Reading an attribute that does not exist yields None — and on None it can yield a STRING

x = None
print(x.foo)          # CPython: AttributeError    pxx: None
print(x.upper())      # CPython: AttributeError    pxx: 'NONE'

class K:
    pass2 = 1
k = K()
print(k.nope)         # CPython: AttributeError    pxx: None

d = {}
print(d.get("k").x)   # CPython: AttributeError    pxx: None

None.upper() returning 'NONE' is the one to look at first: the receiver was stringified to "None" and the string method applied to that text. So a missing value does not merely propagate as None — it can turn into plausible data.

Measured

The subscript / len / iterate / call paths all raise correctly. It is specifically ATTRIBUTE access that answers None.

Why it matters more than it looks

This is the same failure shape as [[bug-nilpy-mixed-type-arithmetic-silently-does-pointer-math]]: a wrong value travelling far from its cause. A typo in an attribute name, or a dict.get that missed, yields None and the program carries on — and the first place it goes wrong is wherever that None is finally used, which may be nowhere near.

It also interacts with the deliberate None -> 0 coercion in pyvar_to_int (kept for the Optional[int] contract uforth depends on): a misspelled attribute becomes None, which then becomes 0 in a numeric context, and the arithmetic silently answers with a zero nobody wrote.

Note on dynamic attributes

NilPy supports genuinely dynamic attributes (obj.name = v stored in a global dict keyed by address + name, used by uforth for lazy state), so "not a declared field" cannot itself be the error condition — the lookup has to miss the dynamic table too before it raises. That is presumably why the miss path returns a default instead of raising, and it is the thing to be careful about when fixing: hasattr must keep working, and getattr(o, n, default) must still return the default rather than raise.

Gate

make test-nilpy + self-host byte-identical, plus a .npy covering the table above with CPython's own output, and explicitly: hasattr(o, 'x') False without raising, getattr(o, 'x', d) returning d, and a dynamic attribute set then read back.

Fixed separately while measuring this

None() raised a bare Exception carrying the text TypeError: ... rather than a TypeError, so except TypeError: missed it while except Exception: caught it. That raise lived in pyeval.pas and was the last of the family the [[bug-nilpy-pytypeerror-halts-instead-of-raising]] sweep missed (it only swept pylib.pas). Converted, and covered by test/test_nilpy_typeerror_is_catchable.npy.

CLOSED

Both measured causes fixed. pydynattr_get's miss branch now raises AttributeError — confirmed safe for hasattr/getattr, which resolve through pydynattr_has directly and never reach the miss branch of pydynattr_get at all. Split into two pylib procs so the error MESSAGE is safe to build in both cases the ticket's receivers can take: a class-typed receiver (obj is always a genuine object or nil) keeps pydynattr_get; a variant receiver (a for-loop element, d.get(k), an unannotated parameter) gets a new pydynattr_get_v, which checks the runtime TAG before ever touching .ClassName — the naive unwrap would have ClassName'd a str/int/ float/bool variant's reinterpreted payload, a new crash the ticket's own matrix does not cover but the general receiver shape (for variables, d.get()) very much reaches.

None.upper() (and the same for any str method on a non-str variant) is fixed separately: PyParseVariantMethod used to stringify the receiver via pystr_of unconditionally before dispatching the method, and pystr_of correctly renders None as 'None' (that part is not a bug — str(None) is supposed to say that) — the bug was calling a METHOD through that text. Fixed with a runtime tag guard (pyvar_is_strtag) that raises via a new pydynattr_no_method before the str conversion runs, whenever the receiver is not actually string-shaped. The receiver is hoisted into a hidden temp first since it is now read twice (guard, then conversion) and is not always side-effect-free — d.get(k).upper(), the ticket's own repro of the OTHER half of this bug, is exactly such a case.

Test: test/test_nilpy_missing_attribute_raises.npy, the full table from this ticket plus the hasattr/getattr/dynamic-attribute-round-trip guards the gate asked for and a genuine str-variant method call (the pre-existing, must-keep- working path). Gate: make test-nilpy green, self-host fixedpoint, testmgr --tier quick.

Ticket closed.

Log

• 2026-07-31 — resolved, commit 096035190.