[DRAFT] Dont track source for keyworld fields that are multi-field, unless specifically designated so

[Kubik42]

[DRAFT] (no tests yet)

Consider the following edge case:

"os": {
  "properties": {
    "name": {
      "type": "text",
      "fields": {
        "foo": {
          "type": "keyword",
          "ignore_above": 100
        }
      }
    }

If synthetic source is enabled, we'll have name.foo track source (current behavior). And if name.foo trips ignore_above, we'll create an extra StoredField to continue tracking source within the name.foo.

This however becomes a problem if there are multiple keyword multi fields, all of which trip ignore_above. In such a case, they'll all create an additional StoredField and we'll be storing the source multiple times, which is not space efficient.

This change designates one of the keyword multi fields as the one responsible for tracking source. Upon tripping ignore_above, this field will create a StoredField (same behavior as now), while the remaining keyword multi fields will not (new behavior).

[lkts]

Do you think we can make MultiFields.Builder an interface and do this stuff in a decorator or something inside TextFieldMapper? I know we have hasSyntheticSourceCompatibleKeywordField already but this is growing more and more.

[Kubik42]

I agree that its growing too much. I was also thinking about moving this logic somewhere closer to where it belongs. I'll see what I can do.

[lkts]

I am struggling with this name. Isn't this something like isUsedForSyntheticSourceByParent?

[Kubik42]

yeah... my thinking was to not expose this edge case as much through naming, but I can also see someone misunderstanding what this is used for. I'll change it

[lkts]

Same - usedByParentFieldForSyntheticSource?