← Back to context

Comment by dataflow

12 days ago

> Static typing, for example in C, is essentially just moving the location of the type tag from existing at runtime within the struct, to the "analysis" phase of the compiler.

See also "shift left"

> All the information _has_ to exist somewhere, and the only question is where.

Not entirely sure what you mean about this, but not all type information has to exist somewhere. Sometimes things get encoded into a type that might never be represented otherwise -- like whether an integer is a width or a height, or whether a string is a name or an address.

whether something (a variable) is a width or a height is encoded somewhere in the code because if the variable is a width it makes its usage different than if it is a height.

Then the code is written to make sure you never make the mistake of sending a width to a height, because that would be silly.

The type (width or height) is represented as logic in the code.

theoretically - if you have a type system that allows you to know something is a width or height you can often reduce logic to keep track of these things but my experience is that level of granularity in your typing reduces dynamism too much. I would rather keep track of that in code (as it probably will have to deal with it anyway) than make my type system deal with it.

  • There's some cool mechanisms in type-forward languages that lead to minimal dynamism loss along with safety: For instance, you can do scientific programming that carries units along. So you really can divide a length by a time with no problems: Your results just happens to be a velocity, and will come out in meters per second (or whichever your favorite units are). Adding meters to kilometers? Congrats, you might you have to say which physical representation you want in float-land. But then you are saved from adding seconds to milliliters, because then the types will tell you it's complete nonsense.

    But note that to make all of that work well you really need your language to provide a lot of syntactic sugar features which are missing in most older languages, so all the operations read seamlessly, instead of being full of ceremony. One could do all of this in, say, old Java 1.4 or something, but the amount of type annotation, along with the massive mounts of boilerplate to get all of the operationg working right, and checked at compile time would make it all Not-Worth-It(tm) But with enough language features, and a smart enough compiler, `val acc = mySpeed / someSeconds` will work, straight up, with the safety included.

    • > But then you are saved from adding seconds to milliliters, because then the types will tell you it's complete nonsense.

      I was thinking about this as I was reading your comment, and wondering: is it nonsense? If I have 5 seconds of time as well as 10 milliliters of water... I can consider them as being packaged together... which is addition. And I can subtract 3 milliliters and 6 seconds from them without running out of what I had. Nothing wrong with that, really. Five potatoes and a gallon of water is the same notion, just more familiar. Seems no more nonsensical than dividing length by time, right? Food for thought...

      12 replies →

That just means it exists in your head.

  • Not really. Whether an integer represents a width or a height won't even exist in your head if you're implementing a function like max(a, b).

    • > exist in your head if you're implementing a function like max(a, b)

      surely, you have an idea of what you're using that function for, while you're writing it.

      The compiler is doing a check to see if the function makes some minimal logical sense (ala, a typecheck), but it cannot read your mind. But this implies the types are actually an actualization of the thoughts in your head, in a formal manner.

    • Because inside the abstraction max it doesn't represent a width or height, they are just numbers (or just some "comparable things"). But outside the abstraction, at some level, you'll definitely know what a and b are.

    • The type is encoded in the name of the function.

      Is a pretty bad system because you say "won't even exist in your head" but I dont even know if I should pass it integers, floats a srtring. Or what it will return.

At the extreme end, and with no claims of scalability, you have shell: everything is a string.

I'm skeptical that sophisticated type systems need to exist.

Perhaps I should dust off my copy and start reading SICP