Purity inference

[timonkrebs]

Purity inference in C# could be implemented by providing a new keyword, such as pure, placed in the same location as async. (Because async functions would never be pure, this would not lead to less readable function declarations.) There is only one constraint that must be enforced to achieve this feature: pure functions could only call other pure functions, properties, etc.

public pure int MethodName()

Purity inference would enable a lot of beneficial features:

• You can tell at a glance which functions do and do not have side effects.
• Data races are completely ruled out, eliminating the need for a borrow checker, or actors.
• Tests can be guaranteed to never flake, meaning their outcomes can be cached.
• Constants can be fully evaluated at compile time.
• Improved compiler optimizations.
• Expensive computations can be (automatically) memoized.

And it should be possible to incrementally adopt it, similar to nullable reference types. Additionally, it should be completely inferable.

Probably the biggest benefit would be that it encourages a better coding style by motivating developers to actively separate pure and effectful code, thereby reducing bugs and improving maintainability.

It is in line of adapting functional programming paradigms when beneficial.

Inspired by:
https://www.roc-lang.org/plans#purity-inference

https://docs.google.com/document/u/0/d/1ZVD3h5jLpQNFSDXTg2RkzPhNXz5EErUXBBjN8TuyiqQ/mobilebasic?pli=1

[HaloFour]

See: #1157