The real lesson is for language designers, who for some reason love using equals for assignment. Lisp got it right a hundred years ago and nobody learned.

There is nothing wrong with using equals for assignment, the language designers just have to disallow assignments in if conditions.

It's a little weird. lvalues and rvalues are different things so a biased symbol feels nice to me, <-, e.g.

One more corner case, instead of designing orthogonal features.

It's not a corner case if you do it right.

For example, in C the assignment operator evaluates to the value of the right-hand side expression (so `int x = (y = 2);` initializes `x` to 2). But in Rust, the assignment operator always evaluates to `()` (a.k.a. "unit", the empty tuple), and trust me, this is how you want the language to work in the presence of pervasive move semantics and single ownership. Furthermore, nothing in Rust is "truthy": if-expressions accept a boolean and nothing else (this is also, IMO, exactly what you want in every language, but I'm sure others will disagree :P ). So `if x = 2 {` is a compiler error in Rust because `()` can't be coerced to a boolean (to say nothing of the fact that this reassignment would probably be a compiler error anyway because most variables in Rust are immutable). This doesn't require any acrobatics, it's just the natural fallout of how the rest of the language works.

Thanks, that's indeed a sensible way to do that.

That's a cure worse than the disease. If I faced a false choice between = as an assignment operator or else accepting syntax that distinguishes statements and expressions, I'd just take the = operator.

There's nothing wrong with using assignment in if conditions. For example, `if let Some(thing) = computation_that_may_fail { ... }`.

The operator there is let. The = is just a punctuator which indicates the initial value expression for the let.

Litmus test: if this was rendered into S-expressions, the let would stay, but the = would disappear.

No, that's just because the example I used was Rust. In Ruby it would be `if user = User.find(..)` and the same goes for Go and plenty of other languages.

If user is being freshly bound, that isn't an assignment but an initialization.

If the construct is not allowed when user is already in scope, it reflects the designer's view that an assignment in a conditional isn't a good thing.

If the construct is allowed when user exists already, with no diagnostic, then it carries the pitfalls that this thread is about.

Hacker News is implemented in a dialect of Lisp designed by a guy who wrote multiple books on Lisp. Seems even lispers don't learn:

> The assignment operator is =. I was dubious about this, but decided to try it and see if I got used to it. It turns out to work well, even in prefix. Stripes stand out, which is why they get used on warning signs and poisonous animals.

http://www.paulgraham.com/arcll1.html

That is quite silly. I briefly contemplated this idea for TXR Lisp, but completely rejected it. The = function is numeric comparison, like in Common Lisp. You don't want to turn that into assignment. People used to Lisp dialects where = is numeric comparison will make the mistake:

  (if (= whatever whatever-else) (you-got-burned))

Arc code written by people used to Lisp dialects where = is a pure comparison function, or Arc code converted from other Lisp dialects, has to be carefully reviewed against this.

If I put this into a Lisp dialect, I would make the code walker issue a warning whenever the value of a (= ...) assignment is used, and provide an alternative assignment operator which doesn't have that warning.

I'd say the issue is keyboard designers, if ':' didn't require shift I bet we'd see it for assignment more often.

Well, languages can either use another assignment operator or have type system that wont check that kind of statement. Both ways fix it.

In fact, it's not easy to have this kind of bug on your language. C just has it because people wanted to write stuff like `int a = b = 0`.

`int a <- b <- 0` or `int a := b := 0` would have worked just as well.

Programmers using equals for assignment is a bit of a misunderstanding of what the mathematical idiom "let x = 5" means. The assignment is signaled by "let", not by "=". In fact, several programming languages use "let" exactly for this purpose too.

There are useful cases for this. It's much less ambiguous in those cases if your language allows the definition of the variable in the same location, and scopes it. For example, Perl:

    use strict;
    use warnings;
    
    sub one { 1 }
    
    if ( my $one = one() ) {
        say $one; # prints 1
    }
    
    say $one; # compilation error, "Global symbol "$one" requires explicit package name"

This case is trivial, but when you want a temporary variable and don't want to clutter your scope, it can be useful.

That's not to say assignment and equivalence might not be better off with different operators, as noted by others here.

Declaration and a (mutating) assignment are different: your example demonstrates that Perl signals the declaration with `my`, and other languages offer similar things like `if let ... = ... { ... }`

> Declaration and a (mutating) assignment are different

Yes, that's what I was trying to get at by saying definition, but not very clearly. It's one of the reasons I prefer languages to require and clearly indicate variable declaration in most cases.

Eh, I think it is good in Java in cases where Null input is possible.

Yes, "foo".equals(bar) is useful in Java if bar can be null, but that's a different thing.

Not really different. In fact is is mentioned on the Wikipedia page.

I never got the Java argument. I would much rather fail early and noisily with an NPE than return false and let my program happily chug along when it wasn't expecting a null. If the value is intended to be nullable, I would use an optional.

Well for 95% of Java's live, there was no such thing as Optional.

So say untrusted user input.. many cases where it could be something or null. Yoda checking it was a nice way to save a null check.

Java8 optional is way slower (write a tree alike struct with optional for left/right/parent) and null checks are one of the free ones. (Hardware optimized)

I think it is better than if(x!=null && x.equals("y")

You have to take bits of concision where you can find them in Java.

Counterargument:

Do do this if the coding standards for the language/framework you're working in require it. Like WordPress.

Wordpress is a legacy procedural codebase and should not be used as an example of good development practices.

Regardless of your religious convictions, Wordpress still exists and that work has to be done by somebody.

I'd rather it be done by people who care than those that don't.

How is that a comment on coding standards or best practices? I'm afraid the arguments against procedural code are not religious. Procedural code provides poor encapsulation and modularity. If procedural code was all that great, we wouldn't have needed to have C++, C#, Java, and Objective-C, and any number of other derivative languages. Regardless of my or your opinions, WordPress is a legacy codebase. Procedural code today is all but completely discarded.

While I don't disagree, curious to hear your reasoning for not doing this.

This trick can only save people who remember to use it. Warnings emitted by tooling save everyone.

Also it's slightly awkward to read, as mentioned by others already.

Warnings can only save people who bother to read them. Yoda conditionals save everyone trying to run or compile the code in question. It goes both ways ;)

Re: awkwardness, it doesn't feel awkward at all to me; given that Erlang and Elixir use assignment for destructuring and pattern-matching, Yoda conditionals feel very natural to me in comparison.

For example, this is valid Elixir code:

    foo = {1, 2}
    
    bar = if {baz, 2} = foo do
            525600
          end
    
    IO.inspect foo  # prints {1,2}
    IO.inspect bar  # prints 525600
    IO.inspect baz  # prints 1

The trick saves the person writing the code and anyone modifying it down the road.

You can't force everyone who edits your code to have linters and warnings turned on. Not everyone follows bests practices.

Does it? You also can't force everyone who edits your code to follow your conventions.

actually you can, depending on the management structure in an organization :)

It's unnatural, confusing to read, and won't save you anyway in cases where you're comparing two variables.

It's OK if your tools absolutely can't diagnose accidental if(a = b), but it should be the last resort.

In some languages it's better to use it, eg. in Java you can go with:

   mString != null && mString.equals("test")

or

   "test".equals(mString)

They do the same, but the second one adds hidden `defensive programming` to prevent NPE in `equals`. Saves time AND code.

Intended usage of if(a=b) should be written as if((a=b)). Don't know what tool you use at work or at home, but most static code analysers will point you that. Try SonarQube at least.

> It's unnatural, confusing to read ...

I feel as if that can rephrased "Don't do it because it is not done." A left-hand side constant is a convention that can have practical benefits in limited cases. Which at least should merit consideration.

"Don't do it because it is not done" is a good reason unless you're working solo and are sure your code will never be seen by others. And even then, it's a good idea to stick to what's done so you don't accumulate bad habits for projects that do have others looking at the code.

how would you write a yoda conditional with two variables? which is the yoda conditional (a == b), or (b == a)?

You can't if they're both mutable, since the whole idea of a yoda conditional is to put an immutable thing on the left side. What I meant was that you can't use them in that case, so it's a technique you can't use with 100% consistency.

Also clang warns "using the result of an assignment as a condition without parentheses".

Exactly, and use 'if ((a = b)) ...' in those rare cases when it's appropriate.

Even then I would prefer an explicit comparison:

    if ((a = b) != NULL) ...

I think Rust has a happy medium, where assignment evaluates to (), not the variable (important as this means you don't have an implicit borrow), but the idiom expressed here is usually replaced by:

    if let Some(thing) = fn_that_returns_option_thing() ...

What about PHP? It's in the coding standard in WordPress and Symfony... :(