In the previous post, I explored the idea that some sentences change meaning if you switch between different names for the same object. Some other sentences don’t change meaning – these ones are referentially transparent, which in non-posh-speak means “you don’t notice if I switch to a different name”.
I’ve now asked some questions on haskell-cafe, and consulted various gurus at my work so I’m a little less confused that I was yesterday. 🙂
How does this concept apply to programming languages?
In haskell, you could say that the “names” were expressions like “2+2” and “sqrt(9)” and the underlying ‘thing’ which was being named are the values which those expressions evaluate to. So the integer “two” has many names – “1+1” and “sqrt(4)” and many others.
That gives us a nice connection back to the Quine definition of “referentially transparent”. Essentially, when we claim that Haskell is RT, its just a shorthand way of saying “you can replace an expression with any other expression that evaluates to the same result, and it won’t affect the overall meaning of the program”.
However, I feel that I’ve just handwaved over a few interesting points.
It’s reasonable to think of “sqrt(4)” as being a name for “2” in haskell because, well, functions always return the same value in haskell. But in Java or C, what kind of name could “rand()” be? The function evaluates to various values and it does this because it uses some kind of additional runtime context/state. So the basic idea that you can always have “names” and underlying “things” is a bit shaky – and we can’t really use Quine’s definition without them. What does it mean to say that C++ is “refentially opaque” if the very idea of names/references is shaky?
The same problem applies to the english-language examples from yesterday. I claimed that “Scotlands capital” was another name for edinburgh-the-city. That’s certainly true today. But it might not be true in 100 years. And perhaps, someone who lives in Glasgow might even disagree that its true today! So names which are co-referent today might not always be co-referent.
There’s a famous-ish example that’s similar too. In the sentence “The King of France is bald”, what kind of thing is being named by ‘The King of France’, given that France is a republic. That’s the kind of thing which keeps philosophers awake at night.
And if you’re into books, you’ll know that the author “Iain M Banks” writes sci-fi but “Iain Banks” writes mainstream fiction. There’s only one human being, but he writes under two “personas”. This makes me think that you have to be pretty careful about the THINGS that you think you are NAMING. In this example, “Iain M Banks” and “Iain Banks” are coreferent if you are discuss the human being, but they are not coreferent if you are discussing his “writing personas”.
There was one more point that I handwaved over. Why do I believe that Haskell really is referentially transparent? I might not trust Simon Peyton Jones when he claims that it is! But to prove some property of a language, you need a formal semantics for the language. And haskell98 doesn’t have that (unlike Standard ML). So there’s no much hope of doing a bottom-up inductive proof over the structure of the language – not just yet anyway. That’s a pity.
I’ll finish off with a couple of links about what RT means and RT and haskell.
Now I’m going to return to finish watching the SICP video which triggered this whole day-long distraction. 🙂