Suspension: the magic behind composability (or: "The Kyo Monad")

Transcript

1. Suspension: the magic behind composability (or: "The Kyo Monad") Flavio Brasil - LambdaDays 2025
2. Principal Engineer @ Nubank Who am I?
3. Principal Engineer @ Nubank Open Source Developer @ getkyo.io Who am I?
4. Who am I? Principal Engineer @ Nubank Open Source Developer @ getkyo.io Just moved back to Brazil!
5. Who am I? Principal Engineer @ Nubank Open Source Developer @ getkyo.io Just moved back to Brazil! An engineer, not a mathematician
6. Introduction An overview of the project
7. • A monadic effect system based on algebraic effects What is Kyo?
8. • A monadic effect system based on algebraic effects • Leverages Scala 3's advanced type system to provide seamless effect tracking What is Kyo?
9. • A monadic effect system based on algebraic effects • Leverages Scala 3's advanced type system to provide seamless effect tracking • Properly suspends side effects and supports multi-shot continuations What is Kyo?
10. What is Kyo? • A monadic effect system based on algebraic effects • Leverages Scala 3's advanced type system to provide seamless effect tracking • Properly suspends side effects and supports multi-shot continuations • Provides a separate direct syntax dialect module and an advanced combinators module
11. What is Kyo? • A monadic effect system based on algebraic effects • Leverages Scala 3's advanced type system to provide seamless effect tracking • Properly suspends side effects and supports multi-shot continuations • Provides a separate direct syntax dialect module and an advanced combinators module • A real-world engineering project. Not experimental!
12. • 3 years in development (first year in private) Project Status
13. • 3 years in development (first year in private) • Evolving quickly with 22 releases so far Project Status
14. • 3 years in development (first year in private) • Evolving quickly with 22 releases so far • An active core developers team Project Status
15. Project Status • 3 years in development (first year in private) • Evolving quickly with 22 releases so far • An active core developers team • Comprehensive set of primitives
16. Project Status • 3 years in development (first year in private) • Evolving quickly with 22 releases so far • An active core developers team • Comprehensive set of primitives • Close to the 1.0-RC1 release
17. The magic behind composability Suspension
24. The "Pending" type
37. But… are we just tracking effects? 🤔
43. • Continuations are first-class values The power of suspension
44. • Continuations are first-class values • Can resume with arbitrary values. For example, injecting or transforming values The power of suspension
45. • Continuations are first-class values • Can resume with arbitrary values. For example, injecting or transforming values • Can resume zero, one, or n times The power of suspension
46. • Continuations are first-class values • Can resume with arbitrary values. For example, injecting or transforming values • Can resume zero, one, or n times • Unlocks next-level expressivity without language changes The power of suspension
47. Kyo from scratch Complete and minimalistic algebraic effect kernel
57. Expressing effects
67. Let's build the kernel :)
71. Transformation
76. Effect handling
85. Evaluation and automatic lifting
90. Example effects
100. The first of its kind? The Laws of Kyo
107. Where is the proof? 😬
108. An "approximate proof" 😅
109. Bridging the real-world gap Extending the kernel
110. • Stack safety is handled by safepoints before each transformation Bridging the real-world gap
111. • Stack safety is handled by safepoints before each transformation • Unboxed representation of plain pure values Bridging the real-world gap
112. • Stack safety is handled by safepoints before each transformation • Unboxed representation of plain pure values • Inlining as computation staging Bridging the real-world gap
113. • Stack safety is handled by safepoints before each transformation • Unboxed representation of plain pure values • Inlining as computation staging • Two kinds of effect: ContextEffect and ArrowEffect for performance + forking Bridging the real-world gap
114. • Stack safety is handled by safepoints before each transformation • Unboxed representation of plain pure values • Inlining as computation staging • Two kinds of effect: ContextEffect and ArrowEffect for performance + forking • Isolates to enable forking with effects Bridging the real-world gap
115. Conclusion
116. • Thank you to all contributors, in special Adam Hearn and John Hungerford for the long-time steady contributions! Conclusion
117. • Thank you to all contributors, in special Adam Hearn and John Hungerford for the long-time steady contributions! • Feedback is appreciated, even if negative :) Conclusion
118. • Thank you to all contributors, in special Adam Hearn and John Hungerford for the long-time steady contributions! • Feedback is appreciated, even if negative :) • We'd love to collaborate on a formal proof Conclusion
119. • Thank you to all contributors, in special Adam Hearn and John Hungerford for the long-time steady contributions! • Feedback is appreciated, even if negative :) • We'd love to collaborate on a formal proof • Keep an eye out for the 1.0-RC1! Conclusion
120. Thank you! 🙏