How to use a property from parent resource in child?

How can I use the property nextVersion from resource mlEnvironment from my template below:

@description('The list of Azure ML workspace names to deploy environments to.')
param workspaceName string

@description('The list of environments to create. Each object should have: environmentName, environmentVersion.')
param environments array

resource mlWorkspace 'Microsoft.MachineLearningServices/workspaces@2025-06-01' existing = {
  name: workspaceName
}

resource mlEnvironment 'Microsoft.MachineLearningServices/workspaces/environments@2025-07-01-preview' = [
  for env in environments: if(length(environments)>0) {
    parent: mlWorkspace
    name: env.environmentName
    properties: {
      isArchived: false
  }
}]

resource mlEnvironmentVersion 'Microsoft.MachineLearningServices/workspaces/environments/versions@2025-07-01-preview' = [
  for (env, i) in environments: if(length(environments)>0) {
    parent: mlEnvironment[i]
    name: mlEnvironment[i].properties.nextVersion
    properties: {
      description: env.environmentDescription
      condaFile: env.condaFile
    osType: env.osType
    isAnonymous: false
    isArchived: false
    image: (empty(env.environmentImage) ? null : env.environmentImage)
  }
}]

If I do name: mlEnvironment[i].properties.nextVersion from template above, I get this error:

This expression is being used in an assignment to the "name" property of the "Microsoft.MachineLearningServices/workspaces/environments/versions" type, which requires a value that can be calculated at the start of the deployment. Properties of mlEnvironment which can be calculated at the start include "apiVersion", "id", "name", "type".bicepBCP120
The value of type "Microsoft.MachineLearningServices/workspaces/environments | null" may be null at the start of the deployment, which would cause this access expression (and the overall deployment with it) to fail.bicepBCP318

I tried using user defined functions but also didn't work.

Solution

You have to use modules to achieve that. It is not possible without using modules.

main.bicep

@description('The list of Azure ML workspace names to deploy environments to.')
param workspaceName string

@description('The list of environments to create. Each object should have: environmentName, environmentVersion.')
param environments array

resource mlWorkspace 'Microsoft.MachineLearningServices/workspaces@2025-06-01' existing = {
  name: workspaceName
}

resource mlEnvironment 'Microsoft.MachineLearningServices/workspaces/environments@2025-07-01-preview' = [
  for env in environments: if(length(environments)>0) {
    parent: mlWorkspace
    name: env.environmentName
    properties: {
      isArchived: false
  }
}]

module versions 'versions.bicep' = [for (env, i) in environments: if(length(environments)>0) {
  name: 'version-${env.environmentName}-${i}'
  params: {
    env: env
    workspaceName: workspaceName
    nextVersion: mlEnvironment[i].properties.nextVersion
  }
}]

versions.bicep

param env object
param workspaceName string
param nextVersion string

resource mlWorkspace 'Microsoft.MachineLearningServices/workspaces@2025-06-01' existing = {
  name: workspaceName
}

resource mlEnvironment 'Microsoft.MachineLearningServices/workspaces/environments@2025-07-01-preview' = {
  parent: mlWorkspace
  name: env.environmentName
  properties: {
    isArchived: false
  }
}

resource mlEnvironmentVersion 'Microsoft.MachineLearningServices/workspaces/environments/versions@2025-07-01-preview' = {
  parent: mlEnvironment
  name: nextVersion
  properties: {
    description: env.environmentDescription
    condaFile: env.condaFile
    osType: env.osType
    isAnonymous: false
    isArchived: false
    image: (empty(env.environmentImage) ? null : env.environmentImage)
  }
}