Segfaults when loading model with local functions, works fine if model is inlined by ONNX

[BowenBao]

Describe the issue

As title, here is the example model.

To reproduce

Download model from link above. Install onnx from source (for inliner, this step is not necessary to repro the segfault), onnxruntime from nightly.

pip install --index-url=https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/ ort-nightly==1.16.0.dev20230528001

import onnxruntime

# Segfaults
sess = onnxruntime.InferenceSession("segfault.onnx", providers=["CPUExecutionProvider"])

# Works if model is inlined first.
# Needs onnx main branch for inliner.
import onnx
import onnx.inliner
model_proto = onnx.load("segfault.onnx")
inlined_model_proto = onnx.inliner.inline_local_functions(model_proto)
sess = onnxruntime.InferenceSession(inlined_model_proto.SerializeToString(), providers=["CPUExecutionProvider"])

Urgency

No response

Platform

Linux

OS Version

20.04

ONNX Runtime Installation

Built from Source

ONNX Runtime Version or Commit ID

1.16.0.dev20230528001

ONNX Runtime API

Python

Architecture

X64

Execution Provider

Default CPU

Execution Provider Library Version

No response

[BowenBao]

gdb shows this is likely the same issue as onnx/onnx#5212, so should be resolvable by bumping onnx submodule commit to onnx/onnx@213b525

Thread 1 "python" received signal SIGSEGV, Segmentation fault.
onnx::shape_inference::ShapeInferenceImplBase::process (this=this@entry=0x7fffffff9e50, func_proto=..., ctx=...)
    at /home/bowbao/onnxruntime/build/Linux/RelWithDebInfo/_deps/onnx-src/onnx/shape_inference/implementation.cc:591
591           if (type->value_case() == TypeProto::kTensorType && ctx.getInputData(i) != nullptr) {

[snnn]

@pranavsharma , how should we proceed?
@liqunfu , will ONNX have a patch release for this bug? Or will they cherry-pick the fix to a release branch?

[pranavsharma]

As discussed we've an internal work item to add a cmake option to allow build ORT with a different ONNX Commit.

[BowenBao]

@pranavsharma thanks for info, is there an eta for this item?

[thiagocrepaldi]

FYI Onnx 1.14.1 is coming by the end of August (onnx/onnx#5468)

[pranavsharma]

@pranavsharma thanks for info, is there an eta for this item?

Check with @snnn