super() in expression position: the refusal is right, the diagnostic is not

• Type: bug (Track N) — filed by Track D from [[docs-verify-nil-python-page-against-the-compiler]].
• Found: 2026-08-19 against pin v363. Low priority: the limitation is documented and has a working spelling; it is the diagnostic that costs time.

Repro

class P1:
    def hi(self) -> str:
        return "P"
class C1(P1):
    def hi(self) -> str:
        return "C"
    def viasuper(self) -> str:
        return super().hi()
print(C1().viasuper())

pascal26:1: error: Nil Python: annotate the type / too dynamic [a=22 b=8]
  near: >>>  P1

Line 1 is class P1:; the offending construct is on line 8. The message asks for an annotation, which cannot fix it, and prints raw node numbers. The supported form — Parent.method(self) — compiles and returns P correctly, so the fix is a diagnostic that says "super() is a statement form; write P1.hi(self)" at the real line.

Second, related finding

With multiple bases, calling the second base's initialiser explicitly is refused:

class C(A, B):
    def __init__(self) -> None:
        A.__init__(self)
        B.__init__(self)   # pascal26:10: error: class method not found: __init__

CPython runs this and sets both fields. Multiple bases otherwise work on v363 (methods from both are callable; first base wins on a name clash), so this is the one hole in them — it is what stops a two-base class from chaining both constructors.

Log

• 2026-08-19 — filed from a Track D documentation verification pass.