pacomb 1.0 · OCaml Package

Combinator library, using continuation

As usual left recursion is not supported, but the library is intended to be used through the Grammar module that provides elimination of left recursion. However, a cache combinatr is supported to overcome the cost of backtracking.

function and type usefull to the end-user

type 'a t

The type of combinator

val partial_parse_buffer : 
  'a t ->
  Blank.t ->
  ?blank_after:bool ->
  Lex.buf ->
  Lex.pos ->
  'a * Lex.buf * Lex.pos

Partial parsing. Beware, the returned position is not the maximum position that can be reached by the grammar.

val parse_all_buffer : 'a t -> Blank.t -> Lex.buf -> Lex.pos -> 'a list

Returns all possible parse trees. Usefull for natural languages but also to debug ambiguity in a supposed non ambiguous grammar. If end of input is not parsed in some ways, some value may correspond to only the beginning of the input. Except when debugging or testing, you should rather use cache/merge anyway.

combinator constructors, normally not needed by the casual user

val fail : 'a t

fail is a parser rejecting every input (it always fails).

val error : string -> 'a t

Fails and report an error

val assert_false : 'a t

val empty : 'a -> 'a t

empty v is a parser that only accepts the empty input and returns v as its semantic value.

val lexeme : 'a Lex.lexeme -> 'a t

lexeme l is a parser accepting the lexeme (or terminal) l, and returns the corresponding semantic value.

val seq : 'a t -> ('a -> 'b) t -> 'b t

seq g1 g2 sequences the parsers g1 and g2. The resulting parser starts by parsing using g1, and then parses the rest of the input using g2. The result of parsing with g2 is then apply to the result of g1.

val dseq : ('a * 'b) t -> ('a -> ('b -> 'c) t) -> 'c t

dseq c1 c2 is a dependant sequence, contrary to seq c1 c2, the combinator used to parse after c1 depends upon the first value returned by c1. It is a good idea to memoize the function c2. The separation of 'a and 'b in the smeantics of g1 allows to depend on the smallest set of possible vaue which is important in case of memoisation.

val alt : Charset.t -> 'a t -> Charset.t -> 'a t -> 'a t

Combinator parsing with the first combinator and in case of failure with the second from the same position. The optionnal charset corresponds to the charaters accepted at the beginning of the input for each combinators. The charset must be Charset.full if the corresponding combinator accept the empty input

val alts : (Charset.t * 'a t) list -> 'a t

same as above but with a list of alternative.

val option : 'a -> Charset.t -> 'a t -> 'a t

option a ~cs c is an optimisation for alt (empty a) ~cs c. In fact it is better to use alt with grammar not accepting empty and use option to deal with an empty case

val app : 'a t -> ('a -> 'b) -> 'b t

Parses with the given combinator and transforms the semantics with the given function

val eval : 'a t -> 'a t

forces immediate evaluation of the action just after parsing

val left_pos : (Pos.t -> 'a) t -> 'a t

Parses as the given combinator and give the position to the left of the parsing input as argument to the action

val right_pos : (Pos.t -> 'a) t -> 'a t

Same as above with the position to the right

val read_pos : Pos.t Assoc.key -> (Pos.t -> 'a) t -> 'a t

To eliminate left recursion, lpos has to be left factored. if lpos is one single combinator, this adds a lot of closures in action code. To solve this problem, lpos is splitted in two combinators, one that pushes the position to a stack and pops after parsing and another that reads the position.

type 'a key = 'a Stdlib.Lazy.t Assoc.key

key used by lr below

val lr : 'a t -> 'a key -> 'a t -> 'a t

lr c1 v c2 is an optimized version of let rec r = seq c1 (seq r c2) which is illegal as it is left recursive and loops. The optional charset indicates the characteres accepted by c2 at the beginning of input. v is like variable bound in c2, see read_tbl below

val lr_pos : 'a t -> 'a key -> Pos.t Assoc.key -> 'a t -> 'a t

Same as above, but also store the position

type mlr_left =

| LNil : mlr_left
| LCns : 'a key * Charset.t * 'a t * mlr_left -> mlr_left

type to represent the left prefix of a mutually recursive grammar. the key represents the produced grammar for each left prefix.

type mlr_right =

| RNil : mlr_right
| RCns : 'a key * 'b key * Charset.t * 'b t * mlr_right -> mlr_right

type of the suffix to be repeted in a mutually recursive grammar. the first key represents the grammar that parsed the input before the second key represents the produced grammar.

Somehow, mlr_right is a matrix R, the two keys being the index of the coefficient and mlr_left is a vector L. Parsing, will somehow use L R^n for n large enough;

val mlr : ?lpos:Pos.t Assoc.key -> mlr_left -> mlr_right -> 'a key -> 'a t

The combinator itself. The optionnal argument indicated that we need the position before parsing

val read_tbl : 'a key -> 'a t

combinator to access the value stored by lr. It must be uses as prefix of c2 in lr c1 c2. For instance, the coding of let rec r = seq c1 (seq r c2) is let k = Assoc.new_key () in lr c1 k (seq (read_tbl k) c2). Here we ommited the actions. This way of coding left recursion avoids to transform the action and produce closure. The code for elimination of left recursion is also much simpler

val test_before : 
  (Lex.buf -> Lex.pos -> Lex.buf -> Lex.pos -> bool) ->
  'a t ->
  'a t

Allow to test the blank characteres before a grammar and more

val test_after : 
  ('a -> Lex.buf -> Lex.pos -> Lex.buf -> Lex.pos -> bool) ->
  'a t ->
  'a t

Allow to test the blank characteres after a grammar and more

val deref : 'a t Stdlib.ref -> 'a t

Access to a reference to a combinator, used by Grammar.compile for recursive grammars (not for left recursion

val change_layout : ?config:Blank.layout_config -> Blank.t -> 'a t -> 'a t

Change the blank function used to parse with the given combinator. we can choose which blank to use at the boundary with the optional parameters.

val cache : ?merge:('a -> 'a -> 'a) -> 'a t -> 'a t

Combinator that caches a grammar to avoid exponential behavior. parsing with the grammar from each position is memoized to avoid parsing twice the same sequence with the same grammar.

package pacomb

Combinator library, using continuation

function and type usefull to the end-user

combinator constructors, normally not needed by the casual user