Migrate translated entity reference from D7 to D8

Task: Migrate multilingual content types A and B from D7 to D8. Content type A has a reference to content type B.

Note: In this scenario, the Content Translation module is being used for D7, so each translation is an individual node.

My current solution/workarounds for this issue:

• Create 4 migrations: original content type A, translations of content type A and original content type B, translations of content type B. I needed to do that because if I specify that I want to migrate translations, the original node(n.nid = n.tnid) isn't migrated.
• Migration for original content type A references original content type B with a migration process plugin.
• Migration for translations of content type A references translations of content type B with a migration process plugin.
• Need patch to remove langcode as an ID.

If there's a better solution/workaround for this, I would be glad to know and document it somewhere. Here are a sample of my migration config files:

File: migrate_plus.migration.node_a.yml

id: node_a
label: Node A
migration_group: example
source:
  plugin: d7_node
  key: migrate
  node_type: node_a
destination:
  plugin: entity:node
process:
  title: title
  langcode: language
  type:
    plugin: default_value
    default_value: node_a
  field_reference_node_b:
    plugin: iterator
    source: field_reference_node_b
    process:
      target_id:
        plugin: migration
        migration: node_b
        source: target_id

File: migrate_plus.migration.node_a_translations.yml

id: node_a_translations
label: Node A Translations
migration_group: example
source:
  plugin: d7_node
  key: migrate
  node_type: node_a
  translations: true
destination:
  plugin: entity:node
  translations: true
  content_translation_update_definitions:
    - node
migration_dependencies:
  required:
    - node_a
process:
  nid:
    plugin: migration
    migration: node_a
    source: tnid
  title: title
  langcode: language
  content_translation_source: source_langcode
  type:
    plugin: default_value
    default_value: node_a
  field_reference_node_b:
    plugin: iterator
    source: field_reference_node_b
    process:
      target_id:
        plugin: migration
        migration: node_b_translations
        source: target_id

File: migrate_plus.migration.node_b.yml

id: node_b
label: Node B
migration_group: example
source:
  plugin: d7_node
  key: migrate
  node_type: node_b
destination:
  plugin: entity:node
process:
  title: title
  type:
    plugin: default_value
    default_value: node_b

File: migrate_plus.migration.node_b_translations.yml

id: node_b_translations
label: Node B translations
migration_group: example
source:
  plugin: d7_node
  key: migrate
  node_type: node_b
  translations: true
destination:
  plugin: entity:node
  translations: true
process:
  nid:
    plugin: migration
    migration: node_a
    source: tnid
  title: title
  langcode: language
  type:
    plugin: default_value
    default_value: node_b

@Gabor, comment #2 was to show the solution/workaround I got so far. For the "main content type" I do the migration using the tnid as the new nid and it works fine. The problem happens when I add the node reference field, because out of the box when I migrate the referenced content and use the migration process to get the referenced nid, the target_id is returned as an array of id and langcode, instead of an int/string with the id.

So, the workaround outlined in #2 was to remove langcode as an ID(with the patch attached) and import each referenced entity with a new nid. Because I thought(wrongly) I would have problems using repeated nid's as target ID.

I did some new tests and it actually works if I migrate all referenced translations to one node entity, so that isn't an issue, I updated the comment my latest solution, sorry about the mix up.

The actual issue now is the error in the description. The patch on #2 makes it work on my case, but I imagine it's a bad idea to just drop the langcode as an destination ID? Maybe we should do this conditionally? Or at the getValue function for the entity reference field use only the first value if an array is given?

The language is dropped as a destination id but it's still used to create/update the entity. So for an example in D7 where nid 123 is the original en content and 124 is the fr translation, the mapping table would look something like:

   souceid1 | destinationid1
    123     |        1
    124     |        1

And then the entity destination plugin uses the given langcode to decide if it will create/update a node or a translation. I imagine that in this case where in the D7 source translations have different nid's we don't need the langcode as a destinationid2, but maybe it would be useful in a setup where the D7 source has the same nid for all translations, like with Entity Translation, I think.

That makes sense, though it looks like that the Entity Reference field doesn't need the language to select the proper translation for the given field, so I'm not sure why we need to send the language as an ID to the destination. Maybe we could just have it as an ID for mapping?(Not sure if such concept exists in the Migrate API)