mirror of
https://github.com/c-cube/ocaml-containers.git
synced 2025-12-06 03:05:28 -05:00
527 lines
19 KiB
OCaml
527 lines
19 KiB
OCaml
|
|
(* This file is free software, part of containers. See file "license" for more details. *)
|
|
|
|
(** Basic String Utils (Labeled version of {!CCString}) *)
|
|
|
|
type 'a iter = ('a -> unit) -> unit
|
|
(** Fast internal iterator.
|
|
@since 2.8 *)
|
|
|
|
type 'a gen = unit -> 'a option
|
|
|
|
include module type of struct include StringLabels end
|
|
(** {{: https://caml.inria.fr/pub/docs/manual-ocaml/libref/StringLabels.html} Documentation for the standard StringLabels module} *)
|
|
|
|
val length : t -> int
|
|
(** [length s] returns the length (number of characters) of the given string [s]. *)
|
|
|
|
val blit : src:t -> src_pos:int -> dst:Bytes.t -> dst_pos:int -> len:int -> unit
|
|
(** [blit ~src ~src_pos ~dst ~dst_pos ~len] copies [len] characters from string [src] starting at character indice [src_pos],
|
|
to the Bytes sequence [dst] starting at character indice [dst_pos].
|
|
Like {!String.blit}.
|
|
Compatible with the [-safe-string] option.
|
|
@raise Invalid_argument if indices are not valid. *)
|
|
|
|
(*
|
|
val blit_immut : t -> int -> t -> int -> int -> string
|
|
(** Immutable version of {!blit}, returning a new string.
|
|
[blit a i b j len] is the same as [b], but in which
|
|
the range [j, …, j+len] is replaced by [a.[i], …, a.[i + len]].
|
|
@raise Invalid_argument if indices are not valid. *)
|
|
*)
|
|
|
|
val fold : f:('a -> char -> 'a) -> init:'a -> t -> 'a
|
|
(** [fold ~f ~init s] folds on chars by increasing index. Computes [f(… (f (f init s.[0]) s.[1]) …) s.[n-1]].
|
|
@since 0.7 *)
|
|
|
|
val foldi : f:('a -> int -> char -> 'a) -> 'a -> t -> 'a
|
|
(** [foldi ~f init s] is just like {!fold}, but it also passes in the index of each chars
|
|
as second argument to the folded function [f].
|
|
@since 3.3 *)
|
|
|
|
(** {2 Conversions} *)
|
|
|
|
val to_gen : t -> char gen
|
|
(** [to_gen s] returns the [gen] of characters contained in the string [s]. *)
|
|
|
|
val to_iter : t -> char iter
|
|
(** [to_iter s] returns the [iter] of characters contained in the string [s].
|
|
@since 2.8 *)
|
|
|
|
val to_seq : t -> char Seq.t
|
|
(** [to_seq s] returns the [Seq.t] of characters contained in the string [s].
|
|
Renamed from [to std_seq] since 3.0.
|
|
@since 3.0 *)
|
|
|
|
val to_list : t -> char list
|
|
(** [to_list s] returns the [list] of characters contained in the string [s]. *)
|
|
|
|
val pp_buf : Buffer.t -> t -> unit
|
|
(** [pp_buf buf s] prints [s] to the buffer [buf].
|
|
Renamed from [pp] since 2.0. *)
|
|
|
|
val pp : Format.formatter -> t -> unit
|
|
(** [pp f s] prints the string [s] within quotes to the formatter [f].
|
|
Renamed from [print] since 2.0. *)
|
|
|
|
(** {2 Strings} *)
|
|
|
|
val equal : string -> string -> bool
|
|
(** [equal s1 s2] returns [true] iff the strings [s1] and [s2] are equal. *)
|
|
|
|
val compare : string -> string -> int
|
|
(** [compare s1 s2] compares the strings [s1] and [s2] and returns an integer that indicates
|
|
their relative position in the sort order. *)
|
|
|
|
val is_empty : string -> bool
|
|
(** [is_empty s] returns [true] iff [s] is empty (i.e. its length is 0).
|
|
@since 1.5 *)
|
|
|
|
val hash : string -> int
|
|
(** [hash s] returns the hash value of [s]. *)
|
|
|
|
val rev : string -> string
|
|
(** [rev s] returns the reverse of [s].
|
|
@since 0.17 *)
|
|
|
|
val pad : ?side:[`Left|`Right] -> ?c:char -> int -> string -> string
|
|
(** [pad ?side ?c n s] ensures that the string [s] is at least [n] bytes long,
|
|
and pads it on the [side] with [c] if it's not the case.
|
|
@param side determines where padding occurs (default: [`Left]).
|
|
@param c the char used to pad (default: ' ').
|
|
@since 0.17 *)
|
|
|
|
val of_char : char -> string
|
|
(** [of_char 'a'] is ["a"].
|
|
@since 0.19 *)
|
|
|
|
val of_gen : char gen -> string
|
|
(** [of_gen gen] converts a [gen] of characters to a string. *)
|
|
|
|
val of_iter : char iter -> string
|
|
(** [of_iter iter] converts an [iter] of characters to a string.
|
|
@since 2.8 *)
|
|
|
|
val of_seq : char Seq.t -> string
|
|
(** [of_seq seq] converts a [seq] of characters to a string.
|
|
Renamed from [of_std_seq] since 3.0.
|
|
@since 3.0 *)
|
|
|
|
val of_list : char list -> string
|
|
(** [of_list lc] converts a list of characters [lc] to a string. *)
|
|
|
|
val of_array : char array -> string
|
|
(** [of_array ac] converts an array of characters [ac] to a string. *)
|
|
|
|
val to_array : string -> char array
|
|
(** [to_array s] returns the array of characters contained in the string [s]. *)
|
|
|
|
val find : ?start:int -> sub:(string [@keep_label]) -> string -> int
|
|
(** [find ?start ~sub s] returns the starting index of the first occurrence of [sub] within [s] or [-1].
|
|
@param start starting position in [s]. *)
|
|
|
|
val find_all : ?start:int -> sub:(string [@keep_label]) -> string -> int gen
|
|
(** [find_all ?start ~sub s] finds all occurrences of [sub] in [s], even overlapping instances
|
|
and returns them in a generator [gen].
|
|
@param start starting position in [s].
|
|
@since 0.17 *)
|
|
|
|
val find_all_l : ?start:int -> sub:(string [@keep_label]) -> string -> int list
|
|
(** [find_all_l ?start ~sub s] finds all occurrences of [sub] in [s]
|
|
and returns them in a list.
|
|
@param start starting position in [s].
|
|
@since 0.17 *)
|
|
|
|
val mem : ?start:int -> sub:(string [@keep_label]) -> string -> bool
|
|
(** [mem ?start ~sub s] is [true] iff [sub] is a substring of [s].
|
|
@since 0.12 *)
|
|
|
|
val rfind : sub:(string [@keep_label]) -> string -> int
|
|
(** [rfind ~sub s] finds [sub] in string [s] from the right, returns its first index or [-1].
|
|
Should only be used with very small [sub].
|
|
@since 0.12 *)
|
|
|
|
val replace : ?which:[`Left|`Right|`All] -> sub:(string [@keep_label]) -> by:(string [@keep_label]) -> string -> string
|
|
(** [replace ?which ~sub ~by s] replaces some occurrences of [sub] by [by] in [s].
|
|
@param which decides whether the occurrences to replace are:
|
|
{ul
|
|
{- [`Left] first occurrence from the left (beginning).}
|
|
{- [`Right] first occurrence from the right (end).}
|
|
{- [`All] all occurrences (default).}
|
|
}
|
|
@raise Invalid_argument if [sub = ""].
|
|
@since 0.14 *)
|
|
|
|
val is_sub : sub:(string [@keep_label]) -> sub_pos:int -> string -> pos:int -> sub_len:(int [@keep_label]) -> bool
|
|
(** [is_sub ~sub ~sub_pos s ~pos ~sub_len] returns [true] iff the substring of [sub]
|
|
starting at position [sub_pos] and of length [sub_len] is a substring of [s]
|
|
starting at position [pos]. *)
|
|
|
|
val repeat : string -> int -> string
|
|
(** [repeat s n] creates a string by repeating the string [s] [n] times. *)
|
|
|
|
val prefix : pre:(string [@keep_label]) -> string -> bool
|
|
(** [prefix ~pre s] returns [true] iff [pre] is a prefix of [s]. *)
|
|
|
|
val suffix : suf:(string [@keep_label]) -> string -> bool
|
|
(** [suffix ~suf s] returns [true] iff [suf] is a suffix of [s].
|
|
@since 0.7 *)
|
|
|
|
val chop_prefix : pre:(string [@keep_label]) -> string -> string option
|
|
(** [chop_prefix ~pre s] removes [pre] from [s] if [pre] really is a prefix
|
|
of [s], returns [None] otherwise.
|
|
@since 0.17 *)
|
|
|
|
val chop_suffix : suf:(string [@keep_label]) -> string -> string option
|
|
(** [chop_suffix ~suf s] removes [suf] from [s] if [suf] really is a suffix
|
|
of [s], returns [None] otherwise.
|
|
@since 0.17 *)
|
|
|
|
val take : int -> string -> string
|
|
(** [take n s] keeps only the [n] first chars of [s].
|
|
@since 0.17 *)
|
|
|
|
val drop : int -> string -> string
|
|
(** [drop n s] removes the [n] first chars of [s].
|
|
@since 0.17 *)
|
|
|
|
val take_drop : int -> string -> string * string
|
|
(** [take_drop n s] is [take n s, drop n s].
|
|
@since 0.17 *)
|
|
|
|
val lines : string -> string list
|
|
(** [lines s] returns a list of the lines of [s] (splits along '\n').
|
|
@since 0.10 *)
|
|
|
|
val lines_gen : string -> string gen
|
|
(** [lines_gen s] returns a generator [gen] of the lines of [s] (splits along '\n').
|
|
@since 0.10 *)
|
|
|
|
val lines_iter : string -> string iter
|
|
(** [lines_iter s] returns the [iter] of the lines of [s] (splits along '\n').
|
|
@since 3.2 *)
|
|
|
|
val lines_seq : string -> string Seq.t
|
|
(** [lines_seq s] returns the [Seq.t] of the lines of [s] (splits along '\n').
|
|
@since 3.2 *)
|
|
|
|
val concat_iter : sep:string -> string iter -> string
|
|
(** [concat_iter ~sep iter] concatenates all strings of [iter], separated with [sep].
|
|
@since 3.2 *)
|
|
|
|
val concat_gen : sep:(string [@keep_label]) -> string gen -> string
|
|
(** [concat_gen ~sep gen] concatenates all strings of [gen], separated with [sep].
|
|
@since 0.10 *)
|
|
|
|
val concat_seq : sep:string -> string Seq.t -> string
|
|
(** [concat_seq ~sep seq] concatenates all strings of [seq], separated with [sep].
|
|
@since 3.2 *)
|
|
|
|
val unlines : string list -> string
|
|
(** [unlines ls] concatenates all strings of [ls], separated with '\n'.
|
|
@since 0.10 *)
|
|
|
|
val unlines_gen : string gen -> string
|
|
(** [unlines_gen gen] concatenates all strings of [gen], separated with '\n'.
|
|
@since 0.10 *)
|
|
|
|
val unlines_iter : string iter -> string
|
|
(** [unlines_iter iter] concatenates all strings of [iter], separated with '\n'.
|
|
@since 3.2 *)
|
|
|
|
val unlines_seq : string Seq.t -> string
|
|
(** [unlines_seq seq] concatenates all strings of [seq], separated with '\n'.
|
|
@since 3.2 *)
|
|
|
|
val set : string -> int -> char -> string
|
|
(** [set s i c] creates a new string which is a copy of [s], except
|
|
for index [i], which becomes [c].
|
|
@raise Invalid_argument if [i] is an invalid index.
|
|
@since 0.12 *)
|
|
|
|
val iter : f:(char -> unit) -> string -> unit
|
|
(** [iter ~f s] applies function [f] on each character of [s].
|
|
Alias to {!String.iter}.
|
|
@since 0.12 *)
|
|
|
|
val filter_map : f:(char -> char option) -> string -> string
|
|
(** [filter_map ~f s] calls [(f a0) (f a1) … (f an)] where [a0 … an] are the characters of s.
|
|
It returns the string of characters [ci] such as [f ai = Some ci] (when [f] returns [None],
|
|
the corresponding element of [s] is discarded).
|
|
@since 0.17 *)
|
|
|
|
val filter : f:(char -> bool) -> string -> string
|
|
(** [filter ~f s] discards characters of [s] not satisfying [f].
|
|
@since 0.17 *)
|
|
|
|
val uniq : eq:(char -> char -> bool) -> string -> string
|
|
(** [uniq ~eq s] remove consecutive duplicate characters in [s].
|
|
@since 3.4 *)
|
|
|
|
val flat_map : ?sep:string -> f:(char -> string) -> string -> string
|
|
(** [flat_map ?sep ~f s] maps each chars of [s] to a string, then concatenates them all.
|
|
@param sep optional separator between each generated string.
|
|
@since 0.12 *)
|
|
|
|
val for_all : f:(char -> bool) -> string -> bool
|
|
(** [for_all ~f s] is [true] iff all characters of [s] satisfy the predicate [f].
|
|
@since 0.12 *)
|
|
|
|
val exists : f:(char -> bool) -> string -> bool
|
|
(** [exists ~f s] is [true] iff some character of [s] satisfy the predicate [f].
|
|
@since 0.12 *)
|
|
|
|
val drop_while : f:(char -> bool) -> t -> t
|
|
(** [drop_while ~f s] discards any characters of [s] starting from the left,
|
|
up to the first character [c] not satisfying [f c].
|
|
@since 2.2 *)
|
|
|
|
val rdrop_while : f:(char -> bool) -> t -> t
|
|
(** [rdrop_while ~f s] discards any characters of [s] starting from the right,
|
|
up to the first character [c] not satisfying [f c].
|
|
@since 2.2 *)
|
|
|
|
val ltrim : t -> t
|
|
(** [ltrim s] trims space on the left (see {!String.trim} for more details).
|
|
@since 1.2 *)
|
|
|
|
val rtrim : t -> t
|
|
(** [rtrim s] trims space on the right (see {!String.trim} for more details).
|
|
@since 1.2 *)
|
|
|
|
(** {2 Operations on 2 strings} *)
|
|
|
|
val map2 : f:(char -> char -> char) -> string -> string -> string
|
|
(** [map2 ~f s1 s2] maps pairs of chars.
|
|
@raise Invalid_argument if the strings have not the same length.
|
|
@since 0.12 *)
|
|
|
|
val iter2: f:(char -> char -> unit) -> string -> string -> unit
|
|
(** [iter2 ~f s1 s2] iterates on pairs of chars.
|
|
@raise Invalid_argument if the strings have not the same length.
|
|
@since 0.12 *)
|
|
|
|
val iteri2: f:(int -> char -> char -> unit) -> string -> string -> unit
|
|
(** [iteri2 ~f s1 s2] iterates on pairs of chars with their index.
|
|
@raise Invalid_argument if the strings have not the same length.
|
|
@since 0.12 *)
|
|
|
|
val fold2: f:('a -> char -> char -> 'a) -> init:'a -> string -> string -> 'a
|
|
(** [fold2 ~f ~init s1 s2] folds on pairs of chars.
|
|
@raise Invalid_argument if the strings have not the same length.
|
|
@since 0.12 *)
|
|
|
|
val for_all2 : f:(char -> char -> bool) -> string -> string -> bool
|
|
(** [for_all2 ~f s1 s2] returns [true] iff all pairs of chars satisfy the predicate [f].
|
|
@raise Invalid_argument if the strings have not the same length.
|
|
@since 0.12 *)
|
|
|
|
val exists2 : f:(char -> char -> bool) -> string -> string -> bool
|
|
(** [exists2 ~f s1 s2] returns [true] iff a pair of chars satisfy the predicate [f].
|
|
@raise Invalid_argument if the strings have not the same length.
|
|
@since 0.12 *)
|
|
|
|
(** {2 Ascii functions}
|
|
|
|
Those functions are deprecated in {!String} since 4.03, so we provide
|
|
a stable alias for them even in older versions. *)
|
|
|
|
val capitalize_ascii : string -> string
|
|
(** [capitalize_ascii s] returns a copy of [s] with the first character set to uppercase
|
|
using the US-ASCII character set.
|
|
See {!String}.
|
|
@since 0.18 *)
|
|
|
|
val uncapitalize_ascii : string -> string
|
|
(** [uncapitalize_ascii s] returns a copy of [s] with the first character set to lowercase
|
|
using the US-ASCII character set.
|
|
See {!String}.
|
|
@since 0.18 *)
|
|
|
|
val uppercase_ascii : string -> string
|
|
(** [uppercase_ascii s] returns a copy of [s] with all lowercase letters translated to uppercase
|
|
using the US-ASCII character set.
|
|
See {!String}.
|
|
@since 0.18 *)
|
|
|
|
val lowercase_ascii : string -> string
|
|
(** [lowercase_ascii s] returns a copy of [s] with all uppercase letters translated to lowercase
|
|
using the US-ASCII character set.
|
|
See {!String}.
|
|
@since 0.18 *)
|
|
|
|
val equal_caseless : string -> string -> bool
|
|
(** [equal_caseless s1 s2] compares [s1] and [s2] without respect to {b ascii} lowercase.
|
|
@since 1.2 *)
|
|
|
|
(** {2 Finding}
|
|
|
|
A relatively efficient algorithm for finding sub-strings.
|
|
@since 1.0 *)
|
|
|
|
module Find : sig
|
|
type _ pattern
|
|
|
|
val compile : string -> [ `Direct ] pattern
|
|
|
|
val rcompile : string -> [ `Reverse ] pattern
|
|
|
|
val find : ?start:int -> pattern:(([`Direct] pattern) [@keep_label]) -> string -> int
|
|
(** [find ?start ~pattern s] searches for [pattern] in the string [s], left-to-right.
|
|
@return the offset of the first match, -1 otherwise.
|
|
@param start offset in string at which we start. *)
|
|
|
|
val rfind : ?start:int -> pattern:(([`Reverse] pattern) [@keep_label]) -> string -> int
|
|
(** [rfind ?start ~pattern s] searches for [pattern] in the string [s], right-to-left.
|
|
@return the offset of the start of the first match from the right, -1 otherwise.
|
|
@param start right-offset in string at which we start. *)
|
|
end
|
|
|
|
(** {2 Splitting} *)
|
|
|
|
module Split : sig
|
|
(** Specification of what to do with empty blocks, as in [split ~by:"-" "-a-b-"].
|
|
|
|
- [{first=false; last=false}] will return [""; "a"; "b"; ""]
|
|
- [{first=true; last=false}] will return ["a"; "b" ""]
|
|
- [{first=false; last=true}] will return [""; "a"; "b"]
|
|
- [{first=true; last=true}] will return ["a"; "b"]
|
|
|
|
The default value of all remaining functions is [Drop_none].
|
|
@since 1.5
|
|
*)
|
|
type drop_if_empty = {
|
|
first: bool;
|
|
last: bool;
|
|
}
|
|
|
|
val no_drop : drop_if_empty
|
|
(** [no_drop] does not drop any group, even empty and on borders.
|
|
@since 1.5 *)
|
|
|
|
val list_ : ?drop:drop_if_empty -> by:(string [@keep_label]) -> string -> (string*int*int) list
|
|
(** [list_ ?drop ~by s] splits the given string [s] along the given separator [by].
|
|
Should only be used with very small separators, otherwise use {!Containers_string.KMP}.
|
|
@return a [list] of slices [(s,index,length)] that are
|
|
separated by [by]. {!String.sub} can then be used to actually extract
|
|
a string from the slice.
|
|
@raise Failure if [by = ""]. *)
|
|
|
|
val gen : ?drop:drop_if_empty -> by:string -> string -> (string*int*int) gen
|
|
(** [gen ?drop ~by s] splits the given string [s] along the given separator [by].
|
|
Returns a [gen] of slices. *)
|
|
|
|
val iter : ?drop:drop_if_empty -> by:string -> string -> (string*int*int) iter
|
|
(** [iter ?drop ~by s] splits the given string [s] along the given separator [by].
|
|
Returns an [iter] of slices.
|
|
@since 2.8 *)
|
|
|
|
val seq : ?drop:drop_if_empty -> by:string -> string -> (string*int*int) Seq.t
|
|
(** [seq ?drop ~by s] splits the given string [s] along the given separator [by].
|
|
Returns a [Seq.t] of slices.
|
|
Renamed from [std_seq] since 3.0.
|
|
@since 3.0 *)
|
|
|
|
(** {4 Copying functions}
|
|
|
|
Those split functions actually copy the substrings, which can be
|
|
more convenient but less efficient in general. *)
|
|
|
|
val list_cpy : ?drop:drop_if_empty -> by:string -> string -> string list
|
|
(** [list_cpy ?drop ~by s] splits the given string [s] along the given separator [by].
|
|
Returns a [list] of strings. *)
|
|
|
|
val gen_cpy : ?drop:drop_if_empty -> by:string -> string -> string gen
|
|
(** [gen_cpy ?drop ~by s] splits the given string [s] along the given separator [by].
|
|
Returns a [gen] of strings. *)
|
|
|
|
val iter_cpy : ?drop:drop_if_empty -> by:string -> string -> string iter
|
|
(** [iter_cpy ?drop ~by s] splits the given string [s] along the given separator [by].
|
|
Returns an [iter] of strings.
|
|
@since 2.8 *)
|
|
|
|
val seq_cpy : ?drop:drop_if_empty -> by:string -> string -> string Seq.t
|
|
(** [seq_cpy ?drop ~by s] splits the given string [s] along the given separator [by].
|
|
Returns a [Seq.t] of strings.
|
|
Renamed from [std_seq_cpy] since 3.0.
|
|
@since 3.0 *)
|
|
|
|
val left : by:(string [@keep_label]) -> string -> (string * string) option
|
|
(** [left ~by s] splits on the first occurrence of [by] from the leftmost part
|
|
of the string [s].
|
|
@since 0.12 *)
|
|
|
|
val left_exn : by:(string [@keep_label]) -> string -> string * string
|
|
(** [left_exn ~by s] splits on the first occurrence of [by] from the leftmost part
|
|
of the string [s].
|
|
@raise Not_found if [by] is not part of the string [s].
|
|
@since 0.16 *)
|
|
|
|
val right : by:(string [@keep_label]) -> string -> (string * string) option
|
|
(** [right ~by s] splits on the first occurrence of [by] from the rightmost part
|
|
of the string [s].
|
|
@since 0.12 *)
|
|
|
|
val right_exn : by:(string [@keep_label]) -> string -> string * string
|
|
(** [right_exn ~by s] splits on the first occurrence of [by] from the rightmost part
|
|
of the string [s].
|
|
@raise Not_found if [by] is not part of the string [s].
|
|
@since 0.16 *)
|
|
|
|
end
|
|
|
|
val split_on_char : by:char -> string -> string list
|
|
(** [split_on_char ~by s] splits the string [s] along the given char [by].
|
|
@since 1.2 *)
|
|
|
|
val split : by:(string [@keep_label]) -> string -> string list
|
|
(** [split ~by s] splits the string [s] along the given string [by].
|
|
Alias to {!Split.list_cpy}.
|
|
@since 1.2 *)
|
|
|
|
(** {2 Utils} *)
|
|
|
|
val compare_versions : string -> string -> int
|
|
(** [compare_versions s1 s2] compares {i version strings} [s1] and [s2],
|
|
considering that numbers are above text.
|
|
@since 0.13 *)
|
|
|
|
val compare_natural : string -> string -> int
|
|
(** [compare_natural s1 s2] is the Natural Sort Order, comparing chunks of digits as natural numbers.
|
|
https://en.wikipedia.org/wiki/Natural_sort_order
|
|
@since 1.3 *)
|
|
|
|
val edit_distance : ?cutoff:int -> string -> string -> int
|
|
(** [edit_distance ?cutoff s1 s2] is the edition distance between the two strings [s1] and [s2].
|
|
This satisfies the classical distance axioms: it is always positive, symmetric, and satisfies
|
|
the formula [distance s1 s2 + distance s2 s3 >= distance s1 s3].
|
|
@param cutoff if provided, it's a cap on both the number of iterations,
|
|
and on the result. (since 3.0). This is useful if you just want to
|
|
check whether the edit distance is less or equal than 2 (use cutoff of 3).
|
|
*)
|
|
|
|
(** {2 Infix operators}
|
|
|
|
@since 3.0 *)
|
|
|
|
module Infix : sig
|
|
val (=) : t -> t -> bool
|
|
(** @since 3.0 *)
|
|
|
|
val (<>) : t -> t -> bool
|
|
(** @since 3.0 *)
|
|
|
|
val (<) : t -> t -> bool
|
|
(** @since 3.0 *)
|
|
|
|
val (<=) : t -> t -> bool
|
|
(** @since 3.0 *)
|
|
|
|
val (>=) : t -> t -> bool
|
|
(** @since 3.0 *)
|
|
|
|
val (>) : t -> t -> bool
|
|
(** @since 3.0 *)
|
|
end
|
|
|
|
include module type of Infix
|