A shim class reports its Pascal spelling from `type(x).name`

Measured 2026-09-21 at pxx 8e60c44be, and again under the pinned compiler, which is how it was attributed: it turned up as a false hit in an unrelated change's positive control, and reproducing it on the pin is what separated "pre-existing" from "I just broke this".

import collections
d = collections.deque()
print(type(d).__name__)

CPython 3.14.4 : deque
pxx            : TPyDeque

Why it exists

A shim cannot always spell the Python name directly — lib/rtl/pil.pas records the sharp version of this, where declaring a class named after a unit in its own uses clause silently miscompiles the constructor into a self-call. The established workaround is to declare the class under a Pascal-ish name and publish a type ALIAS:

Image = TPILImage;

The alias is enough for construction, for annotation and for qualified access, so everything a program normally does keeps working. __name__ is derived from the DECLARED class, so it keeps reporting the internal spelling.

Why the prio is 25 and not lower

It is introspection-only and no demo depends on it, which is the whole argument for a low number. What keeps it from being rejected/:

__name__ is an instrument, and this is an instrument lying. The tree's own SizeOf ruling says a truthful instrument reporting a surprising answer is not a defect — "sizeof reported CORRECTLY about the accurate type". This is the other case: TPyDeque is not a true statement about anything the program can name. There is no pxx-level type called TPyDeque that a NilPy program could have written or could reference.
It is a CLASS and it grows. Every alias-around-a-collision shim has it. Two exist today (deque, TPILImage); the pattern is the recommended workaround, so the population only goes up.
The failure is silent and the value is plausible, which is this tree's expensive shape.

Not the fix

Renaming the Pascal classes to their Python spellings. That is what the alias exists to avoid, and in pil.pas's case it reintroduces a silent miscompile.

What would probably fix it

Carry the Python-facing name on the class and have __name__ prefer it — the alias already records the intended spelling, so the information exists and is simply not reaching the accessor. Whoever takes this should check what __qualname__, repr() and isinstance error text do with the same name, since a partial fix here is the sibling-spelling trap.

What would retire this