Github user justinleet commented on the issue:
https://github.com/apache/metron/pull/995
I agree, I'm not seeing how partial updates solves the general problem of
"We need to avoid updates against derived fields". End users aren't going to
necessarily know what a generated field is. Say a data scientist writes a
machine learning model, and an analyst wants to update some output of the
model. They may not know, or even be able to easily discover, what fields are
poly types and what aren't. They'll just get an error. I'm not even convinced
it'll be an easily understandable error to anyone not familiar with Solr.
It seems like there one real, solution to this problem that doesn't
compromise usability: Don't pass sub fields back in queries.
This can be something in the schema itself (e.g. does docValues change
work). It could also be something determined while running (e.g. LukeRequest
if it turns out to be viable or some other request / query).
---
