Comment by zemo | Hacker News Reader

Comment by zemo | Hacker News Reader

Parent Full thread

unless I'm missing something, that seems to address a separate difference in the specifics of Go and Haskell's type systems, namely that Go's interfaces and Haskell's sum types are different. Yes, they're different. I really only meant to say that the article is predicated on the notion that we're only able to branch on booleans instead of predicates. As for checking for null values...

    var x *MyType

    ...

    switch x {
    case nil:
        // blah
    default:
        // blah
    }

tada, no boolean required, no having to "establish the provenance" of any bits. Sure, inside the nil case, the compiler doesn't actually stop you from unsafely attempting to access the fields of x, and, in that way, Haskell is safer than Go.

I'm not trying to assert that Go is better than Haskell or that they are the same; I've never programmed Haskell so I'm not qualified to make such a statement. I'm merely addressing the specific "Boolean Blindness" criticism.

I don't really think that criticism is in any way valid, since "Boolean Blindness" isn't a criticism of a language design, it's a criticism of how one might use a particular language. To say that Go committed "Boolean Blindness" is predicated on the following assumptions:

- you can only branch on a boolean

- given a value and a type, the only thing you can do is obtain a boolean indicating whether or not that value is of that type

as far as Go is concerned, those statements are factually inaccurate.

Given some variable `x` of unknown type, and a type `t`, is there some way, in Haskell, to create a boolean `v` such that the value of `v` is `true` when `x` is of type `t` and `false` otherwise? And if so, is there some way to branch on this boolean value `v`? Because if that is true, then how has Haskell not committed the same atrocity of "Boolean Blindness"?

> tada, no boolean required, no having to "establish the provenance" of any bits. Sure, inside the nil case, the compiler doesn't actually stop you from unsafely attempting to access the fields of x,

But you do have to establish the provenance of your conditional. You have to keep track of whether you compared against nil yourself.

Whether this is expressed as if(x != nil) or as a boolean-blind switch case, the information is lost and it is up to the user to keep track of the meaning of 'x' under all the conditions it is run.

Of course Haskell is capable of expressing boolean blind and unsafe code. But Haskell, unlike go is also capable of expressing non-blind, safe code.

Go is only able to express boolean blind code (again, it is not about a boolean condition but about whether branching buys you type information). When you compare against nil, it is up to you to keep track of the provenance.

it is up to the user to keep track of the meaning of 'x' under all the conditions it is run

OK, so it's a bit like you're chaining assumptions together, rather than putting them next to one another? This makes Monads, esp. `>>=`, make a bunch more sense to me.

Like, the difference is between chaining/linking two operations (incl. checking what `x` is), by using the result of one check as the input to the other, and basically putting two operations next to one another (not to say that the `if` is meaningless, just that once the `if` is evaluated, the associated block has to take the `if`'s word for it).

The bonus is that Haskell handles the first part, the check, for you if you let it. It's an implicit (well sorta) and immutable precondition to whatever's next, at a logical/linguistic level.

And I guess the reason why booleans in particular are problematic is because you're sorta bolting them on, and they inherently compress information into yes/no, which is otherwise not terribly useful for the block following?

> You have to keep track of whether you compared against nil yourself.

that's true in the case of comparing against nil, which I've already agreed with. I never said that Go is as safe as or safer than Haskell, I said that your assertion that there's "no way to branch and get type information" is a factual inaccuracy that misrepresents the language.

The fact that you haven't commented on type switches in any way shows that you're not really interested in anything other than your own sense of superiority.

Haskell is a better programming language. As a programmer of a programming language that is not Haskell, I am inferior to you.

There.

It's not just comparing against nil, it's any comparison or switch-case besides a type-switch.

A type-switch seems to be non-boolean-blind indeed, but it seems to have too much syntactic overhead, as Go code generally uses if branches when a type-safe case is due (e.g: When analyzing return codes).

I _think_ the complaint is the following:

if x == nil { x.doSomethingCrazy(); // Runtime error? } else { ... }

"Boolean blindness" (bah) is that either the type system should catch this at compile time (because we know x is nil), or the language should make it impossible to phrase (pattern matching to pull the information out, such as with the Maybe type).

Go has many type system oddities (reflection instead of generics, for instance), but the things they _have_ done (auto-implementation of interfaces, for example) are pretty cool.

I find it unforgivable for a language in 2012 to repeat decades old design mistakes, and for no apparent benefit.

What's worse is that it appears this was all done out of ignorance, not conscious choice.

You think Rob Pike and Ken Thompson made what you consider a mistake out of ignorance? Just because you don't like it doesn't make it a mistake.

What benefit does it have? What purpose does it serve?

As far as I know, these guys have a reputation in industry moreso than in academic circles. It seems entirely believable that they have taken no interest in language design outside of industry, where the state of the art lags a few decades behind that of academia.

The benefit is not having to code it. The most bug-free components are the ones that aren't there. Also, it's only v1. I don't think anybody's said that they're opposed to putting it in, so give it time. Maybe file a feature request.

I think if they add sum types and pattern matching or otherwise make this feature possible, it will completely change the way people write Go code and it would become a very different language.

Consider all the error-checks in Go -- these are all incorrect from a boolean-blindness perspective, and that would be a lot of code to rewrite if they add proper constructs to Go. Thus, I really believe it is just a mistake -- one that is only possible to make out of ignorance.

Because Go is concurrent and has mutable types, race conditions could cause the assumptions gleaned from the booleans to be false in the very next statement. Boolean blindness is a result of the language.

No, that's false -- consider the Maybe case. If you carry around the content of the Maybe, it will still remain valid even after the Maybe itself is overwritten. Races are of course still possible but they don't relate to boolean blindness.

So you want to copy the data every time there's an if? Some other part of the program could have a pointer to it otherwise.

Reply on news.ycombinator.com