Partial Evaluation Cedar Meeting

Attendees

Experimental feature for the time being
Some customers are playing with it
By the time customers try it though, they find it too raw and give up
Common use cases
- Search: how do you map residual fragments to the relevant query language?
- Search: there is a risk you might hit a non-indexed field in the underlying DB
  -Hopelessness check: I don't have an entire request and I don't want to incur the cost of retrieving all attributes if I don't have all the information if I know I will get access denied.
- Impact analysis
  - What if I change this policy, how will access be impacted?
- Access reviews
  - What can Alice do? What can manager do?
Cedar already produces a JSON version of its AST that represents a partial evaluation response
- https://cedarland.blog/usage/partial-evaluation/content.html
We can compare it with the draft spec

The usefulness and scope of partial evaluation depends on the fact the underlying implementation is stateful or stateless

https://github.com/stalniy/ucast
https://github.com/StyraInc/ucast-linq
Integration with Prisma
- https://www.npmjs.com/package/@styra/ucast-prisma
- https://www.prisma.io/
Challenge
- ucast is not a standard
- ucast doesn't define a serialization format in the OS project
- check with Styra
- the last commit is nearly 2 years old

If partial evaluation is about data filtering, then the target is data platforms (in a broad sense) such as SQL DB vendors, data platforms (Trino, Immuta, Snowflake), or DB SaaS (Athena, RDS…)