Language.Prolog.Read
This package contains an implementation of a tool
(pl2curry
) to transform Prolog programs to Curry programs.
The idea of this tool is to demonstrate the advantages of functional
logic languages compared to purely logic languages. Thus, the tool
translates only pure logic programs (without side effecting predicates
etc). The initial ideas of this tool are described in detail in a paper presented at ICLP
2022 and published in TPLP.
The tool has various options to influence the kind of transformation, e.g.:
--conservative
: transform each Prolog predicate into
a Curry predicate
Note that this is always possible since non-linear left-hand sides are allowed in Curry (in contrast to Haskell).
with functions (this is the default provided that
--conservative
is not set): specify some predicates as
functions by:
:- function p/n.
: The last argument is the result.:- function p/n: i.
: The i-th argument is the result
(i=n: last argument).:- function p/n: [i1,...,ik].
: Argument positions
[i1,…,ik] are put as result arguments.For instance, if p/3
is a function and the last argument
is the result, then a goal p(X,Y,Z)
is transformed into
z =:= p x y
.
Note that it is necessary to define function
as an
operator in Prolog in order to read Prolog programs with such
directives. This can be done by adding the following directives at the
beginning of the Prolog program:
:- op(1150,fx,function).
function(_).
with demand (this is the default provided that
--nodemand
is not set): similarly to functions, but
function calls are transformed into local variable bindings rather than
unifications. Hence, functions are evaluated only if its result is
demanded (due to Curry’s lazy evaluation strategy).
with inlining (this is the default provided that
--noinline
is not set): similarly to demand, but bindings
are inlined (if possible) to obtain a more compact source code
Since adding function
directives to specify result
argument positions is tedious, the tool also contains an analysis to
derive automatic function
directives (if not already
explicitly provided and if the option --noanalysis
is not
set) for standard functions (i.e., where only the last argument is a
result position). It is based on the following principle:
p
is an n-ary predicate and there is a (minimal) set
of argument position such that the rules for p
are
inductively sequentially defined on this set of argument positions (in
particular, non-overlapping w.r.t. these arguments), p
is
considered as a function (where the last argument is the result argument
position, or the maximum of the remaining argument positions if the
option --anyresult
is set).The information about the inferred sets of inductively sequential argument and result arguments of functions is printed if the verbosity is larger than 2.
A special case of the previous criterion are predicates defined by a single rule, e.g., predicates which define constants, as
two(s(s(o))).
This will be translated into
two = S (S O)
Although this is correct, it is sometimes unintended, e.g., if
p
is defined by the single clause
p(X,Y) :- q(X,Z), r(Z,Y).
In order to keep such predicates as predicates on the Curry level, the following heuristic is used. A predicate defined by a single rule is transformed into a function only if the last argument in the left-hand side is not a variable or a variable which occurs in a result argument position in the rule’s body.
Although these heuristics provide expected transformations in most
case, one can always override them using an explicit
function
directive.
To provide a more reasonable translation to Curry, the translation tool considers also type declarations. These declarations are similarly to polymoprhic algebraic data types but use a Prolog-like syntax. For instance, the Prolog program could contain the directives
:- type nat = o ; s(nat).
:- type tree(A) = leaf(A) ; node(tree(A),tree(A)).
These are translated into the Curry type declarations
All other constructors occurring in the logic program (except for
true
and false
and list constructors, see
below) are declared in a single type named Term
in
Curry.
Note that it is necessary to define type
as an operator
in Prolog in order to read Prolog programs with such directives. This
can be done by adding the following directives at the beginning of the
Prolog program:
:- op(1150,fx,type).
type(_).
Due to the fact the Prolog programs are transformed (in the default
case) into nested functions which are lazily evaluated in Curry, it
might be the case that the Curry program computes more answers than the
original Prolog program. This might be the case if a failing or
non-terminating predicate is evaluated in Prolog but not demanded in the
transformed Curry program. In general, this can be considered as an
advantage of functional logic programming compared to pure logic
programming. However, if it is intended to keep the same answer
semantics between Prolog and the generated Curry programs, one can
specify a set of failing functions (i.e., functions that are
not totally defined due to partial pattern matching or infinite
computations). In this case, any occurrence of such a failing function
will be strictly evaluated in the Curry program. This fail-sensitvie
transformation will be used if the option
--failfuncs=F
is provided. In this case, the file
F
contains in each line a failing function in the form
Mod f
, i.e., f
is a function defined in module
Mod
. Such files can be generated by the tool
curry-calltypes
(see Curry package
verify-non-fail
). Thus, the shell script
scripts/pl2curry-failsensitive.sh
can be used to apply the
fail-sensitive transformation with these tools.
Prolog atoms true
and false
are
translated into the Curry Boolean constants True
and
False
.
Prolog lists are transformed into Curry lists (unless the option
--nolists
is set). This should yield the intended code but
might produce type errors for strange uses of Prolog lists, e.g.,
.(1,.(2,3))
. If Curry lists are not used (by setting the
option --nolists
), Prolog lists are transformed into Curry
terms by using the constructors NIL
and CONS
,
e.g., the Prolog list [1,2]
is transformed into
CONS 1 (CONS 2 NIL)
.
In the default case, only the last argument of a predicate will be considered as a result argument position for inferred functions. Hence, if the last argument is contained in a minimal set of inductively sequential arguments, it will not be transformed into a function.
This behavior can be changed by setting the option
--anyresult
. In this case, a maximum argument will be
selected. If this is not the last one, the index of the result argument
position is added to indicate that the order of arguments has been
changed in the transformed function. See examples/demand.pl
for such examples.