hasattr with a COMPUTED name cannot see a builtin method

Filed 2026-08-26 while resolving [[bug-n-hasattr-through-an-untyped-parameter-is-always-false]] — measured, not inferred, and deliberately NOT folded into that fix because the mechanism is a different one.

Measured (self-hosted at the fix's sha, and identical before it)

def lit(x):
    return hasattr(x, 'keys')          # the LITERAL path
def comp(x, n):
    return hasattr(x, n)               # the COMPUTED path

d = {'a': 1}
print(lit(d), comp(d, 'keys'))         # CPython True True   pxx True False
print(lit('s'), comp('s', 'upper'))    # CPython False True  pxx False False

Why the two paths differ

They are two mechanisms answering one question — the exact double case devdocs/dev/normalise-dont-special-case.md is about, and the computed one is the arm that stayed broken.

• Literal name → pasparser_expr.inc / PyParseFactorCore's hasattr arm, which asks the FRONTEND's tables (PyMethNameFor / PyPylibMethodAlias, PyStrMethodInfo, PyIsIntMethodName, PyIsFloatMethodName) and, for a variant receiver, emits PyHasAttrRuntimeChain — run-time tag tests built from those same tables.
• Computed name → PyMakeDynAttrByExpr → pylib's pydynattr_has_any_v, which is pure run time: the dynamic store, PyDeclaredAttrGet, PyPropertyGet, PyFindMethByName(GetInstanceRTTI(obj), …), __getattr__. It has no alias table (the RTTI knows keylist, the user wrote keys) and it Exits outright when pyvartag(v) <> 7, so no scalar — str, int, float, bool — is ever consulted at all.

Shape of the fix — and the reason it is a separate ticket

The frontend cannot answer a computed name, so this half has to be answered in pylib, which means pylib needs what only the frontend has today: the Python→Pascal method alias table and the str/int/float method name sets. Hand-copying them into pylib is exactly the drift that [[bug-nilpy-hasattr-on-a-builtin-container-or-str-answers-false]] was, so the honest options are

1. Emit the tables — have the frontend lower a per-program table (or a generated pyattr_exists_static(tag, name) helper) that both paths consult, so the alias/str/int/float knowledge has exactly one author.
2. Widen the RTTI for pylib container classes so the Python spelling is findable at run time, and give pydynattr_has_any_v scalar-tag arms.

(1) keeps one source of truth and is the direction [[bug-nilpy-hasattr-on-a-builtin-container-or-str-answers-false]]'s resolution already argued for. Either way it is a pylib change, so it needs a pin to reach Track B/E.

Gate

The table above as a .npy with a CPython-generated .expected, each computed row beside its literal sibling so the two arms disagreeing is what fails; plus getattr(x, n) on the same receivers, which is this predicate's partner and should be measured at the same time.