[Do Not Merge]Migrate onnxscript converter to use onnx ir

[gramalingam]

Migrate onnxscript converter to use onnx ir.

PLEASE DO NOT MERGE YET. May want to wait for release 0.6

[codecov]

Codecov Report

❌ Patch coverage is 82.46154% with 57 lines in your changes missing coverage. Please review.
✅ Project coverage is 70.10%. Comparing base (c1bfdfc) to head (86e10dd).
⚠️ Report is 3 commits behind head on main.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
onnxscript/irbuilder.py	73.52%	15 Missing and 3 partials ⚠️
onnxscript/converter.py	89.87%	9 Missing and 7 partials ⚠️
onnxscript/values.py	76.19%	9 Missing and 6 partials ⚠️
onnxscript/converter_test.py	80.76%	3 Missing and 2 partials ⚠️
...ewriter/ort_fusions/fused_matmul_rule_sets_test.py	0.00%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2706      +/-   ##
==========================================
+ Coverage   70.04%   70.10%   +0.06%     
==========================================
  Files         226      226              
  Lines       27177    27132      -45     
  Branches     2734     2739       +5     
==========================================
- Hits        19036    19022      -14     
+ Misses       7197     7165      -32     
- Partials      944      945       +1     

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

To fix this issue, the empty except AttributeError: block should be replaced with exception handling that does something useful. The most appropriate fix is to log a warning (using the module's logger) indicating that an AttributeError was encountered while setting the type info, which helps diagnose and debug the cause without terminating program execution. This must be done in the function set_type_info in onnxscript/converter.py. No new methods or imports are required, as the logger logger is already defined and available in the module. The fix is local: replace pass with a logging call in the specified region.

The best fix is to properly handle the exception by notifying the user or logging the error, so that missing or erroneous type information does not go unnoticed. At minimum, add a logging statement at the appropriate level (e.g., warning or error), describing what happened and including useful contextual information. This way, if an AttributeError is raised, it is recorded and is much less likely to go undetected.

Specifically, edit the file onnxscript/irbuilder.py, in the function set_type_info, replacing the empty except AttributeError: block (lines 130–131) with an except block that logs the exception with a helpful message, including details about the value and typeinfo involved.

No new imports are needed since logging is already imported, and the logger onnxscript is already defined as logger.

[justinchuby]

Probably default the minimal version to a more recent one, like v10, such that the generated model is compatible with newer tool chains (metadata)? This can also be a good default to push for adoption

[justinchuby]

I would consider adding inline noqas for TID251 so we allow the helper module in a case by case basis. There is also a possibility that it is not needed.

[justinchuby]

Curious why we do this instead of converting from an IR Model?