literal() still misses tuple-unpack rebinding and del of the prefix lists

[realmarcin]

literal() in tests/test_fleet_page.py walks the tree for Assign, AugAssign and AnnAssign, and separately flags subscript assignment and mutating method calls. That kills the five shapes in #99. These still survive, each leaving the test reading the original literal while the module uses something else:

mutant	why it survives	worth closing
VOC, _X = [...], 1	tuple-unpack target is a Tuple, not a Name, so getattr(target, "id") is None	yes — VOC, PREF = load() is a natural refactor
del VOC[0] / del VOC[:]	ast.Delete is not examined at all	yes — cheap to add
VOC[0] += "X"	the AugAssign branch tests node.target.id, so a Subscript target is skipped. Asymmetric: plain VOC[0] = is caught	moderate
VOC[0], _x = "BOGUS", 1	subscript store nested inside a Tuple target	low
print((VOC := [...]))	ast.NamedExpr not handled	low
for VOC in [...], with ... as VOC at module level	neither is Assign	low
globals()["VOC"] = ..., exec(...), list.append(VOC, x)	dynamic	no — accept

Suggested fix

Collect stores rather than top-level target ids, which covers the first five in one pass:

for node in ast.walk(tree):
    targets = (node.targets if isinstance(node, (ast.Assign, ast.Delete))
               else [node.target] if isinstance(node, (ast.AugAssign, ast.AnnAssign, ast.NamedExpr))
               else [])
    for target in targets:
        for sub in ast.walk(target):
            if isinstance(sub, ast.Name) and sub.id == name and isinstance(sub.ctx, (ast.Store, ast.Del)):
                ...
            elif isinstance(sub, ast.Subscript) and getattr(sub.value, "id", None) == name:
                mutations.append("subscript assignment")

The ctx guard matters: in VOC[0] = x the Name node is a Load, so without it the subscript store would be miscounted as a rebinding.

Also worth a line each

• assertGreater(len(value), 10) does not check the value is a list. VOC = "abcdefghijklmnop" — a 16-character string — survives, as does a duplicated entry inside the literal. assertIsInstance(value, list) plus a duplicate check is cheap.
• A bare VOC: list with no value makes ast.literal_eval(None) raise ValueError: malformed node or string rather than a clean assertion. It still fails, just untidily.

None of these shapes exists in either script today. The real fix remains #97 — make the modules importable so the tests read the actual objects instead of parsing text for them; every entry above then becomes moot.

Found by an adversarial review of #100.

[realmarcin]

Closed by #108 (main 26e02a6): the prefix-list tests now import build_data and build_subsets and read the real VOC and PREF objects, so the source-parsing hole this recorded no longer exists. GitHub only honours the first number after a closing keyword, which is why the merge left this open.