[jayzhou2309]

Description

Fixes #5488.

With --preview, this assignment loses the parentheses around its right-hand side and the first line grows to 96 characters:

self.a_long_class_attribute_to_use = A_LONG_TEMPLATE_CONSTANT_NAME_HERE.safe_substitute(
    {
        "n": (
            "long text that breaks the ternary" if A_LONG_CONSTANT_NAME else ""
        ),
    },
)

Stable style keeps the parentheses because of the second opinion in run_transformer: when the first line of a right-hand split is still too long, it retries with FORCE_OPTIONAL_PARENTHESES and takes that result if every line fits. The retry is skipped when one of the line's invisible parens has turned visible, because that means the split already used them.

wrap_long_dict_values_in_parens makes the dict value's own parens invisible, so they sit in the same bracket_tracker.invisible list as the line's optional parens. Splitting the sub-line for the dict value makes them visible again, the guard reads that as "the optional parens were used", and the retry never runs. The line's own optional parens are still invisible at that point.

Under wrap_long_dict_values_in_parens, the guard now looks only at the line's top-level invisible parens (bracket_depth == 0). The depths are read before the transform runs, because the sub-line splits re-mark the same leaves. Stable style takes the same code path as before, since the filter only applies with the preview feature on. The output now matches stable's shape:

self.a_long_class_attribute_to_use = (
    A_LONG_TEMPLATE_CONSTANT_NAME_HERE.safe_substitute(
        {
            "n": (
                "long text that breaks the ternary" if A_LONG_CONSTANT_NAME else ""
            ),
        },
    )
)

The existing Random.func case in preview_long_dict_values.py had the same bug: its expected output had a 90-character first line. It now wraps the right-hand side and fits in 88, so this PR updates that expectation.

Verification:

• The new case in preview_long_dict_values.py fails on main and passes with the fix. It lands in its own commit before the fix.
• The full test suite passes: 551 passed, 3 skipped. That includes preview_long_strings__regression, which the length-based approach discussed in the issue broke.
• --preview output is byte-identical before and after the change across 2,268 files (installed site-packages, celery, agno, mempalace).
• A second review ran 57 extra inputs (chained, annotated and augmented assignments, return, del, for, with, lambdas, comments, type: ignore) at line lengths 1, 40, 60, 88 and 120 in preview and unstable. None crashed, all were stable on a second pass, and none came out wider than on main.
• pre-commit (isort, flake8, mypy, prettier) passes.

AI disclosure: an AI agent (Claude Code) reproduced the issue, traced the cause, wrote the fix, the test, and this description, and ran the checks above (pytest tests -n auto, pre-commit run, the corpus comparison against main).

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?

🤖 Written and posted by an AI agent (Claude Code) on behalf of @jayzhou2309.

[github-actions]

diff-shades results comparing this PR (bbf7b32) to main (b801776):

--preview style: no changes

--stable style: no changes

---

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

[cobaltt7]

Thanks!