Preserve blank lines before fmt-off classes

[sjh9714]

Description

Fixes #5225.

When a top-level class starts inside a # fmt: off block after an import, the
converted block looks like a standalone comment to the empty-line tracker. This caused
Black to keep only one blank line instead of the two required before a top-level class.

Preserve the converted block's original class token when deciding import separation.
The change is limited to a new preview feature, so stable formatting remains unchanged,
and a focused fixture covers the reported input.

Testing

• .venv/bin/pytest tests/test_format.py -k preview_fmt_off_class_blank_lines -q
• .venv/bin/pytest tests/test_format.py -q
• env -u NO_COLOR .venv/bin/pytest -q
• env -u NO_COLOR .venv/bin/pre-commit run mypy --files src/black/lines.py src/black/mode.py
• env -u NO_COLOR .venv/bin/tox -e run_self
• .venv/bin/tox -e generate_schema

env -u NO_COLOR .venv/bin/pre-commit run --all-files passed every hook except
repository-wide mypy, which reports an unchanged action/main.py tomllib
ignore/stub mismatch under Python 3.14. The targeted mypy hook above passes for both
changed Python modules.

Note: I used Codex while preparing this change, reviewed the final diff, and ran the
checks listed above locally.

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• Add new / update outdated documentation?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]

diff-shades results comparing this PR (c8f5587) to main (fcab648):

--preview style: no changes

--stable style: no changes

---

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

[cobaltt7]

What logic normally handles newlines before classes? Why isn't it triggering here? Would it make morse sense to update it?

[sjh9714]

Good question — I dug into where this comes from.

The logic that normally handles blank lines before classes is EmptyLineTracker._maybe_empty_lines, which routes any line whose is_decorator/is_def/is_class is true to _maybe_empty_lines_for_class_or_def (src/black/lines.py:1111).

It doesn't trigger here because normalize_fmt_off collapses the whole # fmt: off region into a single STANDALONE_COMMENT leaf (convert_one_fmt_off_pair in src/black/comments.py), keeping the original first token only in fmt_pass_converted_first_leaf. So the line's first leaf is a STANDALONE_COMMENT, not a NAME class — is_class is False, the class/def dispatch is skipped, and the line falls through to the top-level import rule just below (src/black/lines.py:1128), which returns 1 blank line. That's exactly the reported behavior.

On updating the normal logic instead: I considered teaching is_class (or the dispatch) to see through fmt-pass-converted lines, but is_class/is_def are consulted in many other places (stub-class handling, previous_defs tracking, docstring logic), and routing an opaque fmt:off block into _maybe_empty_lines_for_class_or_def would let other rules act on content that # fmt: off promises to leave alone. The two import rules directly below already special-case is_fmt_pass_converted(first_leaf_matches=is_import), so I kept the fix in the same shape and place: a narrow guard for the import → fmt-converted-class boundary.

That said, if you'd rather see this handled inside _maybe_empty_lines_for_class_or_def (or generalized beyond the after-imports case), I'm happy to rework it that way.

[cobaltt7]

Sounds good. Thank you for investigating & fixing this!

[cobaltt7]

Whoops, wrong button