Alexis King

I think copilot is solving the wrong problem. A future of programming where we're higher up the abstraction tree is absolutely something I want to see. I am taking advantage of that right now -- I'm a decently good programmer, in the sense that I can write useful, robust, reliable software, but I'm pretty high up the stack, working in languages like Java or even higher up the stack that free me from worrying about the fine details of memory allocation or the particular architecture of the hardware my code is running on.

Copilot is NOT a shift up the abstraction tree. Over the last few years, though, I've realized the the concept of typing is. Typed programming is becoming more popular and prominent beyond just traditional "typed" languages -- see TypeScript in JS land, Sorbet in Ruby, type hinting in Python, etc. This is where I can see the future of programming being realized. An expressive type system lets you encode valid data and even valid logic so that the "building blocks" of your program are now bigger and more abstract and reliable. Declarative "parse don't validate"[1] is where we're eventually headed, IMO.

An AI that can help us to both _create_ new, useful types, and then help us _choose_ the best type, would be super helpful. I believe that's beyond the current abilities of AI, but can imagine that in the future. And that would be amazing, as it would then truly be moving us up the abstraction tree in the same way that, for instance, garbage collection has…

Note that the division-by-zero example used in this article is not the best example to demonstrate "Parse, Don't Validate," because it relies on encapsulation. The principle of "Parse, Don't Validate" is best embodied by functions that transform untrusted data into some data type which is correct by construction.

Alexis King, the author of the original "Parse, Don't Validate" article, also published a follow-up, "Names are not type safety" [0] clarifying that the "newtype" pattern (such as hiding a nonzero integer in a wrapper type) provide weaker guarantees than correctness by construction. Her original "Parse, Don't Validate" article also includes the following caveat:

> Use abstract datatypes to make validators “look like” parsers. Sometimes, making an illegal state truly unrepresentable is just plain impractical given the tools Haskell provides, such as ensuring an integer is in a particular range. In that case, use an abstract newtype with a smart constructor to “fake” a parser from a validator.

So, an abstract data type that protects its inner data is really a "validator" that tries to resemble a "parser" in cases where the type system itself cannot encode the invariant.

The article's second example, the non-empty vec, is a better example, because it encodes within the type system the invariant that one element must exist. The crux of Alexis King's article is that programs should be structured so that functions return data types designed to be correct by…

You make your life substantially easier if all the functions/methods/procedures/blocks/... in your code are "total" -- if they transform every possible input into the output one would expect from the name and type signature. With that as a guiding principle, I tend toward the following strategies, the particular choice of which depends on other details. Most often I'll combine 2/5, throwing a compile-time error for the broken branches, later realizing that the abstraction I chose was stupid, and then writing code that actually does what I want.

1. Inline the work being done. If you've only implemented a single case it doesn't deserve a named function that does something different from what the name indicates. Refactor it later if you ever need when you have more cases.

2. Ensure the error happens at compilation. C# has diagnostic directives. Zig has the @compileError builtin. C/C++ can fail using the preprocessor. It's a relatively common feature, and if you ensure that when your buggy/unfinished function compiles it's correct then that mitigates the harm you might cause (and serves as documentation: @compileError("TODO: ...")).

3. Ensure the type signature captures the problem. If some inputs result in crashes, return an error-code/error-condition/failure-union/optional/nullable/... so that the caller knows shit might happen and can deal with it at runtime if it does.

4. Make it abundantly clear that there's an issue. Name it…

> no commonly used statically typed language helps here because they all allow null references

This is false.

TypeScript, Swift, and Rust are commonly used and support non-nullable references.

> those languages, don't help you because you are dealing with real world data where inputs to your system can be null or not so you end up using some type system escape hatch anyway.

You don't need escape hatches to deal with "real world data" that may be missing some values. This blog post is my favorite detailed comparison of handling "real world data" in static vs dynamic languages: https://lexi-lambda.github.io/blog/2020/01/19/no-dynamic-type-systems-are-not-inherently-more-open/

Rust's serde_json docs on "Operating on untyped JSON values" are also a pretty good description of working with "real world data" where you want to examine an arbitrary document: https://docs.serde.rs/serde_json/#operating-on-untyped-json-values

The only requirement for working with "real world data" in a static non-nullable language is to choose whatever kind of behaviour you want when working with the data. Everything you can do with null references, you can do better with option types; there is nothing that null references uniquely permit.

Very nice article, I liked it a lot! It personally resonated with me and my own conclusion that the core "benefit" of FP is (for lack of a better work, stupid Bitcoin) "proof of work".

Writing functions FP is essentially all about returning results from a function, which is proof that that a computation has occurred. If you don't have that return value, that proof, then obviously the rest of your code can't and shouldn't continue, and FP makes it obvious compared to more traditional imperative approaches.

And this idea extends into so many other things that people consider core, or at least originating from FP. Result/Option types come to mind, making the possible failure of "proof of work" explicit in the type signature, so people are forced to consciously handle it. It also leads into the whole idea of type driven design, one of my favorite articles, "Parse, don’t validate"[1], describes this as well, making types that clearly restrict and set the expectations needed for the "proof of work" to always be returned from a function.

[1] https://lexi-lambda.github.io/blog/2019/11/05/parse-don-t-validate/

Zod allows for more complex validation like "input cannot exceed 256 chars", "number must be positive", or even arbitrary "refinements". I could, of course, simply write a function that will `Abort` instead, but tRPC+Zod handles that automatically - zero boilerplate. My current understanding is that I'll need to invoke Abort manually if the Zod.parse fails.

Importantly Zod philosophically follows "parse, don't validate" https://lexi-lambda.github.io/blog/2019/11/05/parse-don-t-validate/ which means that once validation is done, you get a type - a parsed value. This means you can validate clientside and serverside using the same types. Again, sure you could do the same with Telefunc, but you still need to invoke Abort manually and every single RPC definition you have needs to be followed by Zod.parse. A small cost to be sure, but what are the advantages of Telefunc over tRPC? I'm not seeing any right now.

tRPC also isn't strongly tied to Zod, it also works with yup, superstruct, and custom validators.