onto a new line instead ([#&#8203;5435](psf/black#5435))
- Remove redundant parentheses around individual variables in unpacking targets (e.g.
  `for (x), (y) in points:` becomes `for x, y in points:`) ([#&#8203;5416](psf/black#5416))
- Normalize uppercase `T` prefixes on t-strings to lowercase ([#&#8203;5425](psf/black#5425))
- Remove redundant parentheses around generator expressions ([#&#8203;5304](psf/black#5304), [#&#8203;5369](psf/black#5369))
- Preserve two blank lines before a top-level class starting inside a `# fmt: off` block
  after an import ([#&#8203;5238](psf/black#5238))
- Fix unnecessary parentheses around short right-hand expressions in indexed assignments
  (e.g. `x[key] = expr`) ([#&#8203;5095](psf/black#5095))
- Parenthesize tuple expressions in `yield` statements for consistency with function
  calls and returns ([#&#8203;5170](psf/black#5170))
- Stop splitting between a variable and its operator (`not in`, `==`, `is`, ...) when
  the right-hand side is a bracketed expression, and instead split inside the brackets
  ([#&#8203;5135](psf/black#5135))
- In `.pyi` stub files, enforce a blank line after a function or method that has a
  docstring-only body when another comment or statement follows it ([#&#8203;5158](psf/black#5158))

##### Updates to existing preview features

- Fix crash in stub files when `# fmt: skip` is placed on a function in a group of
  same-name decorated functions (e.g. `@overload`s or a property setter) ([#&#8203;5430](psf/black#5430))
- Keep the parentheses around a lambda used as the iterable of a comprehension (e.g.
  `[x for x in (lambda: 0) if x]`) ([#&#8203;5176](psf/black#5176), [#&#8203;5200](psf/black#5200))

##### Updates to existing unstable features

- Do not treat multi-line expressions with merged strings and trailing `# type: ignore`
  comments as single-line unsplittable expressions ([#&#8203;5466](psf/black#5466))
- Fix duplicated inline comment when stripping the parentheses around a string or
  merging a backslash-continued string on the same line ([#&#8203;5449](psf/black#5449))
- Split long stringified return annotations even when the function has parameters
  ([#&#8203;5427](psf/black#5427))
- Don't hug brackets when doing so would join two `# type: ignore` comments onto one
  line ([#&#8203;5271](psf/black#5271))
- Fix a crash when `# type: ignore` is lost during formatting of a long parenthesized
  string ([#&#8203;5329](psf/black#5329))
- Fix only the first part of an implicitly concatenated unmergeable string (e.g.
  `r"..." r"..."`) being wrapped in parentheses ([#&#8203;5434](psf/black#5434))

##### Configuration

- Add support for `NO_COLOR` environment variable to disable ANSI output ([#&#8203;5129](psf/black#5129))
- Remove spurious target version warning when runtime version is included in a
  `--target-version` flag ([#&#8203;5167](psf/black#5167))
- Fix `--force-exclude` not excluding files whose path contains `..` (e.g.
  `black ../generated/file.py` run from a subdirectory) ([#&#8203;5471](psf/black#5471))
- Fall back to the default configuration, with a warning, when the given sources share
  no common project root (e.g. they are on different drives on Windows) instead of
  crashing ([#&#8203;5386](psf/black#5386))
- Fix loading a stale cached `pyproject.toml` path when `--code` is used from different
  working directories in the same process ([#&#8203;5152](psf/black#5152))
- Add validation for `--line-ranges` values ([#&#8203;5107](psf/black#5107))
- Ignore empty cache files instead of raising an `EOFError` ([#&#8203;5192](psf/black#5192))
- Reject non-string `include` and `force-exclude` values in `pyproject.toml` ([#&#8203;5193](psf/black#5193))
- Validate `BLACK_NUM_WORKERS` values and report invalid values as usage errors instead
  of crashing ([#&#8203;5211](psf/black#5211))
- Ignore permission errors when reading cache ([#&#8203;5258](psf/black#5258))

##### Packaging

- Reduce the size of Linux standalone binaries by stripping debug symbols during the
  PyInstaller release build ([#&#8203;5223](psf/black#5223))
- Black is now released using GitHub
  [Immutable Releases](https://docs.github.com/code-security/concepts/supply-chain-security/immutable-releases)
  ([#&#8203;5296](psf/black#5296))

##### Performance

- Fix superlinear runtime growth with the number of input files ([#&#8203;5450](psf/black#5450))
- Improve performance on strings containing many consecutive backslashes ([#&#8203;5163](psf/black#5163))
- Improve performance on files with many `# fmt: skip`/`# fmt: off`/`# fmt: on` comments
  ([#&#8203;5169](psf/black#5169), [#&#8203;5190](psf/black#5190), [#&#8203;5232](psf/black#5232))
- Improve performance on long calls and collections ([#&#8203;5177](psf/black#5177))
- Improve performance on multiline strings inside large collections (e.g. a dict literal
  with triple-quoted strings as values) ([#&#8203;5188](psf/black#5188))
- Improve performance on files with many soft keywords (e.g. `match`/`case` blocks)
  ([#&#8203;5186](psf/black#5186))
- Improve performance on long semicolon-separated statements (in the stable style) and
  large dict literals (in `--preview`) ([#&#8203;5184](psf/black#5184))
- Improve performance of `--line-ranges` on files with many sibling blocks (a long
  `if`/`elif` chain, a `match` with many cases, or many top-level definitions) ([#&#8203;5213](psf/black#5213))
- Improve performance on deeply nested bracketed expressions ([#&#8203;5171](psf/black#5171), [#&#8203;5242](psf/black#5242))
- Improve performance on lists and subscripts holding one long expression without
  operators (e.g. a long run of implicitly concatenated strings inside `[]`) ([#&#8203;5239](psf/black#5239))
- Improve performance on deeply chained operations (e.g. a long `a ** b ** c ** ...`
  chain) ([#&#8203;5235](psf/black#5235))
- Improve performance on long `if`/`elif` chains and other compound statements with many
  clauses ([#&#8203;5322](psf/black#5322))
- Improve performance of `string_processing`:
  - when merging implicitly concatenated f-strings containing long string literals
    ([#&#8203;5165](psf/black#5165))
  - when merging long implicitly concatenated strings ([#&#8203;5173](psf/black#5173), [#&#8203;5194](psf/black#5194))
  - when rewriting large nodes (e.g. `"%s ..." % (a, b, c, ...)`) ([#&#8203;5178](psf/black#5178), [#&#8203;5199](psf/black#5199), [#&#8203;5220](psf/black#5220))
  - when splitting long string literals ([#&#8203;5183](psf/black#5183))

##### Output

- Report parser failures using editor-friendly `path:line:column` locations ([#&#8203;5237](psf/black#5237))
- Fix crash when writing formatted code or diffs to a `sys.stdout` that has no `buffer`
  attribute (e.g. in Jupyter notebooks) ([#&#8203;5411](psf/black#5411))
- Report parse failures on Black's own output as internal errors ([#&#8203;5383](psf/black#5383))

##### *Blackd*

- Return HTTP 400 instead of 500 when the `X-Python-Variant` header is empty or has an
  empty entry (e.g. a trailing comma) ([#&#8203;5428](psf/black#5428))
- Allow optional whitespace around comma-separated versions and `pyi` in the
  `X-Python-Variant` header ([#&#8203;5441](psf/black#5441))

##### Integrations

- Remove unused `migrate-black` script ([#&#8203;5319](psf/black#5319))
- Support PEP 440 version specifiers in `tool.black.required-version` for the GitHub
  Action ([#&#8203;5399](psf/black#5399))
- Add outputs (`is_formatted`, `change_count`, `same_count`, `failure_count`) to GitHub
  Action runs ([#&#8203;5408](psf/black#5408))

##### Documentation

- Document `vim-python-pep8-indent`, which provides an `indentexpr` for Black-style
  insert-mode indentation ([#&#8203;5288](psf/black#5288))

</details>

---

### Configuration

📅 **Schedule**: (UTC)

- Branch creation
  - At any time (no schedule defined)
- Automerge
  - At any time (no schedule defined)

🚦 **Automerge**: Disabled by config. Please merge this manually once you are satisfied.

♻ **Rebasing**: Whenever PR is behind base branch, or you tick the rebase/retry checkbox.

🔕 **Ignore**: Close this PR and you won't be reminded about this update again.

---

 - [ ] <!-- rebase-check -->If you want to rebase/retry this PR, check this box

---

This PR has been generated by [Mend Renovate CLI](https://github.com/renovatebot/renovate).
<!--renovate-debug:eyJjcmVhdGVkSW5WZXIiOiI0NC4xMzIuNSIsInVwZGF0ZWRJblZlciI6IjQ0LjEzMi41IiwidGFyZ2V0QnJhbmNoIjoibWFpbiIsImxhYmVscyI6WyJ0eXBlL2RlcGVuZGVuY2llcyJdfQ==-->

Reviewed-on: https://git.tainton.uk/luke/instant-msg-api/pulls/290
Co-authored-by: renovate[bot] <renovate-bot@git.tainton.uk>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@sahvx655-wq @cobaltt7