A bare tuple returned from a mimic-module method becomes its repr

Repro — nine lines, against the real shim

from xml.sax.xmlreader import AttributesNSImpl
XLINK = "http://www.w3.org/1999/xlink"
attrs  = {(None, "class"): "hero",  (XLINK, "href"): "http://example.org/"}
qnames = {(None, "class"): "class", (XLINK, "href"): "xlink:href"}
a = AttributesNSImpl(attrs, qnames)
r = a.getNameByQName("xlink:href")
print("type:", type(r).__name__)
print("repr:", repr(r))
print("eq:",   r == (XLINK, "href"))

The shim is not at fault: mimic_xml_sax_xmlreader.py:122 iterates self._qnames and returns the KEY, which is the right answer and the right object.

THE ROUTE IS THE VARIABLE — three controls, and two of them do NOT reproduce

Written out because each one removes a suspect, and the first two are where the obvious hypotheses die:

• Dict iteration over tuple keys — CORRECT. Inline, with the test's exact shapes including the (None, "class") key: type: tuple, right repr, eq: True. This was my first hypothesis and it is wrong.
• A plain local imported module — CORRECT. A hand-written mymod.py with a class whose method returns a dict key, imported normally, returns a real tuple. So it is not "crossing a module boundary" and not "a method return".
• The mimic route — WRONG. Same construct, same types, through from xml.sax.xmlreader import ….

And a tuple inside a returned LIST survives. In the same run, (XLINK,"href") in a.getNames() is ok — getNames() returns a list of tuples and the tuples are intact. Only the BARE tuple return degrades, which is the sharpest clue available: something on the mimic return path marshals a top-level tuple to text while leaving a container's elements alone.

That asymmetry is also the positive control this ticket needs — 24 of 25 assertions pass, including every string, int, bool, and list-of-tuples row, so a fix that merely makes the file green without restoring type(r) is tuple has not fixed it.

Provenance — read this before re-attributing it

Auto-filed by Track T as regression-lib-test-lib-mimic-xml-sax-xmlreader at 2b2ec3fee1c5. That sha cannot be the cause and the auto-ticket says so itself: the job builds with $(PXX_STABLE) and that commit touched only compiler/** and devdocs/** — zero files the job reads. Verified with git show --name-only. The named sha is the one that was TESTED, not bisected.

Nor is it a v411 regression in the "was working" sense: v410's binary cannot compile the test at all against today's lib/ tree (AttributesNSImpl fails), so the pin is what made this row reachable. It is a pre-existing gap the pin EXPOSED, not one it introduced, and the honest framing is that 24 of 25 newly pass where previously none did.