Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/alhassy/agdacheatsheet

Basics of the dependently-typed functional language Agda ^_^
https://github.com/alhassy/agdacheatsheet

agda cheatsheet emacs html pdf

Last synced: 18 days ago
JSON representation

Basics of the dependently-typed functional language Agda ^_^

Awesome Lists containing this project

README 

        
# Created 2019-10-04 Fri 16:08

#+OPTIONS: toc:nil d:nil

#+OPTIONS: toc:nil d:nil

#+TITLE: Agda CheatSheet

#+AUTHOR: [[http://www.cas.mcmaster.ca/~alhassm/][Musa Al-hassy]]

#+export_file_name: README.org



Basics of Agda, a total and dependently-typed functional language (•̀ᴗ•́)و



*The listing sheet, as PDF, can be found

 [[file:CheatSheet.pdf][here]]*,

 or as a [[file:CheatSheet_Portrait.pdf][single column portrait]],

 while below is an unruly html rendition.



This reference sheet is built from a

[[https://github.com/alhassy/CheatSheet][CheatSheets with Org-mode]]

system. This is a /literate/ Agda file written in Org-mode using

[[https://github.com/alhassy/org-agda-mode][org-agda-mode]].



#+toc: headlines 2



#+macro: blurb Basics of Agda, a total and dependently-typed functional language (•̀ᴗ•́)و



#+latex_header: \usepackage{titling,parskip}

#+latex_header: \usepackage{eufrak} % for mathfrak fonts

#+latex_header: \usepackage{multicol,xparse,newunicodechar}



#+latex_header: \usepackage{etoolbox}



#+latex_header: \newif\iflandscape

#+latex_header: \landscapetrue



#+latex_header_extra: \iflandscape \usepackage[landscape, margin=0.5in]{geometry} \else \usepackage[margin=0.5in]{geometry} \fi



#+latex_header: \def\cheatsheetcols{2}

#+latex_header: \AfterEndPreamble{\begin{multicols}{\cheatsheetcols}}

#+latex_header: \AtEndDocument{ \end{multicols} }



#+latex_header: \let\multicolmulticols\multicols

#+latex_header: \let\endmulticolmulticols\endmulticols

#+latex_header: \RenewDocumentEnvironment{multicols}{mO{}}{\ifnum#1=1 #2 \def\columnbreak{} \else \multicolmulticols{#1}[#2] \fi}{\ifnum#1=1 \else \endmulticolmulticols\fi}



#+latex_header: \def\maketitle{}

#+latex: \fontsize{9}{10}\selectfont



#+latex_header: \def\cheatsheeturl{}



#+latex_header: \usepackage[dvipsnames]{xcolor} % named colours

#+latex: \definecolor{grey}{rgb}{0.5,0.5,0.5}



#+latex_header: \usepackage{color}

#+latex_header: \definecolor{darkgreen}{rgb}{0.0, 0.3, 0.1}

#+latex_header: \definecolor{darkblue}{rgb}{0.0, 0.1, 0.3}

#+latex_header: \hypersetup{colorlinks,linkcolor=darkblue,citecolor=darkblue,urlcolor=darkgreen}



#+latex_header: \setlength{\parindent}{0pt}



#+latex_header: \def\cheatsheetitemsep{-0.5em}

#+latex_header: \let\olditem\item

#+latex_header_extra: \def\item{\vspace{\cheatsheetitemsep}\olditem}



#+latex_header: \usepackage{CheatSheet/UnicodeSymbols}



#+latex_header: \makeatletter

#+latex_header: \AtBeginEnvironment{minted}{\dontdofcolorbox}

#+latex_header: \def\dontdofcolorbox{\renewcommand\fcolorbox[4][]{##4}}

#+latex_header: \makeatother



#+latex_header: \RequirePackage{fancyvrb}

#+latex_header: \DefineVerbatimEnvironment{verbatim}{Verbatim}{fontsize=\scriptsize}



#+latex_header: \def\cheatsheeturl{https://github.com/alhassy/AgdaCheatSheet}



#+latex_header: \def\cheatsheetcols{2}

#+latex_header: \landscapetrue

#+latex_header: \def\cheatsheetitemsep{-0.5em}



#+latex_header: \newunicodechar{‼}{\ensuremath{!\!!}}

#+latex_header: \newunicodechar{𝕨}{\ensuremath{\mathbb{w}}}

#+latex_header: \newunicodechar{≈}{\ensuremath{\approx}}

#+latex_header: \newunicodechar{ℓ}{\ensuremath{\ell}}

#+latex_header: \newunicodechar{ω}{\ensuremath{\omega}}

#+latex_header: \newunicodechar{⁰}{\ensuremath{^0}}

#+latex_header: \newunicodechar{⁴}{\ensuremath{^4}}

#+latex_header: \newunicodechar{♯}{\ensuremath{\sharp}}

#+latex_header: \newunicodechar{α}{\ensuremath{\alpha}}

#+latex_header: \newunicodechar{β}{\ensuremath{\beta}}



#+latex_header: \newunicodechar{⇨}{\ensuremath{\circlearrowright}}



#+begin_quote

- [[#administrivia][Administrivia]]

- [[#dependent-functions][Dependent Functions]]

- [[#reads][Reads]]

- [[#dependent-datatypes][Dependent Datatypes]]

- [[#the-curry-howard-correspondence----propositions-as-types][The Curry-Howard Correspondence ---“Propositions as Types”]]

  - [[#adding-to-the-table][Adding to the table]]

- [[#equality][Equality]]

- [[#break][break]]

- [[#modules----namespace-management][Modules ---Namespace Management]]

  - [[#anonymous-modules-and-variables][Anonymous Modules and Variables]]

  - [[#break][break]]

  - [[#module-keywords][Module Keywords]]

- [[#records][Records]]

- [[#interacting-with-the-real-world----compilation-haskell-and-io][Interacting with the real world ---Compilation, Haskell, and IO]]

- [[#absurd-patterns][Absurd Patterns]]

  - [[#preconditions-as-proof-object-arguments][Preconditions as proof-object arguments]]

- [[#mechanically-moving-from-bool-to-set----avoiding-boolean-blindness][Mechanically Moving from ~Bool~ to ~Set~ ---Avoiding “Boolean Blindness”]]

#+end_quote



* Administrivia



#+latex: \hspace{-1.3em}

Agda is based on  intuitionistic type theory.



| Agda | ≈ | Haskell + Harmonious Support for Dependent Types |



In particular, /types ≈ terms/ and so, for example,

~ℕ ∶ Set = Set₀~ and ~Setᵢ ∶ Setᵢ₊₁~.

One says /universe/ ~Setₙ~ has /level/ $n$.



⇨ It is a programming language and a proof assistant.

#+latex: \newline {\color{white}.}\hspace{0.3em}

A proposition is proved by writing a program of the corresponding type.



⇨ Its Emacs interface allows programming by gradual refinement

  of incomplete type-correct terms. One uses the “hole” marker ~?~

  as a placeholder that is used to stepwise write a program.



⇨ Agda allows arbitrary mixfix Unicode lexemes, identifiers.

- Underscores are used to indicate where positional arguments.

- Almost anything can be a valid name; e.g., ~[]~ and ~_∷_~ below.

  Hence it's important to be liberal with whitespace: ~e:T~ is a valid identifier

  whereas ~e ∶ T~ declares ~e~ to be of type ~T~.



#+latex: \begin{parallel}



#+latex: \begin{tiny}

#+begin_src agda

module CheatSheet where



open import Level using (Level)

open import Data.Nat

open import Data.Bool hiding (_>_~ and ~_>>=_~ need to be in scope ---that is all.

The type of ~IO._>>_~ takes two “lazy” IO actions and yield a non-lazy IO action.

The one below is a homogeneously typed version.



#+begin_src agda

infixr 1 _>>=_ _>>_



_>>=_ : ∀ {ℓ} {α β : Set ℓ} → IO α → (α → IO β) → IO β

this >>= f = ♯ this IO.>>= λ x → ♯ f x



_>>_ : ∀{ℓ} {α β : Set ℓ} → IO α → IO β → IO β

x >> y = x >>= const y

#+end_src



Oddly, Agda's standard library comes with ~readFile~ and

~writeFile~, but the symmetry ends there since it provides ~putStrLn~

but not [[https://hackage.haskell.org/package/base-4.12.0.0/docs/Prelude.html#v:getLine][~getLine~]]. Mimicking the ~IO.Primitive~ module, we define /two/

versions ourselves as proxies for Haskell's ~getLine~ ---the second one

below is bounded by 100 characters, whereas the first is not.



#+begin_src agda

postulate

  getLine∞ : Primitive.IO Costring



{-# FOREIGN GHC

  toColist :: [a] -> MAlonzo.Code.Codata.Musical.Colist.AgdaColist a

  toColist []       = MAlonzo.Code.Codata.Musical.Colist.Nil

  toColist (x : xs) =

    MAlonzo.Code.Codata.Musical.Colist.Cons x (MAlonzo.RTE.Sharp (toColist xs))

#-}



{- Haskell's prelude is implicitly available; this is for demonstration. -}

{-# FOREIGN GHC import Prelude as Haskell #-}

{-# COMPILE GHC getLine∞  = fmap toColist Haskell.getLine #-}



-- (1)

-- getLine : IO Costring

-- getLine = IO.lift getLine∞



getLine : IO String

getLine = IO.lift

  $ getLine∞ Primitive.>>= (Primitive.return ∘ S.fromList ∘ B.toList ∘ take 100)

#+end_src

We obtain ~MAlonzo~ strings, then convert those to colists, then

eventually lift those to the wrapper ~IO~ type.



Let's also give ourselves Haskell's ~read~ method.

#+begin_src agda

postulate readInt  : L.List Char → ℕ

{-# COMPILE GHC readInt = \x -> read x :: Integer  #-}

#+end_src



Now we write our ~main~ method.

#+begin_src agda

main : Primitive.IO ⊤

main = run do putStrLn "Hello, world! I'm a compiled Agda program!"



              putStrLn "What is your name?"

              name ← getLine



              putStrLn "Please enter a number."

              num ← getLine

              let tri = show $ sum $ upTo $ suc $ readInt $ S.toList num

              putStrLn $ "The triangle number of " ++ num ++ " is " ++ tri



              putStrLn "Bye, "

              -- IO.putStrLn∞ name  {- If we use approach (1) above. -}

              putStrLn $ "\t" ++ name

#+end_src

For example, the $12^{th}$ [[https://en.wikipedia.org/wiki/Triangular_number][triangle number]] is $\sum_{i=0}^{12} i = 78$.

Interestingly, when an integer parse fails, the program just crashes!

Super cool dangerous stuff!



Calling this file ~CompilingAgda.agda~, we may compile then run it with:

#+begin_src shell :tangle no

NAME=CompilingAgda; time agda --compile $NAME.agda; ./$NAME

#+end_src



The very first time you compile may take ∼80 seconds since some prerequisites need to be compiled,

but future compilations are within ∼10 seconds.



The generated Haskell source lives under the newly created MAlonzo directory; namely

~./MAlonzo/Code/CompilingAgda.hs~. Here's some fun: Write a parameterised module with multiple declarations,

then use those in your ~main~; inspect the generated Haskell to see that the module is thrown away in-preference

to top-level functions ---as mentioned earlier.



- When compiling you may see an error ~Could not find module ‘Numeric.IEEE’~.

- Simply open a terminal and install the necessary Haskell library:

  #+begin_src shell :tangle no

  cabal install ieee754

  #+end_src



* Absurd Patterns



#+latex: \hspace{-1.3em}

When there are no possible constructor patterns, we may match on the pattern ~()~

and provide no right hand side ---since there is no way anyone could provide an argument

to the function.



For example, here we define the datatype family of numbers smaller than a given natural number:

~fzero~ is smaller than ~suc n~ for any ~n~, and if ~i~ is smaller than ~n~ then ~fsuc i~ is smaller

than ~suc n~.



#+latex: \begin{parallel}

#+begin_src agda

{- Fin n  ≅  numbers i with i < n -}

data Fin : ℕ → Set where

  fzero : {n : ℕ} → Fin (suc n)

  fsuc  : {n : ℕ}

        → Fin n → Fin (suc n)

#+end_src

#+latex: \columnbreak



For each $n$, the type ~Fin n~ contains $n$ elements;

e.g., ~Fin 2~ has elements ~fsuc fzero~ and ~fzero~,

whereas ~Fin 0~ has no elements at all.



#+latex: \end{parallel} \vspace{-1em}



Using this type, we can write a safe indexing function that never “goes out of bounds”.

#+begin_src agda



_‼_ : {A : Set} {n : ℕ} → Vec A n → Fin n → A

[] ‼ ()

(x ∷ xs) ‼ fzero  = x

(x ∷ xs) ‼ fsuc i = xs ‼ i

#+end_src



When we are given the empty list, ~[]~, then ~n~ is necessarily ~0~,

but there is no way to make an element of type ~Fin 0~ and so we have the absurd pattern.

That is, since the empty type ~Fin 0~ has no elements there is nothing to define

---we have a definition by /no cases/.



Logically [[https://en.wikipedia.org/wiki/Principle_of_explosion][“anything follows from false”]] becomes the following program:

#+begin_src agda

data False : Set where



magic : {Anything-you-want : Set} → False → Anything-you-want

magic ()

#+end_src



Starting with ~magic x = ?~ then casing on ~x~ yields the program above

since there is no way to make an element of ~False~

---we needn't bother with a result(ing right side), since there's no way to make

an element of an empty type.



** Preconditions as proof-object arguments



Sometimes it is not easy to capture a desired precondition in the types, and

an alternative is to use the following ~isTrue~-approach of passing around

explicit proof objects.



#+latex: \begin{parallel}

#+begin_src agda

{- An empty record has only

   one value: record {} -}

record True : Set where



isTrue : Bool → Set

isTrue true  = True

isTrue false = False

#+end_src

#+latex: \columnbreak

#+begin_src agda

_<₀_ : ℕ → ℕ → Bool

_ <₀ zero      = false

zero <₀ suc y  = true

suc x <₀ suc y = x <₀ y

#+end_src

#+latex: \end{parallel} \vspace{-1em}



#+begin_src agda

find : {A : Set} (xs : List A) (i : ℕ) → isTrue (i <₀ length xs) → A

find [] i ()

find (x ∷ xs) zero pf    = x

find (x ∷ xs) (suc i) pf = find xs i pf



head′ : {A : Set} (xs : List A) → isTrue (0 <₀ length xs) → A

head′ [] ()

head′ (x ∷ xs) _ = x

#+end_src



Unlike the ~_‼_~ definition, rather than there being no index into the empty list,

there is no proof that a natural number ~i~ is smaller than 0.



* Mechanically Moving from ~Bool~ to ~Set~ ---Avoiding “Boolean Blindness”



#+latex: \hspace{-1.3em}

In Agda we can represent a proposition as a type whose elements denote proofs

of that proposition. Why would you want this? Recall how awkward it was to request

an index be “in bounds” in the ~find~ method, but it's much easier to encode this

using ~Fin~ ---likewise, ~head′~ obtains a more elegant type when the non-empty precondition

is part of the datatype definition, as in ~head~.



Here is a simple recipe to go from Boolean functions to inductive datatype families.

1. Write the Boolean function.

2. Throw away all the cases with right side ~false~.

3. Every case that has right side ~true~ corresponds to a new nullary constructor.

4. Every case that has $n$ recursive calls corresponds to an ~n~-ary constructor.



Following these steps for ~_<₀_~, from the left side of the page, gives us:



#+begin_src agda

data _<₁_ : ℕ → ℕ → Set where

  z< : {y : ℕ} → zero <₁ y

  s< : {x y : ℕ} → x <₁ y → suc x <₁ suc y

#+end_src



To convince yourself you did this correctly, you can prove “soundness”

---constructed values correspond to Boolean-true statements---

and “completeness” ---true things correspond to terms formed from constructors.

The former is ensured by the second step in our recipe!



#+begin_src agda

completeness : {x y : ℕ} → isTrue (x <₀ y) → x <₁ y

completeness {x}     {zero}  ()

completeness {zero}  {suc y} p = z<

completeness {suc x} {suc y} p = s< (completeness p)

#+end_src



We began with ~completeness {x} {y} p = ?~, then we wanted to case on ~p~

but that requires evaluating ~x <₀ y~ which requires we know the shapes of ~x~ and ~y~.

/The shape of proofs usually mimics the shape of definitions they use/; e.g., ~_<₀_~ here.