But in my domain I have seen countless one-line functions used in one place each, and it’s frustrating having to jump around the code (and add to my mental context stack*) rather than just putting some curly braces and a comment. Functions aren’t primarily meant as a hygiene mechanism. I really like the curly brace scoping in Rust.
I found my style partly by reading John Carmack and Casey Muratori’s coding philosophies and putting them to the test for myself.
* For a natural language analogy, consider the difference between parenthetical remarks and asterisks. You have to remember your place in the text.
I much prefer breaking problems down to the level where I can reason in terms of sets, algebras, and that sort of thing. It's much easier for my brain to rely on properties and the ability to substitute terms in equations than it is to reconstruct a giant state machine and execute it in my head. Much more so when we add concurrency to the mix.
At the end of the day though I don't think there's, "one true way."
IMHO that is a side effect of how we teach people how to program mixed in with orgs cargo culting acronyms and not the core concepts of ideas.
Want people in a large organization to ignore your ideas? Speak them out loud.
Want them to follow them as dogma? Write a blog post under a pseudonym.
I always try to write code like
fn foo() {
let foo = do_first_thing();
let bar = do_second_thing();
do_third_thing(foo, bar);
}whenever possible. Then you can see the abstract overview of what the code does at a glance, but you can drill down into the details if you need to.
An ability that is only valuable if it really is a procedure, some sequence of operations, rather than a function, some forming of a value. I'd agree that procedures should generally not be split up into subprocedures; a procedure has to be read from top to bottom, so factoring out subprocedures only makes it harder to understand. But a function can be understood without understanding the smaller functions that it calls, so breaking functions up into smaller functions - and breaking small functions out of procedures, where possible - makes it easier to understand.
Having a language that makes an explicit distinction between procedures and functions - in particular, where calling a procedure looks different from calling a function - helps a lot (e.g. Haskell).
Sometimes you need explicit scope for a bunch of things and the code would benefit from it. That doesn't justify having explicit scopes for everything all the time.
Nothing in the middle of this code prevents using ‘a’ even though it isn’t relevant to ‘c’. And there’s no way to arrange it to change that.
How does that prevent 'a' from being in the scope where 'c' is defined? It is not clear to me what you mean by defining 'c' later. Later where? And how do we ensure that 'a' is not accessible there?
I think lex-lightning's comment has a point. We need to define 'a' and we need to define 'c' and then we need to print 'c' before printing 'a', so although 'c' does not depend on 'a', we still need to keep 'a' in scope because we cannot print it before we print 'c'.
Perhaps a convoluted solution to the problem posed in lex-lightning's comment would be something like this:
#include <stdio.h>
int getint() {
return 42;
}
int main()
{
int a = getint();
int b = getint() + a;
int c;
{
// Shadow 'a' here to make the outer 'a' inaccessible in this scope.
int *a = NULL;
(void) a;
c = getint();
}
int d = b / c;
printf("%d\n", d);
printf("%d\n", c);
printf("%d\n", b);
printf("%d\n", a);
return 0;
}
Output: $ cc -std=c99 -Wall -Wextra -pedantic foo.c && ./a.out
2
42
84
42One note: each variable only gets assigned to once, so FP won’t magic that away.
The side effects might have a different way of management in FP. But the IO will happen in the same order in reality anyway.
So I don’t think there’s a way out, but I honestly would be interested to see it.
I’d argue that even declaring ‘c’ is against the spirit though. To say nothing of the eldritch shadow cast there (no disrespect, I think we’re both enjoying playing around here)
Yes, I agree. My code example is meant to show the best we can possibly do to solve the problem in your comment and how contrived that solution is. The fact that this solution is contrived and diverges from the intended spirit of the problem only emphasizes the point of your comment.
That's why I am eager to understand what zelphirkalt really meant when they wrote, "the re-arrangement seems simple: define `c` later". It seems far from simple and, in fact, impossible if we want to avoid contrived solutions.
def doD(b: Int, c: Int) = puts(b / c)
def doCD(b: Int) = getInt >>= {c => doD(b, c) >> puts(c)}
def doBCD(a: Int) = getInt >>= {tb => val b = tb+a; doCD(b) >> puts(b)}
getInt >>= {a => doBCD(a) >> puts(a)}But I never got into Haskell so I don’t really know.
I don't think so? If you write it as a "flat" block using do notation then anything from earlier in the block is accessible later in the block. But local functions are real functions with lexical scope, they don't get hoisted up to the containing scope or anything. (And FWIW that's Scala-style syntax, although the ideas are very Haskell-like)
Have you written much Prolog?
I was thinking about in Prolog terms and I think it’s the hardest paradigm to avoid the scoping problem in
But still, the original post said that they have to deal with this:
> because I always see about 10 uninitialized variables at the start of each function, where half of those are used only once
That's bad!
Sure, if you write C like it's 1995, it's not going to be very readable. You don't need to define variable at the beginning of a scope, and you can use {} to scope variables to tighter blocks. And functions can be small.
(Sometimes you need to keep variables for longer than you would like for cleanup reason, that's true.)