[WakaTaira]

dict and set replace the TypeError from a failed key hash with one naming the container, which has dropped the original traceback, cause and notes since 3.14. The original exception now survives when it holds something worth keeping, and {[]: 1} keeps its single-line message.

The original TypeError has	Result
A traceback or a cause (a Python __hash__)	Chained as __cause__
Only notes (a C-level failure, such as the tuple note in #161496)	Notes copied to the replacement
Neither	Same output as today

Output for a __hash__ that raises a TypeError with the note within Key.first:

Traceback (most recent call last):
  File "ks.py", line 5, in __hash__
    raise exc
TypeError: unhashable field
within Key.first

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "ks.py", line 7, in <module>
    {Key(): 1}
TypeError: cannot use 'Key' as a dict key (unhashable field)

Before this change, only the last line and its traceback appear.

Start with _PyErr_ReplaceTypeError in Python/errors.c. Three decisions are worth checking:

• __context__ does not trigger chaining. A C-level TypeError raised inside an except block gets the handled exception as its context, so checking it brings the duplicated unhashable type: 'list' message back. The replacement gets the same context anyway.
• Always chaining would print the inner message twice for {([1],): 1}, as described in the Exception notes discarded from TypeError within __hash__() #149313 discussion.
• Formatting errors stay untouched. If building the new message fails, for example because __str__ raises, that error propagates with its own cause.

Verified on Linux with a debug build: ./python -m test -j0, where only test_perf_profiler and test_profiling fail because of the local NixOS on WSL environment, and -R 3:3 on test_dict and test_set. I did not test the free-threaded build locally; it passes in this PR's CI.

A personal note: gh-161433 was my first issue on a project this large. I had a tuple-note prototype ready locally and planned to wait for comments before opening a PR. When I woke up, @localzet had already opened #161496, and @anujbharambe's #149505 and @localzet's fork experiment were there to learn from. This PR builds on both of them, and I wrote it now because I wanted to see this change through to a PR myself.

I used Claude Code to help with the implementation and tests, and I reviewed the result and can explain it.

🤖 Generated with Claude Code

https://claude.ai/code/session_01YSkzESiXGfCM1jUTyLpxHo

[python-cla-bot]

All commit authors signed the Contributor License Agreement.