Honor # isort: skip when a __future__ import is present (#2092)

[apoorvdarshan]

Fixes #2092.

Problem

A per-line # isort: skip directive was silently ignored when a __future__ import was present in the same block:

from __future__ import annotations

from foo import bar  # isort: skip
from bar import baz

became

from __future__ import annotations

from bar import baz

from foo import bar  # isort: skip

The skipped import is relocated below the block, violating the directive.

Root cause

__future__ imports are always floated to the very top of the file. Internally, isort collapses each contiguous run of imports into a single sorted block that is spliced back in at the position of the first import in the run (parse.py import_index). A # isort: skip line is kept out of the sort and re-emitted as a plain "out line", so it can only ever land before or after the whole block — never in the middle.

When the skipped import is the first line of its section (the common case), the block is inserted after it and it stays put. But as soon as any real import precedes it — most naturally a floated __future__ import — the block is inserted ahead of the skipped line, pushing it below.

Fix

Treat a per-line isort: skip import as an implicit section boundary when earlier imports have already been collected into the current section: flush those imports and emit the skipped statement in place, so nothing can be sorted above it. This mirrors isort's existing # isort: split behavior and keeps the skipped line exactly where the user put it.

The contains_imports guard is important — it means a skipped import that is already alone in its section (e.g. after a non-import statement) keeps flowing through the existing path unchanged, so trailing-whitespace normalization and other established behaviors are untouched.

Verification

• Added test_isort_skip_is_honored_with_future_import_issue_2092 to tests/unit/test_regressions.py. It fails on main (the skipped line is relocated) and passes with this change. It also covers an interleaved skip (import / skip / import) and asserts idempotency.
• The reported case is now a no-op, and the fix is stable across re-runs (isort.check_code(...) passes).
• Full unit suite green: pytest tests/unit/ — 576 passed (the only non-passing tests are pre-existing and environment-specific: test-ordering pollution in test_importable, which passes in isolation, and tests that shell out to the isort CLI / require the example plugins).
• ruff check and ruff format --check clean on the changed files (the process mccabe complexity stays within the configured limit — the new predicate is folded into the existing branch), and mypy isort/core.py reports no issues.
• Added a CHANGELOG.md entry.

---

Disclosure: this change was prepared with the assistance of an AI tool (Claude Code). I reproduced the issue, reviewed and verified the fix, the test, and the full test run, and take responsibility for the contribution and will respond to review feedback personally.

[DanielNoord]

Please remove the changelog line :)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.26%. Comparing base (a0cc97f) to head (793a91e).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2574   +/-   ##
=======================================
  Coverage   99.25%   99.26%           
=======================================
  Files          41       41           
  Lines        3108     3111    +3     
  Branches      674      674           
=======================================
+ Hits         3085     3088    +3     
  Misses         13       13           
  Partials       10       10           

🚀 New features to boost your workflow:

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

[apoorvdarshan]

Done — removed the changelog line. Thanks for the review!