Segfault in TensorRT 10.13.3 ONNX parser

Description

There seems to be a regression in the TensorRT 10.13 ONNX parser causing a segmentation fault when parsing a simple identity model (a toy model I'm using in a unit-test). The crash occurs when calling tensorrt.OnnxParser.parse() in Python, but it's also reproducible with trtexec:

$ /usr/src/tensorrt/bin/trtexec --onnx=ident.onnx --minShapes=__input:1x512x512x3 --optShapes=__input:1x512x512x3 --maxShapes=__input:1x512x512x3 --shapes=__input:1x512x512x3
...
[09/24/2025-11:57:43] [I] Start parsing network model.
[09/24/2025-11:57:43] [I] [TRT] ----------------------------------------------------------------
[09/24/2025-11:57:43] [I] [TRT] Input filename:   ident.onnx
[09/24/2025-11:57:43] [I] [TRT] ONNX IR version:  0.0.7
[09/24/2025-11:57:43] [I] [TRT] Opset version:    13
[09/24/2025-11:57:43] [I] [TRT] Producer name:    tf2onnx
[09/24/2025-11:57:43] [I] [TRT] Producer version: 1.9.1
[09/24/2025-11:57:43] [I] [TRT] Domain:           
[09/24/2025-11:57:43] [I] [TRT] Model version:    0
[09/24/2025-11:57:43] [I] [TRT] Doc string:       
[09/24/2025-11:57:43] [I] [TRT] ----------------------------------------------------------------
Segmentation fault

This error doesn't occur with TensorRT 10.12 or 10.11.

Environment

TensorRT installed on Ubuntu via APT.

TensorRT Version: 10.13.3

NVIDIA GPU: RTX 4080

NVIDIA Driver Version: 575.57.08

CUDA Version: 12.9

CUDNN Version: 8.9.7

Operating System: Ubuntu 24.04

Python Version (if applicable): 3.13

Relevant Files

ident.onnx in ident.zip or as Python bytes

b"\x08\x07\x12\x07tf2onnx\x1a\x051.9.1:\x8f\x01\x12\x07tf2onnxR\x18converted from TestModelZ4\n\x05input\x12+\n)\x08\x01\x12%\n\t\x12\x07unk__29\n\t\x12\x07unk__30\n\t\x12\x07unk__31\n\x02\x08\x03b4\n\x05input\x12+\n)\x08\x01\x12%\n\t\x12\x07unk__32\n\t\x12\x07unk__33\n\t\x12\x07unk__34\n\x02\x08\x03B\x02\x10\r"

Steps To Reproduce

Commands or scripts: trtexec --onnx=ident.onnx

Start by reproducing the crash with ident.onnx using trtexec --onnx=ident.onnx, then compare the behavior with TensorRT 10.12 or 10.11. Check the tensorrt.OnnxParser.parse() path as well as the trtexec path; done means the identity model parses without a segmentation fault on TensorRT 10.13.3.