← board

A module-level ANNOTATED global is not visible in a keyword argument

from dataclasses import dataclass, field

@dataclass
class R:
    evidence: list[str] = field(default_factory=list)

pe: dict[str, list[str]] = {}
pe["C"] = []
a = R(evidence=pe.get("C", []))     # error: undefined variable (pe)
print(len(a.evidence))



pascal26:9: error: undefined variable (pe)
  near:  R  evidence  pe >>>  get

The same name is visible everywhere else — len(pe), pe["C"] = [], and a plain call argument f(pe.get("C", [])) all compile. Only the keyword argument of a class construction fails to resolve it.

Dropping the annotation (pe = {}) compiles, so it is the ANNOTATED module-level assignment specifically: that path registers the global somewhere the dataclass keyword-argument parser does not consult.

Found while building a repro for [[bug-nilpy-slice-of-variant-local-returned-is-unusable]]; unrelated to it.

Fixed 2026-07-30 — and it was never about kwargs

Narrowed before fixing: pe2 = pe fails identically, and so does every annotation kind including a bare int. The rule is "a module-level statement whose RIGHT-HAND SIDE is trial-parsed cannot see an annotated global"; the dataclass keyword argument in the original repro was simply the first such statement in that file. print(len(pe)) and pe["C"] = [] are not trial-parsed, which is why they looked fine and made this read as a kwarg bug.

Cause: PyCollectModuleLocalsAST's annotated arm called PyNoteLocalType, which records the name for the NEXT round's seeding loop, and nothing put it in the CURRENT round's scratch scope. The "undefined variable" that followed is a fatal Error, so the round that would have known the name never ran. The arm now allocates it as well, exactly as the seeding loop does.

Gate

The snippet above compiling and printing 0, plus make test-nilpy.

Log