class S(tuple[*tuple[()]]) crashes mypy

[jorenham]

class S(tuple[*tuple[()]]): ...

mypy 1.15 stacktrace (mypy-play):

Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "/layers/google.python.pip/pip/lib/python3.12/site-packages/mypy/__main__.py", line 37, in <module>
    console_entry()
  File "/layers/google.python.pip/pip/lib/python3.12/site-packages/mypy/__main__.py", line 15, in console_entry
    main()
  File "mypy/main.py", line 119, in main
  File "mypy/main.py", line 203, in run_build
  File "mypy/build.py", line 191, in build
  File "mypy/build.py", line 267, in _build
  File "mypy/build.py", line 2937, in dispatch
  File "mypy/build.py", line 3335, in process_graph
  File "mypy/build.py", line 3430, in process_stale_scc
  File "mypy/semanal_main.py", line 97, in semantic_analysis_for_scc
  File "mypy/semanal.py", line 7621, in apply_semantic_analyzer_patches
  File "mypy/semanal.py", line 2587, in <lambda>
  File "mypy/semanal_shared.py", line 305, in calculate_tuple_fallback
NotImplementedError

and on current master (mypy-play)

Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "mypy/__main__.py", line 37, in <module>
    console_entry()
  File "mypy/__main__.py", line 15, in console_entry
    main()
  File "mypy/main.py", line 127, in main
    res, messages, blockers = run_build(sources, options, fscache, t0, stdout, stderr)
                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "mypy/main.py", line 211, in run_build
    res = build.build(sources, options, None, flush_errors, fscache, stdout, stderr)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "mypy/build.py", line 191, in build
    result = _build(
             ^^^^^^^
  File "mypy/build.py", line 267, in _build
    graph = dispatch(sources, manager, stdout)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "mypy/build.py", line 2939, in dispatch
    process_graph(graph, manager)
  File "mypy/build.py", line 3337, in process_graph
    process_stale_scc(graph, scc, manager)
  File "mypy/build.py", line 3432, in process_stale_scc
    mypy.semanal_main.semantic_analysis_for_scc(graph, scc, manager.errors)
  File "mypy/semanal_main.py", line 99, in semantic_analysis_for_scc
    apply_semantic_analyzer_patches(patches)
  File "mypy/semanal.py", line 7740, in apply_semantic_analyzer_patches
    patch_func()
  File "mypy/semanal.py", line 2630, in <lambda>
    self.schedule_patch(PRIORITY_FALLBACKS, lambda: calculate_tuple_fallback(base))
                                                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "mypy/semanal_shared.py", line 305, in calculate_tuple_fallback
    raise NotImplementedError

---

another way to trigger this crash is through a typo in the variadic type parameter name:

class MyTuple[*Ts](tuple[*_Ts]): ...  # oh no a typo

this produces an identical stacktrace (mypy-play)

---

some observations:

• it doesn't matter if typing.Unpack is used instead of *
• I checked all versions from 1.9.0-master, and all crash in the same way
• class S(tuple[()]): ... does not cause a crash
  (although S() is revealed as tuple[(), fallback=__main__.S], which is kinda sus)
• in case you didn't notice; the first example is valid at runtime, the second isn't
• my ear is a bit itchy

And it's probably not what you want to hear, but it might be relevant info, so I'll carefully whisper it: _{Pyright does not have this problem (playground 1, playground 2), and the same goes for ty (playground 1, playground 2)}

---

ref: #19100 (comment)

[sterliakov]

Ough, note this isn't specific to empty tuples, the following crashes just as hard:

class S(tuple[*tuple[int]]): ...

[sterliakov]

Quick braindump: we normalize such trivial tuples in semanal_typeargs, but TypeArgumentsAnalyzer runs after we have calculated all fallbacks. Fallback calculation logic assumes normalized inputs, which is most probably wrong. I'm not sure whether what I did is the best approach, but at least it doesn't crash anymore - let's continue this discussion in the linked PR.