These are some advanced notes extending the material presented this week. Don't worry about mastering this stuff now if you're already feeling saturated. But you will need to learn these things in the near future, so tackle them as soon as you're ready.
+[[!toc]]
### More on multivalues ###
I agree it's annoying that these conventions are so diverse. There are plenty other commenting conventions out there, too.
-
+<a id=funct-declarations></a>
### Matching function values ###
A function value doesn't have any structure---at least none that's visible to the pattern-matching system. You can only match against simple patterns like `_` or the variable `f`.
This special syntax is only permitted in `let`- and `letrec`-constructions, not in `case`-constructions.
+<a id=guards></a>
### Pattern guards ###
In `case` contructions, it's sometimes useful to check not only whether a certain pattern matches, but also whether a certain boolean expression is true, typically where we want some variables in that expression to be bound by the relevant pattern. Thus, for example, if we wanted to count the number of odd numbers in a sequence, we could do this:
If we get to the `y & ys` line in the pattern list, and the pattern-match succeeds, then we check the guard expression `odd? y`, with `y` bound to whatever part of `xs` matched the corresponding part of the pattern `y & ys`. If that boolean expression is `'true`, then we continue to the right-hand side, after the `then`, just as usual. But if the boolean expression is `'false`, then we treat the whole line as a failed match, and proceed on to the next line in the binding list, if any.
+<a id=as-patterns></a>
### As-patterns ###
Sometimes it's useful to bind variables against overlapping parts of a structure. For instance, suppose I'm writing a pattern that is to be matched against multivalues like `([10, 20], 'true)`. And suppose I want to end up with `ys` bound to `[10, 20]`, `x` bound to `10`, and `xs` bound to `[20]`. Using the techniques introduced so far, I have two options. First, I could bind `ys` against `[10, 20]`, and then initiate a second pattern-match to break that up into `10` and `[20]`. Like this:
case ([10, 20], 'true) of
- [ys, _] then case ys of
+ (ys, _) then case ys of
x & xs then ...;
...
end;
Alternatively, I could directly bind `x` against `10` and `xs` against `[20]`. But then I would have to re-cons them together again to get `ys`. Like this:
case ([10, 20], 'true) of
- [x & xs, _] then let
+ (x & xs, _) then let
ys match x & xs
in ...;
...
Both of these strategies work. But they are a bit inefficient. I said you didn't really need to worry about efficiency in this seminar. But these are also a bit cumbersome to write. There's a special syntax that enables us to bind all three of `ys`, `x`, and `xs` in the desired way, despite the fact that they will be matching against overlapping, rather than discrete, parts of the value `[10, 20]`. The special syntax looks like this:
case ([10, 20], 'true) of
- [(x & xs) as ys, _] then ...
+ ((x & xs) as ys, _) then ...
...
end
in (fst, snd, swap, dup)
+<a id=sections></a>
### Sections ###
OCaml and Haskell have a convenient bit of syntax for the common case where you want a function like this: