Improves implementation of aten_index_put

[xadupre]

From #2606.

[codecov]

Codecov Report

❌ Patch coverage is 87.03704% with 7 lines in your changes missing coverage. Please review.
✅ Project coverage is 70.12%. Comparing base (5be9d3b) to head (db53ee1).
⚠️ Report is 9 commits behind head on main.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
onnxscript/function_libs/torch_lib/ops/core.py	87.03%	4 Missing and 3 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2641      +/-   ##
==========================================
- Coverage   70.43%   70.12%   -0.32%     
==========================================
  Files         224      225       +1     
  Lines       26778    27168     +390     
  Branches     2680     2738      +58     
==========================================
+ Hits        18862    19052     +190     
- Misses       6985     7166     +181     
- Partials      931      950      +19     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[gramalingam]

Adding some pointers/info for my own clarification:

• torch.Tensor.index_put_ documentation
• C++ documentation
• C++ implementation

[gramalingam]

Why do we need the above line? Just wondering ... shouldn't it already have dtype set?

[xadupre]

Maybe it is not useful anymore but when I did this PR, it was needed.

To fix the mixed explicit/implicit returns, we should ensure every code path in aten_index_bool ends with an explicit return. Since all explicit returns are of type TensorType, and the function signature declares this, the implicit fallthrough at the end should return a value of the same type. If an error or unexpected condition occurs (e.g., all indices are None and the loop doesn't yield a result), it makes sense to raise an error or explicitly return None (if downstream code safely handles this), but more robustly, an exception is best for unreachable/invalid cases. If returning None is preferable, make it explicit. Given the function signature expects a TensorType, raising an informative error is the most readable solution, but returning None also satisfies the CodeQL requirement for explicitness if that's been the behaviour.

Thus, add return None or raise ValueError("No valid indices provided to aten_index_bool") to the end of the function—choose return None to preserve existing behaviour, unless a contract is required.

Only lines inside aten_index_bool (lines 4366–4404) need fixing in file onnxscript/function_libs/torch_lib/ops/core.py.

---