Add IdentityFixPass to fix invalid graphs with direct input-output connections

[Copilot]

ONNX specification prohibits graph inputs from being directly used as graph outputs without intermediate nodes. This creates invalid graphs that fail validation.

Changes

• New pass: IdentityFixPass in src/onnx_ir/passes/common/identity_fix.py

  • Detects outputs that are direct graph inputs (via is_graph_input())
  • Inserts Identity nodes between them
  • Recursively processes subgraphs and functions
  • Preserves metadata, shapes, types, and doc strings

• Tests: 12 test cases in identity_fix_test.py covering:

  • Single and multiple direct connections
  • Subgraphs (If/Loop nodes)
  • Functions
  • Metadata preservation

• Export: Added to __init__.py for public API

Usage

from onnx_ir.passes.common import IdentityFixPass

# Invalid: input directly used as output
graph = ir.Graph(inputs=[x], outputs=[x], nodes=[])

# Apply pass to fix
result = IdentityFixPass()(ir.Model(graph, ir_version=10))
# Now contains Identity node: input -> Identity -> output

Ref: pytorch/pytorch#167848

Original prompt

---

This section details on the original issue you should resolve

<issue_title>[Passes] Add identity fix</issue_title>
<issue_description>Create a pass that adds an Identity node if an input of a graph is directly set as the output of the graph. This will turn an invalid graph into a valid one.

Ref: pytorch/pytorch#167848</issue_description>

<agent_instructions>Add the pass in src/onnx_ir/passes/common</agent_instructions>

Comments on the Issue (you are @copilot in this section)

• Fixes [Passes] Add identity fix #267

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[codecov]

Codecov Report

❌ Patch coverage is 97.22222% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 77.64%. Comparing base (ac8138e) to head (d247885).
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/onnx_ir/passes/common/output_fix.py	97.14%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #268      +/-   ##
==========================================
+ Coverage   77.33%   77.64%   +0.31%     
==========================================
  Files          40       41       +1     
  Lines        5042     5078      +36     
  Branches     1009     1016       +7     
==========================================
+ Hits         3899     3943      +44     
+ Misses        853      843      -10     
- Partials      290      292       +2     

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