Codacy backlog: 443 issues, mostly Prospector/Pyflakes findings

从 ruff check --select F src/machinevisiontoolbox tests 开始，检查列出的 29 处 F821，并区分实际的运行时问题、死代码或不可达代码以及测试问题。先对这些发现进行分类或修复，然后运行相关测试；这一阶段完成的标准是每一处 F821 发现都已得到处理，而更广泛的 lint 类别仍作为单独的后续工作。

Raised 2026-07-29 when the user pointed at the repo's Codacy dashboard
(443 issues total) and asked how much overlaps with the mixin/hygiene
work happening the same day. Verified: only the bare-except finding
(see git history, since fixed) genuinely overlapped. Everything else is
a distinct, much larger body of work, deliberately not tackled in that
pass — logging the real numbers here instead of re-deriving them from
scratch next time.

Codacy's Python analysis engine is Prospector (bundles Pylint +
Pyflakes + Bandit + pycodestyle + pydocstyle + mccabe) — confirmed via
the dashboard's own "Prospector's documentation" tab, and pattern names
like Avoid Dangerous Mutable Default Arguments / Audit Dangerous Subprocess Usage that are textbook Pylint/Bandit rule names. Codacy's
298-count "Detect Python Source Code..." bucket is all Pyflakes
findings grouped under one umbrella pattern, not broken out by code the
way ruff/raw pyflakes do.

Reproduced locally with ruff check --select F src/machinevisiontoolbox tests against clean origin/main, 2026-07-30: 883 hits (not
directly comparable to Codacy's 298 — different default
exclusions/config, and this sweep includes tests/, which Codacy's
dashboard count may not). By code:

F405/F403 (star-import ambiguity) dominate the count but are mostly
a style/tooling-friction issue, not bugs — this codebase leans on
from machinevisiontoolbox.base import *-style re-exports
deliberately (see the mypy wildcard-re-export issue for the
concrete downside of that pattern). F401/F841/F811 are typical
accumulated-cruft categories, individually low-risk to clean up but
numerous.

F821 (undefined name) is different — this is a real-bug class, not
style: a name that doesn't exist would raise NameError at runtime
if that code path is ever actually executed. All 29 instances, by
location:

• BundleAdjust.py:382,590,592 — undefined c, retain, g2
• ImageSpatial.py:116,121,328,330-332,340-342,1106 — undefined
  _border_opt, border_value, value, a, kv (kv appears 4
  times), conn
• VisualServo.py:186,412,444,1351-1353,1403 — undefined Animate,
  plot, history, camera, SphericalCamera, kwargs, pt
• blocks/camera.py:287,288 — undefined state (x2)
• tests/test_camera.py:191,192,194,195,198,200 — undefined x, y
  (likely a real bug in the test, not production code — check
  whether these lines actually run or are dead/unreachable test code)

Codacy's Pylint/Bandit-derived counts (the non-Pyflakes ~145 of the
443) weren't independently reproduced locally — the dashboard is the
source of truth for those categories (mutable default arguments,
assert usage, subprocess/exec/urlopen auditing, etc.).

Fix

Not a single pass. Suggested order: (1) triage the 29 F821 hits first
— for each, determine real bug vs. genuinely dead/unreachable code, fix
or delete accordingly; (2) F401/F811 next, mechanical and
ruff --fix-automatable for most cases; (3) F841 case-by-case (some
may be intentional, e.g. unpacking for side effects); (4) F405/F403
last and only if the codebase-wide star-import convention itself is
ever reconsidered — otherwise these will just regenerate.

Two more concrete instances, PRs #32/#33, 2026-07-30: Codacy
flagged type shadowing the builtin at ImageWholeFeatures.py:1213
(Histogram.plot's signature, PR #32) and again at :1570
(_compute_plot_series, the extraction in PR #33 that copied plot's
type parameter into a new method). Deliberately not renamed in
either PR — type= is public API (hist.plot(type="pdf")), a rename
needs a proper deprecation cycle. If picked up: this method
already has a precedent for exactly this — bar= is kept as a
deprecated alias for filled= with a DeprecationWarning
(ImageWholeFeatures.py, same method) — mirror that pattern: add
kind= as the real parameter, deprecate type= as an alias. Do this
as its own PR after #32 and #33 are both merged, not before —
branching the rename off pre-#32 main would conflict with both of
those on the same lines.

PR #33 also surfaced 3 more Codacy findings while extracting
Image.__getitem__'s nested closures (ImageCore.py): max as a
parameter name (:2792, _lenkey — carried over verbatim from the
original nested lenkey(key, max), not introduced by the extraction)
and two F405 star-import-ambiguity hits (:352 Dtype, :2812
Any, both from machinevisiontoolbox.mvtb_types's star-import) —
already covered by the F405 finding above, not a new pattern.