Set Builder Notation 672 words published on March 16, 2020.
I’m not a fan of set builder notation, and let me explain why…
I mean, it beats a lot of other things mathematicians were probably doing before they agreed to it, but that could be said about pretty much anything. Just look at how many people love writing Python, and how god awful that language is (at least syntactically). I’d expect a bit more from math notation, but I digress.
We can read a set as “ such that is in ”. Here is an expression which has bound within it.
To fully explain my position, let’s start by looking at a good use of set builder notation.
Here we lead with the interesting stuff, being the , and brush the fact that under the rug. This is more or less OK, because you don’t need to be thinking about what kind of number is to understand the equation in this case. It’s purely a formality, which may be necessary to a larger context.
However, things start getting interesting quickly.
Suddenly the domain of the set and aren’t necessarily the same, yet we would probably assume something reasonable, right? These must all be , or something. Still we’ve tried to lead with what we care about, the in this case. But suddenly the right side matters a lot more.
Next, we start calling functions on the right side.
I like to think someone would stop at this point and restructure the equations. Why am I operating on in before I even know what is?! A quick glance, and you don’t even notice it anymore, it’s lost in the noise on the right.
Finally, we come to what Python starts letting you do with their surprisingly beloved array comprehensions, which are basically just set builder notation in ASCII. In math, at least we can argue about the evaluation order a little bit, but in Python this is just downright wrong!
[bar(x) for x in foo(x)]
foo will execute before the function
bar, and this is not
entirely clear by reading the source. In fact, this makes me wonder what other
things in Python evaluate backwards…
One more thing, what about chained array comprehensions?
[int(b) for x in range(0, 10) for b in bin(x)[2:]]
My head hurts looking at this code. Where is
x bound? Where is
It’s all so unclear, it really makes my wonder how Python is so popular… But
Meanwhile, in Rust things are done properly. We also have something like a
“such that” which is the
where clause, however you still must at least
introduce the name up front. For example:
fn foo<A>(..., a: A, ...) where A: ...
This solves the problem of name resolution while I try and read this function.
Since there’s always a namespace you’re operating within, I must have a way to
A is local, or coming in from elsewhere. Without the leading
I’d be left wondering what
a: A means for far too long.