Currently, we visit fields, variables, arguments, and object fields in DocumentNode, collect their return types and parent types, and turn them into schema coordinates. When a schema is modified, Hive creates a list of touched schema coordinates and checks in ClickHouse if a schema coordinate was registered or not and if it’s a breaking change or not.

This is too safe 🙂

My proposal is to:

• define a set of scenarios
• create a UI to enable/disable some in Hive (if needed)
• create a tool to test each rule, to check if the described scenario is in fact a breaking change or not (document + schema before → document + schema after = error or not)
• think about Hive and how to store data
• lastly, think about implementation (but really at the very very end of the discussion 🙂)
  • consider Rust (synchronous processing will be heavy on the event loop)

Scenarios**:**

Glossary

__typename - means __typename but also fragments

consumer - a code dealing with the GraphQL response object

reachable - reachable at some level by root-level fields

Renaming

Renaming means changing the name but keeping the exact same shape.

Object type

• breaking if GraphQL consumer depends on __typename.

Interface type

• safe, a user never gets to see it.

Input object