[simpleqt]

Summary

multi_line_output=9 (VERTICAL_PREFIX_FROM_MODULE_IMPORT) and multi_line_output=10
(HANGING_INDENT_WITH_PARENTHESES) delete a trailing comment from a from import
when the statement has only one import left to emit. The comment that sits there is
almost always a noqa, so the output silently re-activates the lint error it was
suppressing. For mode 10 the output was also not idempotent: it produced
from a import (b) for a single import, which the next run rewrote to from a import b.

>>> import isort
>>> src = "from a import b  # noqa: F401\n"
>>> first = isort.code(src, multi_line_output=10, line_length=20)
>>> first                      # before the fix: comment gone
'from a import (b)\n'
>>> isort.code(first, multi_line_output=10, line_length=20)   # not idempotent
'from a import b\n'
>>> isort.code(src, multi_line_output=9, line_length=20)      # before the fix: comment gone
'from a import b\n'

Root cause

Both modes attach the collected trailing comment while wrapping the next import:

• hanging_indent_with_parentheses (isort/wrap_modes.py) only reaches
  comments.add_to_line() inside the while imports: loop, and the first-import
  length check ignores the comment, so a lone import that fits under the limit left
  comments unconsumed and the final _add_syntax(statement, ")") never used them.
• vertical_prefix_from_module_import used
  if comments and statement_with_comments: after the loop; statement_with_comments
  is only set by a loop iteration, so with one import the comment was discarded.

isort/output.py deliberately routes a single import that carries a noqa comment
through the wrap-mode formatter (_build_grouped_from_imports, the
has_noqa_comment guard) to keep the comment on the statement line — which is exactly
the path that dropped it. Any other comment on a single import goes through
wrap.line(), which preserves it, so the loss was specific to modes 9/10 (a sweep of
all 12 modes shows 0 losses in the other 10).

Fix

• Mode 10: when the first import is also the last one, attach the comment before the
  fit check. It is hoisted onto the ( line when it does not fit — the shape the mode
  already produces when the import itself is too long — and kept after the closing
  parenthesis when it does fit.
• Mode 9: emit any comment the wrapping loop did not consume, instead of only the one
  carried by the last loop iteration. For statements with several imports this is
  byte-for-byte the previous behaviour (the removed statement_with_comments value was
  comments.add_to_line(comments, output_statement)), so only the single-import case
  changes.

The multi-import comment positions are pinned by a new test, and the 4000-case
differential corpus below confirms that every behavioural difference is a comment that
used to be deleted.

Regression tests

• tests/unit/test_wrap_modes.py::test_hanging_indent_with_parentheses_preserves_comment_of_a_single_import
• tests/unit/test_wrap_modes.py::test_hanging_indent_with_parentheses_comment_position_unchanged_for_many_imports
  (pin: multi-import comment position is unchanged)
• tests/unit/test_wrap_modes.py::test_vertical_prefix_from_module_import_preserves_comment_of_a_single_import
• tests/unit/test_regressions.py::test_wrap_modes_9_and_10_keep_a_trailing_noqa_comment_on_a_single_import
  (public API: exact output, noqa survives, code(code(x)) == code(x) and ast.parse
  over modes 9/10, line_length 12–40, include_trailing_comma, force_grid_wrap=1)

They mirror the neighbouring tests added for mode 0 in PR #2675.

RED — new tests with only isort/wrap_modes.py reverted to main

$ pytest tests/unit/test_wrap_modes.py::test_hanging_indent_with_parentheses_preserves_comment_of_a_single_import \
         tests/unit/test_wrap_modes.py::test_hanging_indent_with_parentheses_comment_position_unchanged_for_many_imports \
         tests/unit/test_wrap_modes.py::test_vertical_prefix_from_module_import_preserves_comment_of_a_single_import \
         tests/unit/test_regressions.py::test_wrap_modes_9_and_10_keep_a_trailing_noqa_comment_on_a_single_import -q
E       AssertionError: assert 'from a import b' == 'from a impor... # noqa: F401'
E         - from a import b  # noqa: F401
E         + from a import b
...
E       AssertionError: assert 'from a import (b)\n' == 'from a impor...401\n    b)\n'
E         + from a import (b)
E         - from a import (  # noqa: F401
E         -     b)
=========================== short test summary info ===========================
FAILED tests/unit/test_wrap_modes.py::test_hanging_indent_with_parentheses_preserves_comment_of_a_single_import
FAILED tests/unit/test_wrap_modes.py::test_vertical_prefix_from_module_import_preserves_comment_of_a_single_import
FAILED tests/unit/test_regressions.py::test_wrap_modes_9_and_10_keep_a_trailing_noqa_comment_on_a_single_import
3 failed, 1 passed in 0.32s

(the 4th, ..._comment_position_unchanged_for_many_imports, passes on main by
design — it pins behaviour that must not change.)

Full transcription: G:\pr\work\out\red.txt

GREEN — same tests with the fix

$ pytest <same four node ids> -q
....                                                                     [100%]
4 passed in 0.66s

Full transcription: G:\pr\work\out\green.txt

Verification

• pytest tests/unit/test_wrap_modes.py tests/unit/test_regressions.py -q → 158 passed

• pytest tests/unit -q → 1 failed, 704 passed, 5 skipped; main (same venv, same
  command) → 1 failed, 700 passed, 5 skipped. The single failure is identical on
  both and is pre-existing: test_windows_diff_too_large_misrepresentative_issue_1348
  rewrites tests/unit/example_crlf_file.py in place, so it fails after an earlier test
  in the same session corrupts that fixture (it passes in isolation, on both revisions).
  The 4 extra passing tests are the new ones.
  (The two example_* plugin packages from the dev dependency group were installed
  locally, otherwise test_sort_configurable_sort_issue_1732 and
  test_isort_supports_shared_profiles_issue_970 fail on base as well.)

• pytest tests/integration/test_hypothesmith.py -q (property test:
  code(code(x)) == code(x) over hypothesmith-generated sources and randomised
  isort.Config) → 1 passed.

• Lint/type checks, run as in tox -e lint:

  • ruff check isort/ tests/ scripts/ → All checks passed!
  • ruff format --check (all project dirs) → 105 files already formatted
  • flake8 isort/ tests/ → clean (exit 0)
  • mypy → 1 pre-existing error, identical on main:
    tests/unit/test_settings.py:101: error: Module has no attribute "mkfifo" (Windows-only)
  • bandit -r isort/ → clean (exit 0)
  • isort --profile hug --check --diff isort/wrap_modes.py tests/unit/test_wrap_modes.py tests/unit/test_regressions.py
    → clean. The repo-wide variant (isort/ tests/ scripts/) reports the CRLF fixture
    tests/unit/example_crlf_file.py as unsorted — identical output with the change
    stashed, a Windows/CRLF artefact of this checkout, not caused by the change.

• Sweep (scratch harness, not committed): 20 source shapes × 30 line lengths × 11
  option sets × all 12 modes = 79,200 runs, checking that a marker comment survives
  the round trip and that the output parses and is stable:

  	marker comment lost	unparsable output	non-idempotent
  before	1,782	0	720
  after	0	0	190

  The remaining 190 non-idempotent cases are unrelated pre-existing cases in other
  modes (e.g. mode 0 with profile=black at very short line_length); a targeted
  base/fixed comparison of that class shows 48 → 36 and no case that is non-idempotent
  only after the fix.

• Differential corpus (4,000 randomised source/config pairs, before vs after): 30
  outputs differ, and every one of them is a previously deleted comment now kept —
  no case changes in any other way.

Risks

• Mode 10 emits from a import (b) # noqa: F401 (parens around a single import) when
  the comment fits; that shape already existed, only the comment is new. The ) is
  always placed before the comment by the existing _add_syntax helper, so the output
  stays valid Python (asserted in the regression test with ast.parse).
• remove_comments/ignore_comments behaviour is unchanged: comments.add_to_line is
  called with removed=remove_comments, so with ignore_comments=True no comment is
  emitted (only the single-import path now adds a call that is a no-op in that case).
• Mode 9 gains a comment on the continuation line of a single over-long import
  (from a import \ / b # noqa), which lengthens that line; it is the same
  fallback shape used for a long single import without a comment.

Artifacts

• Patch: G:\pr\work\out\is-wrap-modes-trailing-comment.patch
• Repro script (uncommitted, kept in the clone): G:\pr\work\is\scratch_repro.py;
  unpatched transcript G:\pr\work\out\scratch_repro_base.txt (4/4 cases lose the
  comment, 2 of them also non-idempotent), patched run → 0/4.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[DanielNoord]

Tests are a bit verbose but oh well.

Please for the sanity of open source maintainers, turn down the verbosity of your PR descriptions. This is really poor.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.38%. Comparing base (f97a190) to head (b22f624).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2733   +/-   ##
=======================================
  Coverage   99.38%   99.38%           
=======================================
  Files          40       40           
  Lines        3253     3255    +2     
  Branches      700      701    +1     
=======================================
+ Hits         3233     3235    +2     
  Misses         12       12           
  Partials        8        8           

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.