Fix parsing of flattened fields within subobjects: false

[felixbarny]

DocumentParser#doParseObject wrongly assumes that after parsing an object, we're not in a leaf object context anymore. While it is usually the case that there are no sub-objects within an object that's mapped as subobjects: false, there's one exception to the rule: when a field is mapped as flattened. While the DocumentParser does successfully parse objects that are mapped as flattened within a subobjects: false context, it sets context.path().setWithinLeafObject(false) after parsing a flattened object so that subsequent fields get their dots expanded.

[elasticsearchmachine]

Pinging @elastic/es-search (Team:Search)

[elasticsearchmachine]

Hi @felixbarny, I've created a changelog YAML for you.

[javanna]

LGTM, I'm afraid that flattened isn't the only exception, there's also rank features and percolator that are hit by the same bug perhaps. Thanks for finding and fixing. Does this fix a related issue? I did not see its link in the description.

[elasticsearchmachine]

💚 Backport successful

Status	Branch	Result
✅	8.13

[felixbarny]

Does this fix a related issue? I did not see its link in the description.

I don't think there's an issue that has been filed for this. I've detected the issue myself when playing around with subobjects: false and flattened fields in the context of #104455. It was a very strange bug to debug as it depends on ordering and because when ingest pipelines are executed, the ordering of fields change because fields are put into a hash map. Therefore, the ordering also depends on the name you choose for field names. It took me longer than I'd like to admit to realize that.