{#measures}
============
Measuring Data Types
--------------------
\begin{code}
module Measures where
import Prelude hiding ((!!), length)
import Language.Haskell.Liquid.Prelude
length :: L a -> Int
(!) :: L a -> Int -> a
insert :: Ord a => a -> L a -> L a
insertSort :: Ord a => [a] -> L a
\end{code}
Measuring Data Types
====================
Recap
-----
1.
**Refinements:** Types + Predicates
2.
**Subtyping:** SMT Implication
Example: Lists
--------------
\begin{code}
infixr `C`
\end{code}
\begin{code}
data L a = N
| C a (L a)
\end{code}
Example: Length of a List
-------------------------
\begin{code}
{-@ measure llen :: (L a) -> Int
llen (N) = 0
llen (C x xs) = 1 + (llen xs) @-}
\end{code}
LiquidHaskell *strengthens* data constructor types
\begin{code}
data L a where
N :: {v: L a | (llen v) = 0}
C :: a -> xs:L a
-> {v:L a | (llen v) = 1 + (llen xs)}
\end{code}
Measures Are Uninterpreted
--------------------------
\begin{code}
data L a where
N :: {v: L a | (llen v) = 0}
C :: a -> xs:L a
-> {v:L a | (llen v) = 1 + (llen xs)}
\end{code}
`llen` is an *uninterpreted function* in SMT logic
Measures Are Uninterpreted
--------------------------
In SMT, [uninterpreted function](http://fm.csl.sri.com/SSFT12/smt-euf-arithmetic.pdf) `f` obeys *congruence* axiom:
`forall x y. (x = y) => (f x) = (f y)`
All other facts about `llen` asserted at *fold* and *unfold*
Measures Are Uninterpreted
--------------------------
All other facts about `llen` asserted at *fold* and *unfold*
\begin{code}**Fold**
z = C x y -- z :: {v | llen v = 1 + llen y}
\end{code}
\begin{code}**Unfold**
case z of
N -> e1 -- z :: {v | llen v = 0}
C x y -> e2 -- z :: {v | llen v = 1 + llen y}
\end{code}
Measured Refinements
--------------------
Now, we can verify:
\begin{code}
{-@ length :: xs:L a -> (EqLen xs) @-}
length N = 0
length (C _ xs) = 1 + (length xs)
\end{code}
Where `EqLen` is a type alias:
\begin{code}
{-@ type EqLen Xs = {v:Nat | v = (llen Xs)} @-}
\end{code}
List Indexing Redux
-------------------
We can type list indexed lookup:
\begin{code}
{-@ (!) :: xs:L a -> (LtLen xs) -> a @-}
(C x _) ! 0 = x
(C _ xs) ! i = xs ! (i - 1)
_ ! _ = liquidError "never happens!"
\end{code}
Where `LtLen` is a type alias:
\begin{code}
{-@ type LtLen Xs = {v:Nat | v < (llen Xs)} @-}
\end{code}
List Indexing Redux
-------------------
Now we can type list indexed lookup:
\begin{code}
{-@ (!) :: xs:L a -> (LtLen xs) -> a @-}
(C x _) ! 0 = x
(C _ xs) ! i = xs ! (i - 1)
_ ! _ = liquidError "never happens!"
\end{code}
Demo:
What if we *remove* the precondition?
Multiple Measures
-----------------
LiquidHaskell allows *many* measures for a type
Multiple Measures
-----------------
**Example:** Nullity of a `List`
\begin{code}
{-@ measure isNull :: (L a) -> Prop
isNull (N) = true
isNull (C x xs) = false @-}
\end{code}
\begin{code} LiquidHaskell **strengthens** data constructors
data L a where
N :: {v : L a | (isNull v)}
C :: a -> L a -> {v:(L a) | not (isNull v)}
\end{code}
Multiple Measures
-----------------
LiquidHaskell *conjoins* data constructor types:
\begin{code}
data L a where
N :: {v:L a | (llen v) = 0
&& (isNull v) }
C :: a
-> xs:L a
-> {v:L a | (llen v) = 1 + (llen xs)
&& not (isNull v) }
\end{code}
Multiple Measures
-----------------
Unlike [indexed types](http://dl.acm.org/citation.cfm?id=270793) ...
+
Measures *decouple* properties from structures
+
Support *multiple* properties over structures
+
Enable *reuse* of structures
Invaluable in practice!
Refined Data Constructors
-------------------------
Can *directly pack* properties inside data constructors
\begin{code}
{-@ data L a = N
| C (x :: a)
(xs :: L {v:a | x <= v}) @-}
\end{code}
Specifies *increasing* Lists
Refined Data Constructors
-------------------------
**Example:** Increasing Lists, with strengthened constructors:
\begin{code}
data L a where
N :: L a
C :: x:a -> xs: L {v:a | x <= v} -> L a
\end{code}
-
LiquidHaskell *checks* property when *folding* `C`
-
LiquidHaskell *assumes* property when *unfolding* `C`
Refined Data Constructors
-------------------------
Demo: Insertion Sort (hover for inferred types)
\begin{code}
insertSort = foldr insert N
insert y (x `C` xs)
| y <= x = y `C` (x `C` xs)
| otherwise = x `C` insert y xs
insert y N = y `C` N
\end{code}
**Problem 1:** What if we need *both* [increasing and decreasing lists](http://web.cecs.pdx.edu/~sheard/Code/QSort.html)?