Previous  Up  Next

Module Std_common.Bigstring

module Bigstring: Bigstring

Types and exceptions

type t = (char, Bigarray.int8_unsigned_elt, Bigarray.c_layout) Bigarray.Array1.t
Type of bigstrings
include Equal.S

Creation and string conversion

val create : ?max_mem_waiting_gc:Byte_units.t -> int -> t
create length
Returns a new bigstring having length. Content is undefined.
max_mem_waiting_gc : default = 256 M in OCaml <= 3.12, 1 G otherwise. As the total allocation of calls to create approach max_mem_waiting_gc, the pressure in the garbage collector to be more agressive will increase.
val init : int -> f:(int -> char) -> t
init n ~f creates a bigstring t of length n, with t.{i} = f i
val of_string : ?pos:int -> ?len:int -> string -> t
of_string ?pos ?len str
Returns a new bigstring that is equivalent to the substring of length len in str starting at position pos.
pos : default = 0
len : default = String.length str - pos
val to_string : ?pos:int -> ?len:int -> t -> string
to_string ?pos ?len bstr
Raises Invalid_argument if the string would exceed runtime limits.
Returns a new string that is equivalent to the substring of length len in bstr starting at position pos.
pos : default = 0
len : default = length bstr - pos

Checking

val check_args : loc:string -> pos:int -> len:int -> t -> unit
check_args ~loc ~pos ~len bstr checks the position and length arguments pos and len for bigstrings bstr.
Raises Invalid_argument if these arguments are illegal for the given bigstring using loc to indicate the calling context.
val get_opt_len : t -> pos:int -> int option -> int
get_opt_len bstr ~pos opt_len
Returns the length of a subbigstring in bstr starting at position pos and given optional length opt_len. This function does not check the validity of its arguments. Use Bigstring.check_args for that purpose.

Accessors

val length : t -> int
length bstr
Returns the length of bigstring bstr.
val sub_shared : ?pos:int -> ?len:int -> t -> t
sub_shared ?pos ?len bstr
Returns the sub-bigstring in bstr that starts at position pos and has length len. The sub-bigstring shares the same memory region, i.e. modifying it will modify the original bigstring. Holding on to the sub-bigstring will also keep the (usually bigger) original one around.
pos : default = 0
len : default = Bigstring.length bstr - pos
val get : t -> int -> char
get t pos returns the character at pos
val set : t -> int -> char -> unit
set t pos sets the character at pos
val is_mmapped : t -> bool
is_mmapped bstr
Returns whether the bigstring bstr is memory-mapped.

Blitting


blit ~src ?src_pos ?src_len ~dst ?dst_pos () blits src_len characters from src starting at position src_pos to dst at position dst_pos.
include Blit.S
module To_string: Blit.S_distinct  with type src := t       with type dst := string
module From_string: Blit.S_distinct  with type src := string  with type dst := t

Reading/writing bin-prot


These functions write the "size-prefixed" bin-prot format that is used by, e.g., async's Writer.write_bin_prot, Reader.read_bin_prot and Unpack_buffer.Unpack_one.create_bin_prot.
val write_bin_prot : t -> ?pos:int -> 'a Bin_prot.Type_class.writer -> 'a -> int
write_bin_prot t writer a writes a to t starting at pos, and returns the index in t immediately after the last byte written. It raises if pos < 0 or if a doesn't fit in t.
val read_bin_prot : t ->
       ?pos:int ->
       ?len:int -> 'a Bin_prot.Type_class.reader -> ('a * int) Or_error.t
The read_bin_prot* functions read from the region of t starting at pos of length len. They return the index in t immediately after the last byte read. They raise if pos and len don't describe a region of t.
val read_bin_prot_verbose_errors : t ->
       ?pos:int ->
       ?len:int ->
       'a Bin_prot.Type_class.reader ->
       [ `Invalid_data of Error.t | `Not_enough_data | `Ok of 'a * int ]

Memory mapping

val map_file : shared:bool -> Unix.file_descr -> int -> t
map_file shared fd n memory-maps n characters of the data associated with descriptor fd to a bigstring. Iff shared is true, all changes to the bigstring will be reflected in the file.

Users must keep in mind that operations on the resulting bigstring may result in disk operations which block the runtime. This is true for pure OCaml operations (such as t.{1} <- 1), and for calls to blit. While some I/O operations may release the OCaml lock, users should not expect this to be done for all operations on a bigstring returned from map_file.



val find : ?pos:int -> ?len:int -> char -> t -> int option
find ?pos ?len char t returns Some i for the smallest i >= pos such that t.{i} = char, or None if there is no such i.
pos : default = 0
len : default = length bstr - pos

Destruction

val unsafe_destroy : t -> unit
unsafe_destroy bstr destroys the bigstring by deallocating its associated data or, if memory-mapped, unmapping the corresponding file, and setting all dimensions to zero. This effectively frees the associated memory or address-space resources instantaneously. This feature helps working around a bug in the current OCaml runtime, which does not correctly estimate how aggressively to reclaim such resources.

This operation is safe unless you have passed the bigstring to another thread that is performing operations on it at the same time. Access to the bigstring after this operation will yield array bounds exceptions.
Raises Failure if the bigstring has already been deallocated (or deemed "external", which is treated equivalently), or if it has proxies, i.e. other bigstrings referring to the same data.


Accessors for parsing binary values, analogous to binary_packing. These are in Bigstring rather than a separate module because:

1) Existing binary_packing requires copies and does not work with bigstrings 2) The accessors rely on the implementation of bigstring, and hence should changeshould the implementation of bigstring move away from Bigarray. 3) Bigstring already has some external C functions, so it didn't require many changes to the OMakefile ^_^.

In a departure from Binary_packing, the naming conventions are chosen to be close to C99 stdint types, as it's a more standard description and it is somewhat useful in making compact macros for the implementations. The accessor names contain endian-ness to allow for branch-free implementations

<accessor> ::= <unsafe><operation><type><endian><int> <unsafe> ::= unsafe_ | '' <operation> ::= get_ | set_ <type> ::= int16 | uint16 | int32 | int64 <endian> ::= _le | _be | '' <int> ::= _int | ''

The "unsafe_" prefix indicates that these functions do no bounds checking. Performance testing demonstrated that the bounds check was 2-3 times slower due to the fact that Bigstring.length is a C call, and not even a noalloc one. In practice, message parsers can check the size of an outer message once, and use the unsafe accessors for individual fields, so many bounds checks can end up being redundant as well. The situation could be improved by having bigarray cache the length/dimensions.

val unsafe_get_int8 : t -> pos:int -> int
val unsafe_set_int8 : t -> pos:int -> int -> unit
val unsafe_get_uint8 : t -> pos:int -> int
val unsafe_set_uint8 : t -> pos:int -> int -> unit
val unsafe_get_int16_le : t -> pos:int -> int
16 bit methods

val unsafe_get_int16_be : t -> pos:int -> int
val unsafe_set_int16_le : t -> pos:int -> int -> unit
val unsafe_set_int16_be : t -> pos:int -> int -> unit
val unsafe_get_uint16_le : t -> pos:int -> int
val unsafe_get_uint16_be : t -> pos:int -> int
val unsafe_set_uint16_le : t -> pos:int -> int -> unit
val unsafe_set_uint16_be : t -> pos:int -> int -> unit
val unsafe_get_int32_le : t -> pos:int -> int
32 bit methods

val unsafe_get_int32_be : t -> pos:int -> int
val unsafe_set_int32_le : t -> pos:int -> int -> unit
val unsafe_set_int32_be : t -> pos:int -> int -> unit
val unsafe_get_uint32_le : t -> pos:int -> int
val unsafe_get_uint32_be : t -> pos:int -> int
val unsafe_set_uint32_le : t -> pos:int -> int -> unit
val unsafe_set_uint32_be : t -> pos:int -> int -> unit

Similar to the usage in binary_packing, the below methods are treating the value being read (or written), as an ocaml immediate integer, as such it is actually 63 bits. If the user is confident that the range of values used in practice will not require 64 bit precision (i.e. Less than Max_Long), then we can avoid allocation and use an immediate. If the user is wrong, an exception will be thrown (for get).
val unsafe_get_int64_le_exn : t -> pos:int -> int
val unsafe_get_int64_be_exn : t -> pos:int -> int
val unsafe_set_int64_le : t -> pos:int -> int -> unit
val unsafe_set_int64_be : t -> pos:int -> int -> unit
val unsafe_get_int64_le_trunc : t -> pos:int -> int
val unsafe_get_int64_be_trunc : t -> pos:int -> int
val unsafe_get_int32_t_le : t -> pos:int -> Int32.t
32 bit methods w/ full precision

val unsafe_get_int32_t_be : t -> pos:int -> Int32.t
val unsafe_set_int32_t_le : t -> pos:int -> Int32.t -> unit
val unsafe_set_int32_t_be : t -> pos:int -> Int32.t -> unit
val unsafe_get_int64_t_le : t -> pos:int -> Int64.t
64 bit methods w/ full precision

val unsafe_get_int64_t_be : t -> pos:int -> Int64.t
val unsafe_set_int64_t_le : t -> pos:int -> Int64.t -> unit
val unsafe_set_int64_t_be : t -> pos:int -> Int64.t -> unit
val get_tail_padded_fixed_string : padding:char -> t -> pos:int -> len:int -> unit -> string
similar to Binary_packing.unpack_tail_padded_fixed_string and .pack_tail_padded_fixed_string.
val set_tail_padded_fixed_string : padding:char -> t -> pos:int -> len:int -> string -> unit
val compare : t -> t -> int
val t_of_sexp : Sexplib.Sexp.t -> t
val sexp_of_t : t -> Sexplib.Sexp.t
val bin_t : t Bin_prot.Type_class.t
val bin_read_t : t Bin_prot.Read.reader
val __bin_read_t__ : (int -> t) Bin_prot.Read.reader
val bin_reader_t : t Bin_prot.Type_class.reader
val bin_size_t : t Bin_prot.Size.sizer
val bin_write_t : t Bin_prot.Write.writer
val bin_writer_t : t Bin_prot.Type_class.writer

Creation and string conversion


create length

init n ~f creates a bigstring t of length n, with t.{i} = f i

of_string ?pos ?len str

to_string ?pos ?len bstr

Checking


check_args ~loc ~pos ~len bstr checks the position and length arguments pos and len for bigstrings bstr.

get_opt_len bstr ~pos opt_len

Accessors


length bstr

sub_shared ?pos ?len bstr

get t pos returns the character at pos

set t pos sets the character at pos

is_mmapped bstr

Blitting


blit ~src ?src_pos ?src_len ~dst ?dst_pos () blits src_len characters from src starting at position src_pos to dst at position dst_pos.

Reading/writing bin-prot


These functions write the "size-prefixed" bin-prot format that is used by, e.g., async's Writer.write_bin_prot, Reader.read_bin_prot and Unpack_buffer.Unpack_one.create_bin_prot.

write_bin_prot t writer a writes a to t starting at pos, and returns the index in t immediately after the last byte written. It raises if pos < 0 or if a doesn't fit in t.

default is 0

The read_bin_prot* functions read from the region of t starting at pos of length len. They return the index in t immediately after the last byte read. They raise if pos and len don't describe a region of t.

Memory mapping


map_file shared fd n memory-maps n characters of the data associated with descriptor fd to a bigstring. Iff shared is true, all changes to the bigstring will be reflected in the file.

Users must keep in mind that operations on the resulting bigstring may result in disk operations which block the runtime. This is true for pure OCaml operations (such as t.{1} <- 1), and for calls to blit. While some I/O operations may release the OCaml lock, users should not expect this to be done for all operations on a bigstring returned from map_file.



find ?pos ?len char t returns Some i for the smallest i >= pos such that t.{i} = char, or None if there is no such i.

Destruction


unsafe_destroy bstr destroys the bigstring by deallocating its associated data or, if memory-mapped, unmapping the corresponding file, and setting all dimensions to zero. This effectively frees the associated memory or address-space resources instantaneously. This feature helps working around a bug in the current OCaml runtime, which does not correctly estimate how aggressively to reclaim such resources.

This operation is safe unless you have passed the bigstring to another thread that is performing operations on it at the same time. Access to the bigstring after this operation will yield array bounds exceptions.

Accessors for parsing binary values, analogous to binary_packing. These are in Bigstring rather than a separate module because:

1) Existing binary_packing requires copies and does not work with bigstrings 2) The accessors rely on the implementation of bigstring, and hence should changeshould the implementation of bigstring move away from Bigarray. 3) Bigstring already has some external C functions, so it didn't require many changes to the OMakefile ^_^.

In a departure from Binary_packing, the naming conventions are chosen to be close to C99 stdint types, as it's a more standard description and it is somewhat useful in making compact macros for the implementations. The accessor names contain endian-ness to allow for branch-free implementations

<accessor> ::= <unsafe><operation><type><endian><int> <unsafe> ::= unsafe_ | '' <operation> ::= get_ | set_ <type> ::= int16 | uint16 | int32 | int64 <endian> ::= _le | _be | '' <int> ::= _int | ''

The "unsafe_" prefix indicates that these functions do no bounds checking. Performance testing demonstrated that the bounds check was 2-3 times slower due to the fact that Bigstring.length is a C call, and not even a noalloc one. In practice, message parsers can check the size of an outer message once, and use the unsafe accessors for individual fields, so many bounds checks can end up being redundant as well. The situation could be improved by having bigarray cache the length/dimensions.

16 bit methods


32 bit methods


Similar to the usage in binary_packing, the below methods are treating the value being read (or written), as an ocaml immediate integer, as such it is actually 63 bits. If the user is confident that the range of values used in practice will not require 64 bit precision (i.e. Less than Max_Long), then we can avoid allocation and use an immediate. If the user is wrong, an exception will be thrown (for get).

32 bit methods w/ full precision


64 bit methods w/ full precision


similar to Binary_packing.unpack_tail_padded_fixed_string and .pack_tail_padded_fixed_string.