See the Dedukti's and Sukerujo's webpages for manuals or tutorials. http://dedukti.gforge.inria.fr/ http://deducteam.gforge.inria.fr/sukerujo/
You can setup Opam to install the development version of Sukerujo instead of the last stable version by typing the following commands in the current directory:
- opam pin -k git add sukerujo .#sukerujo
Warnings are generated but this is a bug of Dedukti (see issue #93)
To compile (and optionally install) Dedukti
you will need:
OCaml >= 4.02
,Menhir
,OCamlBuild
(build only),OCamlFind
(build only).
opam pin add dedukti https://github.com/Deducteam/Dedukti.git
git clone https://github.com/Deducteam/Dedukti.git
cd Dedukti
make
sudo make install
The command
dkcheck examples/append.dk
should output the following.
SUCCESS File 'examples/append.dk' was successfully checked.
The installation provides the following commands:
dkcheck
is the type-checker forDedukti
,dktop
is an interactive wrapper around the type-checker,dkdep
is a dependency generator forDedukti
files,dkindent
is a program to indentDedukti
files.
Then we can declare constants, giving their name and their type.
Dedukti
distinguishes two kinds of declarations:
- declaration of a static symbol
f
of typeA
is writtenf : A
, - declaration of a definable symbol
f
of typeA
is writtendef f : A
.
Definable symbols can be defined using rewrite rules, static symbols can not be defined.
Nat: Type.
zero: Nat.
succ: Nat -> Nat.
def plus: Nat -> Nat -> Nat.
Let's add rewrite rules to compute additions.
[ n ] plus zero n --> n
[ n ] plus n zero --> n
[ n, m ] plus (succ n) m --> succ (plus n m)
[ n, m ] plus n (succ m) --> succ (plus n m).
When adding rewrite rules, Dedukti
checks that they preserve typing.
For this, it checks that the left-hand and right-hand sides of the rules have the same type in some context giving types to the free variables
(in fact, the criterion used is more general, see below), that the free variables occurring in the right-hand side also occur in the left-hand side
and that the left-hand side is a higher-order pattern (see below).
Remark: there is no constraint on the number of rewrite rules associated with a definable symbol.
However it is necessary that the rewrite system generated by the rewrite rules together with beta-reduction
be confluent and terminating on well-typed terms. Confluence can be checked using the option -cc
(see below),
termination is not checked (yet?).
Remark: Because static symbols cannot appear at head of rewrite rules, they are injective with respect to conversion and this information can be exploited by
Dedukti
for type-checking rewrite rules (see below).
A development in Dedukti
is usually composed of several files corresponding to different modules.
Using dkcheck
with the option -e
will produce a file my_module.dko
that exports the constants
and rewrite rules declared in the module my_module
.
Then you can use these symbols in other files/modules using the prefix notation my_module.identifier
.
In Dedukti
comments are delimited by (;
and ;)
.
(; This is a comment ;)
Supported commands are:
#EVAL t. (; evaluate t to its strong normal form and display it. ;)
#EVAL[N]. (; same as above, but evaluate in at most N steps. ;)
#EVAL[STRAT]. (; evaluate t with the strategy STRAT. :)
#EVAL[N,STRAT]. (; same as above, but evaluate in at most N steps. :)
#CHECK t1 == t2. (; display "YES" if t1 and t2 are convertible, "NO" otherwise. ;)
#CHECK t1 : t2. (; display "YES" if t1 has type t2, "NO" otherwise. ;)
#CHECKNOT t1 == t2. (; display "YES" if t1 and t2 are not convertible, "NO" otherwise. ;)
#CHECKNOT t1 : t2. (; display "YES" if t1 does not have type t2, "NO" otherwise. ;)
#ASSERT t1 : t2. (; fail if t1 does not have type t2. ;)
#ASSERT t1 == t2. (; fail if t1 is not convertible with t2. ;)
#ASSERTNOT t1 : t2. (; fail if t1 does have type t2. ;)
#ASSERTNOT t1 == t2. (; fail if t1 is convertible with t2. ;)
#INFER t1. (; infer the type of t1 and display it. ;)
#PRINT s. (; print the string s. ;)
The supported evaluation strategies are:
SNF
(strong normal form),HNF
(head normal form),WHNF
(weak head normal form).
Note that the #INFER
command accepts the same form of configuration as
the #EVAL
command. When given, it is used to evaluate the obtained type.
Dedukti
supports definitions:
def three : Nat := succ ( succ ( succ ( zero ) ) ).
or, omitting the type,
def three := succ ( succ ( succ ( zero ) ) ).
A definition is syntactic sugar for a declaration followed by a rewrite rule. The definition above is equivalent to:
def three : Nat.
[ ] three --> succ ( succ ( succ ( zero ) ) ).
Using the keyword thm
instead of def
makes a definition opaque, meaning that the defined symbol do not reduce
to the body of the definition. This means that the rewrite rule is not added to the system.
thm three := succ ( succ ( succ ( zero ) ) ).
This can be useful when the body of a definition does not matter (only its existence matters), to avoid adding a useless rewrite rule.
When a variable is not used on the right-hand side of a rewrite rule, it can be replaced by an underscore on the left-hand side.
def mult : Nat -> Nat -> Nat.
[ n ] mult zero n --> zero
[ n, m ] mult (succ n) m --> plus m (mult n m).
The first rule can also be written:
[ ] mult zero _ --> zero.
A typical example of the use of dependent types is the type of Vector defined as lists parametrized by their size:
Elt: Type.
Vector: Nat -> Type.
nil: Vector zero.
cons: n:Nat -> Elt -> Vector n -> Vector (succ n).
and a typical operation on vectors is concatenation:
def append: n:Nat -> Vector n -> m:Nat -> Vector m -> Vector (plus n m).
[ n, v ] append zero nil n v --> v
[ n, v1, m, e, v2 ] append (succ n) (cons n e v1) m v2 --> cons (plus n m) e (append n v1 m v2).
These rules verify the typing constraint given above: both left-hand and right-hand sides have the same type.
Also, the second rule is non-left-linear; this is usually an issue because non-left-linear rewrite rules usually generate a non-confluent rewrite system when combined with beta-reduction.
However, because we only intend to rewrite well-typed terms, the rule above is computationally equivalent to the following left-linear rule:
[ n, v1, m, e, v2, x ] append x (cons n e v1) m v2 --> cons (plus n m) e (append n v1 m v2).
Dedukti
will also accept this rule, even if the left-hand side is not well-typed, because it is able to detect that, because of typing
constraints, x
can only be instantiated by a term of the form succ n
(this comes from the fact that Vector
is a static symbol and is
hence injective with respect to conversion: from the type-checking constraint Vector x = Vector (succ n)
, Dedukti
deduces x = succ n
).
For the same reason, it is not necessary to check that the first argument of append
is zero
for the first rule:
[ n, v, x ] append x nil n v --> v.
Using underscores, we can write:
[ v ] append _ nil _ v --> v
[ n, v1, m, e, v2 ] append _ (cons n e v1) m v2 --> cons (plus n m) e (append n v1 m v2).
A different solution to the same problem is to mark with brackets the parts of the left-hand side of the rewrite rules that are constrained by typing.
[ n, v1, m, e, v2 ] append (succ n) (cons {n} e v1) m v2 --> cons (plus n m) e (append n v1 m v2).
The information between brackets will be used when typing the rule but they will not be match against when using the rule (as if they were replaced by fresh variables).
Remark: in order to make this feature type-safe, Dedukti
checks that the typing constraint is verified when using the rule and fails otherwise.
Remark: a variable can occur inside brackets only if it also occurs outside brackets and on the left of the brackets.
By default, Dedukti
rejects non-left-linear rewrite rules because they usually generated non confluent rewrite systems
when combined with beta-reduction. This behaviour can be changed by invoking dkcheck
with the option -nl
.
eq: Nat -> Nat -> Bool.
[ n ] eq n n --> true.
In the previous examples, left-hand sides of rewrite rules were first-order terms.
In fact, Dedukti
supports a larger class of left-hand sides: higher-order patterns.
A higher-order pattern is a beta-normal term whose free variables are applied to (possibly empty) vectors of distinct bound variables.
A classical example of the use of higher-order rules is the encoding the simply types lambda-calculus with beta-reduction:
type: Type.
arrow: type -> type -> type.
term: type -> Type.
def app: a:type -> b:type -> term (arrow a b) -> term a -> term b.
lambda: a:type -> b:type -> (term a -> term b) -> term (arrow a b).
[ f, arg ] app _ _ (lambda _ _ (x => f x)) arg --> f arg.
Remark: type annotations on abstraction must be omitted.
Remark: free variables must be applied to the same number of arguments on the left-hand side and on the right-hand side of the rule.
Remark: with such rewrite rules, matching is done modulo beta in order to preserve confluence.
This means that, in the context (o: type)(c:term o)
, the term App o o (Lam o o (x => x)) c
reduces to c
.
Dedukti
can check the confluence of the rewrite system generated by the rewrite rules and beta-reduction,
using an external confluence checker. For this you need to install a confluence checker for higher-order rewrite systems
supporting the TPDB format, for instance CSI^HO or ACPH.
To enable confluence checking you need to call dkcheck
with the option -cc
followed by the path to the confluence checker:
$ dkcheck -cc /path/to/csiho.sh examples/append.dk
> File examples/append.dk was successfully checked.
Dedukti
is distributed under the CeCILL-B License.