In LF, deductive systems are represented by signatures consisting of constant declarations. Twelf implements declarations in a straightforward way and generalizes signatures by also allowing definitions, which are semantically transparent. Twelf currently does not have module-level constructs so that, for example, signatures cannot be named. Instead, multiple signatures can be manipulated in the programming environment using configurations (see section 10.1 Configurations).
The LF type theory which underlies LF is stratified into three levels: objects M and N, types A and B, and kinds K. Twelf does not syntactically distinguish these levels and simply uses one syntactic category of term. Similarly, object-level constants c and type-level constants a as well as variables share one name space of identifiers.
In explanations and examples we will use letters following the mathematical conventions above to clarify the roles of various terms. We also use U and V to stand for arbitrary terms.
The grammar below defines the non-terminals sig, decl,
term and uses the terminal id which stands for identifers
(see section 2.2 Identifiers). The comments show the meaning in LF. There are
various special declarations %keyword such as %infix
or %theorem which are omitted here and detailed in the
appropriate sections.
sig ::= % Empty signature
| decl sig % Constant declaration
decl ::= id : term. % a : K or c : A
| id : term = term. % d : A = M
| id = term. % d = M
| _ : term = term. % anonymous definition, for type-checking
| _ = term. % anonymous definition, for type-checking
| %abbrev adecl. % abbreviation
| %infix ixdecl. % operator declaration
| %prefix pxdecl. % operator declaration
| %postfix pxdecl. % operator declaration
| %name namepref. % name preference declaration
| %query qdecl. % query declaration
| %solve id : term. % solve declaration
| %mode mdecl. % mode declaration
| %terminates tdecl. % termination declaration
| %reduces rdecl. % reduction declaration
| %theorem thdecl. % theorem declaration
| %prove pdecl. % prove declaration
| %establish pdecl. % prove declaration, do not use as lemma later
| %assert callpats. % assert theorem (requires Twelf.unsafe)
| %use domain. % installs constraint domain
term ::= type % type
| id % variable x or constant a or c
| term -> term % A -> B
| term <- term % A <- B, same as B -> A
| {id : term} term % Pi x:A. K or Pi x:A. B
| [id : term] term % lambda x:A. B or lambda x:A. M
| term term % A M or M N
| term : term % explicit type ascription
| _ % hole, to be filled by term reconstruction
| {id} term % same as {id:_} term
| [id] term % same as [id:_] term
The constructs {x:U} V and [x:U] V bind the identifier
x in V, which may shadow other constants or bound
variables. As usual in type theory, U -> V is treated as an
abbreviation for {x:U} V where x does not appear
in V. However, there is a subtlety in that the latter allows
an implicit argument (see section 4.2 Implicit Arguments) to depend on
x while the former does not.
In the order of precedence, we disambiguate the syntax as follows:
For example, the following are parsed identically:
d : a <- b <- {x} c x -> p x.
d : ({x} c x -> p x) -> b -> a.
d : ((a <- b) <- ({x:_} ((c x) -> (p x)))).
New type families or object constructors can be introduced with
condec ::= id : term. % a : K or c : A
Here a stands for a type family and K for its kind,
whereas c is an objects constructor and A its type.
Identifiers are resolved as follows:
x may be bound by the innermost enclosing binder
for x of the form {x:A} or [x:A].
Twelf supports notational definitions and abbreviations. Semantically, both are completely transparent, that is, both for type checking and the operational semantics, definitions may be expanded.
adecl ::= id : term = term % d : A = M
| id = term % d = M
defn ::= adecl. % definition
| %abbrev adecl. % abbreviation
where the second form of declaration is equivalent to id : _ =
term. Definitions or abbreviations can only be made on the level of
objects, not at the level of type families because the interaction of
such definitions with logic programming search has not been fully
investigated.
In order to avoid the expansion of defined constants as much as
possible, declarations id : term = term. must be strict
(see section 4.4 Strict Definitions). A definition of a constant c is
strict if all arguments to c (implicit or explicit) have at least
one strict occurrence (see section 4.3 Strict Occurrences) in the right-hand
side of the definition, and the right-hand side contains at least one
constant. In practice, most notational definitions are strict. For
some examples, see section 3.6 Sample Signature and section 4.4 Strict Definitions.
Twelf tries to preserve strict definitions as much as possible, instead
of expanding them.
The %abbrev declaration defines an abbreviation
which need not be strict. However, it will be expanded immediately
upon parsing and will not be used in output.
The power of definitions in Twelf, however, is severely limited by the lack of recursion. It should only be thought of as notational definition, not as a computational mechanism. Complex operations need to be defined as logic programs, taking advantage of the operational semantics assigned to signatures (see section 5 Logic Programming).
The user may declare constants to be infix, prefix, or postfix
operators. Operator precedence properties are associated with
constants, which must therefore already have been declared with a type
or kind and a possible definition. It is illegal to shadow an infix,
prefix, or postfix operator with a bound variable. We use
nat for the terminal natural numbers.
assoc ::= none % not associative
| left % left associative
| right % right associative
prec ::= nat % 0 <= prec < 10000
ixdecl ::= assoc prec id
pxdecl ::= prec id
decl ::= ...
| %infix ixdecl.
| %prefix pxdecl.
| %postfix pxdecl.
During parsing, ambiguous successive operators of identical precedence
such as a <- b -> c are flagged as errors. Note that it is not
possible to declare an operator with equal or higher precedence than
juxtaposition or equal or lower precedence than `->' and `<-'.
During printing, Twelf frequently has to assign names to anonymous variables. In order to improve readability, the user can declare a name preference for anonymous variables based on their type. Thus name preferences are declared for type family constants. Note that name preferences are not used to disambiguate the types of identifiers during parsing.
namepref ::= id % existential variables
| id id % existential variables, parameters
decl ::= ...
| %name id namepref.
Following our same conventions, a name preference declaration has the
form %name a id, that is, the first identifier must
be a type family already declared and the second is the name preference
for variables of type a. The second identifier must be uppercase,
that is, start with a letter from `A' through `Z' or an
underscore `_'. Anonymous variables will then be named
id1, id2, etc.
In the second form, we can give a separate name preference
for free (existential) variables and parameters. The second one
will typicall be lowercase, as in %name exp E x.
Below is a signature for intuitionistic first-order logic over an unspecified domain of individuals and atomic propositions. It illustrates constant declarations and definitions and the use of operator precedence and name preference declarations. It may be found in the file `examples/guide/nd.elf'.
%%% Individuals
i : type. %name i T
%%% Propositions
o : type. %name o A
imp : o -> o -> o. %infix right 10 imp
and : o -> o -> o. %infix right 11 and
true : o.
or : o -> o -> o. %infix right 11 or
false : o.
forall : (i -> o) -> o.
exists : (i -> o) -> o.
not : o -> o = [A:o] A imp false.
%%% Natural Deductions
nd : o -> type. %name nd D
impi : (nd A -> nd B) -> nd (A imp B).
impe : nd (A imp B) -> nd A -> nd B.
andi : nd A -> nd B -> nd (A and B).
ande1 : nd (A and B) -> nd A.
ande2 : nd (A and B) -> nd B.
truei : nd (true).
% no truee
ori1 : nd A -> nd (A or B).
ori2 : nd B -> nd (A or B).
ore : nd (A or B) -> (nd A -> nd C) -> (nd B -> nd C) -> nd C.
% no falsei
falsee : nd false -> nd C.
foralli : ({x:i} nd (A x)) -> nd (forall A).
foralle : nd (forall A) -> {T:i} nd (A T).
existsi : {T:i} nd (A T) -> nd (exists A).
existse : nd (exists A) -> ({x:i} nd (A x) -> nd C) -> nd C.
noti : (nd A -> nd false) -> nd (not A)
= [D:nd A -> nd false] impi D.
note : nd (not A) -> nd A -> nd false
= [D:nd (not A)] [E:nd A] impe D E.
Go to the first, previous, next, last section, table of contents.