This repo is an attempt to quantify how thrift types can be changed.

Specifically what is interesting to know is whether a change to a type will cause runtime exceptions or whether thrift/scrooge will attempt to do some sort of corrective action.

This is useful when designing API's and answering questions like

• Should this type be a union or just a struct with optional fields?
• Do I want to use an enum or should we stick with a string and parse the enum later?

Types of mutations to the schema

• Positions
  • input
  • output
• type of type
  • struct
  • enum
  • union
• type of change
  • add "normal"
  • add required
  • add optional
  • change type
  • required -> "normal"
  • required -> optional
  • "normal" -> required
  • "normal" -> optional
  • optional -> required
  • optional -> "normal"
  • add element to co-products
  • remove element from co-product
  • rename element in co-products
  • change type in element in coproduct
  • TypeDef to "regular"
  • "regular" to TypeDef

Conclusions

• Input
  • You cant change types, ever. It will throw.
  • You cant violate the required keyword (neither by removing the value nor making it optional and sending none). It will throw.
  • With Enums you send the number, not the label. Missing label -> Unknown Value
  • With unions a mutated type will throw. Missing element -> Unknown value. Names do not have semantic meaning.
  • When a "normal" value is missing it gets the default value for its type, if such a value exists (see https://twitter.github.io/scrooge/Semantics.html)
• Output
  • You cant change types, ever. It will throw.
  • You cant violate the required keyword (neither by removing the value nor making it optional and sending none). It will throw.
  • With Enums you send the number, not the label. Missing label -> Unknown Value
  • With unions a mutated type (OR NAME) will throw. Missing element -> Unknown value
  • When a "normal" value is missing it gets the default value for its type, if such a value exists (see https://twitter.github.io/scrooge/Semantics.html)

Typedefs doesn't seem to carry any meaning :()