Unexpected behavior with Helm chart values

[kittywaresz]

Describe the bug
Please note, this bug relates only to the Helm chart implementation, not gpu-operator itself

I would like to have an option to install gpu-operator using Helm chart with migManager.config.default set to "" (empty string) , since in my case it is strongly required to not update MIG profiles right after the installation

But currently there is no way to set "" as a value of migManager.config.default

The reason of such behavior is a Helm template rendering logic, when you specify "" as a value of migManager.config.default, it renders as part of the ClusterPolicy resource and looks like this:

spec:
  migManager:
    config:
      default:       <- the problem is here

It means that migManager.config.default property implicitly has a value of null, which is interpreted during the ClusterPolicy creation as an empty value, so the default value comes into place and which is all-disabled

I made a small fix to prevent this behavior and now it looks as expected:

spec:
  migManager:
    config:
      default: ""

To Reproduce
I tried to explicitly set migManager.config.default property to empty string:

cat << EOF | helm --namespace gpu-operator install gpu-operator nvidia/gpu-operator \
  --create-namespace \
  --version=v25.3.3 \
  --values -
mig:
  strategy: mixed
migManager:
  config:
    name: my-custom-conf
    create: false
    default: ""
EOF

Expected behavior
I would like to see ClusterPolicy resource created with migManager.config.default set to "" (empty string), not all-disabled --> no nvidia.com/mig.config label present on any MIG capable nodes, mig-manager pod in running state on these nodes

Environment (please provide the following information):

• GPU Operator Version: v25.3.3
• OS: Ubuntu22.04
• Kernel Version: 5.15.0-157-generic
• Container Runtime Version: v1.7.24 88bf19b2105c8b17560993bee28a01ddc2f97182
• Kubernetes Distro and Version: K8s v1.32.4