[refurb] Implement repeated-global (FURB154)

[alex-700]

Summary

Implement repeated_global (FURB154) lint.
See:

• Support for refurb #1348
• original lint

Test Plan

cargo test

[github-actions]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

ℹ️ ecosystem check detected linter changes. (+15 -0 violations, +0 -0 fixes in 3 projects; 47 projects unchanged)

apache/airflow (+8 -0 violations, +0 -0 fixes)

ruff check --no-cache --exit-zero --ignore RUF9 --output-format concise --preview --select ALL

+ airflow/plugins_manager.py:361:5: FURB154 [*] Use of repeated consecutive `global`
+ airflow/plugins_manager.py:422:5: FURB154 [*] Use of repeated consecutive `global`
+ airflow/plugins_manager.py:476:5: FURB154 [*] Use of repeated consecutive `global`
+ airflow/plugins_manager.py:503:5: FURB154 [*] Use of repeated consecutive `global`
+ airflow/providers/weaviate/hooks/weaviate.py:418:13: FURB154 [*] Use of repeated consecutive `nonlocal`
+ airflow/settings.py:194:5: FURB154 [*] Use of repeated consecutive `global`
+ airflow/settings.py:299:5: FURB154 [*] Use of repeated consecutive `global`
+ airflow/settings.py:426:5: FURB154 [*] Use of repeated consecutive `global`

bokeh/bokeh (+1 -0 violations, +0 -0 fixes)

ruff check --no-cache --exit-zero --ignore RUF9 --output-format concise --preview --select ALL

+ tests/unit/bokeh/document/test_callbacks__document.py:168:13: FURB154 [*] Use of repeated consecutive `nonlocal`

zulip/zulip (+6 -0 violations, +0 -0 fixes)

ruff check --no-cache --exit-zero --ignore RUF9 --output-format concise --preview --select ALL

+ zerver/data_import/slack.py:532:9: FURB154 [*] Use of repeated consecutive `nonlocal`
+ zerver/data_import/slack.py:604:9: FURB154 [*] Use of repeated consecutive `nonlocal`
+ zerver/lib/cache.py:63:5: FURB154 [*] Use of repeated consecutive `global`
+ zerver/lib/markdown/__init__.py:2727:5: FURB154 [*] Use of repeated consecutive `global`
+ zerver/lib/templates.py:106:5: FURB154 [*] Use of repeated consecutive `global`
+ zerver/tests/test_auth_backends.py:7644:13: FURB154 [*] Use of repeated consecutive `nonlocal`

Changes by rule (1 rules affected)

code	total	+ violation	- violation	+ fix	- fix
FURB154	15	15	0	0	0

[charliermarsh]

Since globals and nonlocals are rare, could we try instead running this rule on every global and nonlocal statement, and then, for each statement, mapping back to the containing Suite and looking for global statements that follow it repeatedly?

In other words... instead of iterating over every body / statement, can we instead run this for each global? You can look at traversal::next_sibling or traversal::suite (and their usages) for an example.

[alex-700]

Since globals and nonlocals are rare, could we try instead running this rule on every global and nonlocal statement, and then, for each statement, mapping back to the containing Suite and looking for global statements that follow it repeatedly?

In other words... instead of iterating over every body / statement, can we instead run this for each global? You can look at traversal::next_sibling or traversal::suite (and their usages) for an example.

@charliermarsh done.
Just a note: now the algorithm is quadratic on len(suite) for each suite containing global or nonlocal. Instead of linear (and should be auto vectorized) for each suite.

[charliermarsh]

Thanks @alex-700. Blocked on me, will review shortly.

[codspeed-hq]

CodSpeed Performance Report

Merging #11187 will degrade performances by 9.38%

_{Comparing alex-700:latyshev/furb154 (d66b3f1) with alex-700:latyshev/furb154 (333b7f4)}

Summary

⚡ 9 improvements
❌ 8 regressions
✅ 13 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	alex-700:latyshev/furb154	alex-700:latyshev/furb154	Change
⚡	lexer[large/dataset.py]	1.4 ms	1.1 ms	+28.25%
⚡	lexer[numpy/ctypeslib.py]	285.3 µs	225.8 µs	+26.32%
⚡	lexer[numpy/globals.py]	36.1 µs	29.4 µs	+23.08%
⚡	lexer[pydantic/types.py]	639.6 µs	502.8 µs	+27.22%
⚡	lexer[unicode/pypinyin.py]	98.2 µs	76.6 µs	+28.21%
❌	linter/all-rules[large/dataset.py]	15.5 ms	16.1 ms	-4.27%
❌	linter/all-rules[unicode/pypinyin.py]	2 ms	2.1 ms	-6.2%
❌	linter/default-rules[large/dataset.py]	3.5 ms	3.7 ms	-5.65%
❌	linter/default-rules[numpy/ctypeslib.py]	879.6 µs	945.5 µs	-6.97%
❌	linter/default-rules[pydantic/types.py]	1.7 ms	1.9 ms	-9.38%
❌	linter/default-rules[unicode/pypinyin.py]	341.6 µs	359.8 µs	-5.07%
❌	linter/all-with-preview-rules[numpy/globals.py]	777 µs	812.5 µs	-4.37%
❌	linter/all-with-preview-rules[unicode/pypinyin.py]	2.2 ms	2.4 ms	-6.51%
⚡	parser[large/dataset.py]	5.6 ms	5.3 ms	+5.85%
⚡	parser[numpy/ctypeslib.py]	1.1 ms	1 ms	+7.76%
⚡	parser[numpy/globals.py]	122.3 µs	107 µs	+14.27%
⚡	parser[unicode/pypinyin.py]	372 µs	341.6 µs	+8.89%

[charliermarsh]

I'm confused by the CodSpeed results but my guess is it's just confused by the branch being dated? We didn't touch the parser or lexer here of course.