Keep the required trailing comma of a one-element subscript under --skip-magic-trailing-comma

[LuShadowX]

Description

--skip-magic-trailing-comma drops the trailing comma from a one-element subscript when the line is long enough to be split and contains a power operator, which turns a[x,] (indexing with a one-tuple) into a[x] and trips Black's own AST safety check:

value = alpha[beta,]() ** gamma * delta << epsilon | zeta < eta ^ theta * iota + kappa + mu

$ black --skip-magic-trailing-comma file.py
error: cannot format file.py: INTERNAL ERROR: Black produced code that is not equivalent to the source.

The comma survives on a short line — tests/data/cases/skip_magic_trailing_comma.py opens by asserting exactly that — so this is the same rule failing on a longer one. Under --fast there is no safety net and the changed code is written out.

hug_power_op is the path. It rebuilds the line with leaf.clone() for every leaf, and a clone has no parent, so both tree-based guards that protect a required comma stop working: has_magic_trailing_comma's single-element-subscript branch needs closing.parent.type == syms.trailer, and the one-tuple guard in Line.append needs leaf.parent. Neither can fire, the comma looks magic, and remove_trailing_comma() takes it.

Only the operands either side of a hugged ** actually need their prefix rewritten, so this stops cloning the rest and reuses the leaf, keeping it attached to the tree. Reusing leaves across lines is what bracket_split_build_line already does, and with preformatted=True Line.append does not touch the prefix — it only re-runs BracketTracker.mark, which recomputes the same metadata for an identical leaf sequence.

The one-tuple guard the same clone breaks (foo(lambda x=(1,): x), already in that case file) is fixed by this too; I could not build a failing example for it because the lambda cases are short enough not to reach hug_power_op.

Found with a randomised run of scripts/fuzz.py's property — the committed harness pins derandomize=True, max_examples=1000, so it re-explores the same thousand inputs on every CI run and never reaches this one.

Verification

• Case added to tests/data/cases/skip_magic_trailing_comma.py. It fails on main (alpha[beta]() in the output) and passes with the change.
• pytest: 474 passed, 3 skipped, unchanged from main. black --check ., flake8, isort, mypy clean.
• Formatted 716 files from scrapy, pydantic, Pillow and fontTools with --skip-magic-trailing-comma before and after, comparing output hashes: zero differences, no errors either way. Power-operator hugging is unaffected (x = a ** b still becomes x = a**b).

Checklist - did you ...

• Implement any code style changes under the --preview style, following the stability policy? — no style change; this restores the documented behaviour for a case that was crashing
• Add an entry in CHANGES.md if necessary?
• Add / update tests if necessary?
• Add new / update outdated documentation? — no documented behaviour changes

[LuShadowX]

One more data point: 6000 randomised examples through scripts/fuzz.py's property on top of this branch — the same run that produced the failing input on main — come back clean.

[github-actions]

diff-shades results comparing this PR (85055d1) to main (c7801d9):

--preview style: no changes

--stable style: no changes

---

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