Bug: scope no_color color-stripping to the Outputter lifetime

[danielfromearth]

GitHub Issue: #345

Description

Outputter.__init__ blanked colorama's process-wide Fore/Style singletons
when no_color=True and never restored them, so the change leaked:

• calling compare(..., no_color=True) then compare(...) in the same
  interpreter produced colorless output on the second call; and
• any other in-process colorama user was silently de-colored.

The mutation is now bounded from both ends:

1. __exit__ restores the per-instance snapshot taken in __init__, scoping
   the mutation to the Outputter's lifetime (compare() already uses it as a
   context manager). Restoring the per-instance values rather than pristine ones
   keeps a no-color Outputter nested inside another behaving correctly.
2. Constructing a colorized Outputter restores a pristine snapshot taken at
   import, before any Outputter can touch the singletons. This covers callers
   that never use the context manager — tests/conftest.py builds bare
   Outputters, and a library user can too — for whom __exit__ never runs and
   the leak was still permanent.

In-block behavior is unchanged. Blanking the constants is load-bearing rather
than cosmetic: side_by_side pads the leading gutter by len(Fore.RED) to pay
for the escape sequence it prefixes onto the first column, and blanking makes
that compensation zero. The expanded comment in __init__ records this, and a new
test pins it.

Also drops self._no_color, which was added but never read.

A fuller fix (routing color through the instance instead of mutating globals at
all) remains the follow-up noted in #345.

Local test steps

uv run pytest → 48 passed, 1 deselected (baseline on develop is 44 passed,
1 deselected). uv run pre-commit run --all-files → all hooks pass, mypy included.

Four tests now cover no_color, three of them added in this revision:

• test_no_color_state_is_restored_after_context_exit (tests/test_printing.py,
  from the earlier revision of this PR) — the color constants are blank inside
  with Outputter(no_color=True) and restored afterward.
• test_no_color_leak_is_repaired_by_a_later_colorized_outputter (tests/test_printing.py,
  new) — builds a no-color Outputter without with, so __exit__ never runs,
  then asserts that constructing a colorized Outputter repairs the global state.
• test_no_color_flag_suppresses_ansi_escape_sequences (tests/test_cli.py, new) —
  the --no-color CLI contract, which had no coverage at all. This one needs a
  pseudo-terminal: colorama strips escape sequences when stdout is not a
  terminal, so over a plain pipe a colorized run and a no-color run are
  byte-identical and the flag is untestable. Under a pty the default run emits
  escape sequences and --no-color emits none. Skipped on non-POSIX, since pty
  is Unix-only.
• test_no_color_report_is_aligned_the_same_as_the_colorized_one (tests/test_cli.py,
  new) — pins the gutter arithmetic described above by asserting that stripping ANSI
  from a colorized report reproduces the --no-color report exactly.

Red-before-green, verified by swapping in each earlier version of
printing.py and re-running:

Version of printing.py	..._restored_after_context_exit	..._repaired_by_a_later_colorized_outputter
develop (no restore at all)	fail	fail
this PR's earlier __exit__-only restore	pass	fail
this PR as it now stands	pass	pass

test_no_color_report_is_aligned_the_same_as_the_colorized_one was likewise
confirmed to fail when extra_style_space = " " * len(default_color) is replaced
with "", i.e. it genuinely guards the compensation rather than passing by
construction.

Manual check of the API symptom in the issue, which no test exercises end to end:
compare(a, b, no_color=True) followed by compare(a, b) in one interpreter now
leaves Fore.RED at '\x1b[31m', and unrelated colorama use still colorizes.

Overview of integration done

[n/a] — unit-level only; no external services are involved. The comparison
report itself is unchanged, which the golden-file tests in
tests/test_complete_file_output.py confirm (txt/csv/xlsx all still match
tests/data/a-b_test_golden_file.*), as does the unchanged
EXPECTED_ATL06_DIFFERENCES = 4958.

Worth flagging for the reviewer: until this branch was brought up to date, it had
never been tested by CI at all. The pull-request workflow filters on the base
branch, develop was only recently added to that filter (96a33e9), and a
pull_request run reads its workflow definition from the merge ref — which here was
both stale and conflicted, so no run could be scheduled. The failure mode is that
the check list stays short rather than turning red, which is easy to misread as
"tested".

Merging develop in fixed that. The suite has now run on this PR and passed on
Python 3.11, 3.12, 3.13 and 3.14 (Linux), collecting 48 selected / 1 deselected,
matching the local run — and that is the first Linux run these changes have had,
which matters for anything near the HDF5 stack. codecov/patch and
codecov/project both pass. Confirmed from the CI log rather than assumed: the
pseudo-terminal test executes on Linux rather than silently skipping.

PR Acceptance Checklist

• Unit tests added/updated and passing — 48 passed, 1 deselected locally and on CI across Python 3.11–3.14; each new no-color test confirmed red on the pre-fix code.
• [n/a] Integration testing — no external services; the ATL06 and golden-file regressions are unchanged.
• CHANGELOG.md updated — one bullet under ### Fixed for Bug: no_color=True permanently mutates colorama's global Fore/Style state #345.
• [n/a] Documentation updated (if needed) — no user-facing flag or API change; --no-color's documented behavior is unchanged, only its leakage.

---

📚 Documentation preview 📚: https://ncompare--346.org.readthedocs.build/en/346/

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.61%. Comparing base (de22420) to head (643f163).
⚠️ Report is 1 commits behind head on develop.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #346      +/-   ##
===========================================
+ Coverage    95.81%   96.61%   +0.80%     
===========================================
  Files            9        9              
  Lines          550      562      +12     
===========================================
+ Hits           527      543      +16     
+ Misses          23       19       -4     

Flag	Coverage Δ
python-3.11	96.61% <100.00%> (+0.80%)	⬆️
python-3.12	96.61% <100.00%> (+0.80%)	⬆️
python-3.13	96.61% <100.00%> (+0.80%)	⬆️
python-3.14	96.58% <100.00%> (+0.81%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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