val invalid_arg : ('a, unit, string, 'b) format4 -> 'ainclude module type of struct include Bytes endset s n c modifies s in place, replacing the byte at index n with c.
create n returns a new byte sequence of length n. The sequence is uninitialized and contains arbitrary bytes.
make n c returns a new byte sequence of length n, filled with the byte c.
init n f returns a fresh byte sequence of length n, with character i initialized to the result of f i (in increasing index order).
Return a new byte sequence that contains the same bytes as the given string.
Return a new string that contains the same bytes as the given byte sequence.
sub s pos len returns a new byte sequence of length len, containing the subsequence of s that starts at position pos and has length len.
Same as sub but return a string instead of a byte sequence.
extend s left right returns a new byte sequence that contains the bytes of s, with left uninitialized bytes prepended and right uninitialized bytes appended to it. If left or right is negative, then bytes are removed (instead of appended) from the corresponding side of s.
fill s pos len c modifies s in place, replacing len characters with c, starting at pos.
blit_string src src_pos dst dst_pos len copies len bytes from string src, starting at index src_pos, to byte sequence dst, starting at index dst_pos.
concat sep sl concatenates the list of byte sequences sl, inserting the separator byte sequence sep between each, and returns the result as a new byte sequence.
cat s1 s2 concatenates s1 and s2 and returns the result as a new byte sequence.
iter f s applies function f in turn to all the bytes of s. It is equivalent to f (get s 0); f (get s 1); ...; f (get s
(length s - 1)); ().
Same as iter, but the function is applied to the index of the byte as first argument and the byte itself as second argument.
map f s applies function f in turn to all the bytes of s (in increasing index order) and stores the resulting bytes in a new sequence that is returned as the result.
mapi f s calls f with each character of s and its index (in increasing index order) and stores the resulting bytes in a new sequence that is returned as the result.
fold_left f x s computes f (... (f (f x (get s 0)) (get s 1)) ...) (get s (n-1)), where n is the length of s.
fold_right f s x computes f (get s 0) (f (get s 1) ( ... (f (get s (n-1)) x) ...)), where n is the length of s.
for_all p s checks if all characters in s satisfy the predicate p.
exists p s checks if at least one character of s satisfies the predicate p.
Return a copy of the argument, without leading and trailing whitespace. The bytes regarded as whitespace are the ASCII characters ' ', '\012', '\n', '\r', and '\t'.
Return a copy of the argument, with special characters represented by escape sequences, following the lexical conventions of OCaml. All characters outside the ASCII printable range (32..126) are escaped, as well as backslash and double-quote.
index_opt s c returns the index of the first occurrence of byte c in s or None if c does not occur in s.
rindex s c returns the index of the last occurrence of byte c in s.
rindex_opt s c returns the index of the last occurrence of byte c in s or None if c does not occur in s.
index_from s i c returns the index of the first occurrence of byte c in s after position i. index s c is equivalent to index_from s 0 c.
index_from_opt s i c returns the index of the first occurrence of byte c in s after position i or None if c does not occur in s after position i. index_opt s c is equivalent to index_from_opt s 0 c.
rindex_from s i c returns the index of the last occurrence of byte c in s before position i+1. rindex s c is equivalent to rindex_from s (length s - 1) c.
rindex_from_opt s i c returns the index of the last occurrence of byte c in s before position i+1 or None if c does not occur in s before position i+1. rindex_opt s c is equivalent to rindex_from s (length s - 1) c.
contains_from s start c tests if byte c appears in s after position start. contains s c is equivalent to contains_from
s 0 c.
rcontains_from s stop c tests if byte c appears in s before position stop+1.
Return a copy of the argument, with all lowercase letters translated to uppercase, using the US-ASCII character set.
Return a copy of the argument, with all uppercase letters translated to lowercase, using the US-ASCII character set.
Return a copy of the argument, with the first character set to uppercase, using the US-ASCII character set.
Return a copy of the argument, with the first character set to lowercase, using the US-ASCII character set.
The comparison function for byte sequences, with the same specification as Stdlib.compare. Along with the type t, this function compare allows the module Bytes to be passed as argument to the functors Set.Make and Map.Make.
starts_with ~prefix s is true if and only if s starts with prefix.
ends_with ~suffix s is true if and only if s ends with suffix.
This section describes unsafe, low-level conversion functions between bytes and string. They do not copy the internal data; used improperly, they can break the immutability invariant on strings provided by the -safe-string option. They are available for expert library authors, but for most purposes you should use the always-correct to_string and of_string instead.
Unsafely convert a byte sequence into a string.
To reason about the use of unsafe_to_string, it is convenient to consider an "ownership" discipline. A piece of code that manipulates some data "owns" it; there are several disjoint ownership modes, including:
Unique ownership is linear: passing the data to another piece of code means giving up ownership (we cannot write the data again). A unique owner may decide to make the data shared (giving up mutation rights on it), but shared data may not become uniquely-owned again.
unsafe_to_string s can only be used when the caller owns the byte sequence s -- either uniquely or as shared immutable data. The caller gives up ownership of s, and gains ownership of the returned string.
There are two valid use-cases that respect this ownership discipline:
1. Creating a string by initializing and mutating a byte sequence that is never changed after initialization is performed.
let string_init len f : string =
let s = Bytes.create len in
for i = 0 to len - 1 do Bytes.set s i (f i) done;
Bytes.unsafe_to_string sThis function is safe because the byte sequence s will never be accessed or mutated after unsafe_to_string is called. The string_init code gives up ownership of s, and returns the ownership of the resulting string to its caller.
Note that it would be unsafe if s was passed as an additional parameter to the function f as it could escape this way and be mutated in the future -- string_init would give up ownership of s to pass it to f, and could not call unsafe_to_string safely.
We have provided the String.init, String.map and String.mapi functions to cover most cases of building new strings. You should prefer those over to_string or unsafe_to_string whenever applicable.
2. Temporarily giving ownership of a byte sequence to a function that expects a uniquely owned string and returns ownership back, so that we can mutate the sequence again after the call ended.
let bytes_length (s : bytes) =
String.length (Bytes.unsafe_to_string s)In this use-case, we do not promise that s will never be mutated after the call to bytes_length s. The String.length function temporarily borrows unique ownership of the byte sequence (and sees it as a string), but returns this ownership back to the caller, which may assume that s is still a valid byte sequence after the call. Note that this is only correct because we know that String.length does not capture its argument -- it could escape by a side-channel such as a memoization combinator.
The caller may not mutate s while the string is borrowed (it has temporarily given up ownership). This affects concurrent programs, but also higher-order functions: if String.length returned a closure to be called later, s should not be mutated until this closure is fully applied and returns ownership.
Unsafely convert a shared string to a byte sequence that should not be mutated.
The same ownership discipline that makes unsafe_to_string correct applies to unsafe_of_string: you may use it if you were the owner of the string value, and you will own the return bytes in the same mode.
In practice, unique ownership of string values is extremely difficult to reason about correctly. You should always assume strings are shared, never uniquely owned.
For example, string literals are implicitly shared by the compiler, so you never uniquely own them.
let incorrect = Bytes.unsafe_of_string "hello"
let s = Bytes.of_string "hello"The first declaration is incorrect, because the string literal "hello" could be shared by the compiler with other parts of the program, and mutating incorrect is a bug. You must always use the second version, which performs a copy and is thus correct.
Assuming unique ownership of strings that are not string literals, but are (partly) built from string literals, is also incorrect. For example, mutating unsafe_of_string ("foo" ^ s) could mutate the shared string "foo" -- assuming a rope-like representation of strings. More generally, functions operating on strings will assume shared ownership, they do not preserve unique ownership. It is thus incorrect to assume unique ownership of the result of unsafe_of_string.
The only case we have reasonable confidence is safe is if the produced bytes is shared -- used as an immutable byte sequence. This is possibly useful for incremental migration of low-level programs that manipulate immutable sequences of bytes (for example Marshal.from_bytes) and previously used the string type for this purpose.
split_on_char sep s returns the list of all (possibly empty) subsequences of s that are delimited by the sep character.
The function's output is specified by the following invariants:
sep as a separator returns a byte sequence equal to the input (Bytes.concat (Bytes.make 1 sep)
(Bytes.split_on_char sep s) = s).sep character.Iterate on the string, in increasing index order. Modifications of the string during iteration will be reflected in the sequence.
Iterate on the string, in increasing order, yielding indices along chars
val get_utf_8_uchar : t -> int -> Uchar.utf_decodeget_utf_8_uchar b i decodes an UTF-8 character at index i in b.
set_utf_8_uchar b i u UTF-8 encodes u at index i in b and returns the number of bytes n that were written starting at i. If n is 0 there was not enough space to encode u at i and b was left untouched. Otherwise a new character can be encoded at i + n.
val is_valid_utf_8 : t -> boolis_valid_utf_8 b is true if and only if b contains valid UTF-8 data.
val get_utf_16be_uchar : t -> int -> Uchar.utf_decodeget_utf_16be_uchar b i decodes an UTF-16BE character at index i in b.
set_utf_16be_uchar b i u UTF-16BE encodes u at index i in b and returns the number of bytes n that were written starting at i. If n is 0 there was not enough space to encode u at i and b was left untouched. Otherwise a new character can be encoded at i + n.
val is_valid_utf_16be : t -> boolis_valid_utf_16be b is true if and only if b contains valid UTF-16BE data.
val get_utf_16le_uchar : t -> int -> Uchar.utf_decodeget_utf_16le_uchar b i decodes an UTF-16LE character at index i in b.
set_utf_16le_uchar b i u UTF-16LE encodes u at index i in b and returns the number of bytes n that were written starting at i. If n is 0 there was not enough space to encode u at i and b was left untouched. Otherwise a new character can be encoded at i + n.
val is_valid_utf_16le : t -> boolis_valid_utf_16le b is true if and only if b contains valid UTF-16LE data.
The functions in this section binary encode and decode integers to and from byte sequences.
All following functions raise Invalid_argument if the space needed at index i to decode or encode the integer is not available.
Little-endian (resp. big-endian) encoding means that least (resp. most) significant bytes are stored first. Big-endian is also known as network byte order. Native-endian encoding is either little-endian or big-endian depending on Sys.big_endian.
32-bit and 64-bit integers are represented by the int32 and int64 types, which can be interpreted either as signed or unsigned numbers.
8-bit and 16-bit integers are represented by the int type, which has more bits than the binary encoding. These extra bits are handled as follows:
int values sign-extend (resp. zero-extend) their result.int values truncate their input to their least significant bytes.get_uint8 b i is b's unsigned 8-bit integer starting at byte index i.
get_int8 b i is b's signed 8-bit integer starting at byte index i.
get_uint16_ne b i is b's native-endian unsigned 16-bit integer starting at byte index i.
get_uint16_be b i is b's big-endian unsigned 16-bit integer starting at byte index i.
get_uint16_le b i is b's little-endian unsigned 16-bit integer starting at byte index i.
get_int16_ne b i is b's native-endian signed 16-bit integer starting at byte index i.
get_int16_be b i is b's big-endian signed 16-bit integer starting at byte index i.
get_int16_le b i is b's little-endian signed 16-bit integer starting at byte index i.
get_int32_ne b i is b's native-endian 32-bit integer starting at byte index i.
get_int32_be b i is b's big-endian 32-bit integer starting at byte index i.
get_int32_le b i is b's little-endian 32-bit integer starting at byte index i.
get_int64_ne b i is b's native-endian 64-bit integer starting at byte index i.
get_int64_be b i is b's big-endian 64-bit integer starting at byte index i.
get_int64_le b i is b's little-endian 64-bit integer starting at byte index i.
set_uint8 b i v sets b's unsigned 8-bit integer starting at byte index i to v.
set_int8 b i v sets b's signed 8-bit integer starting at byte index i to v.
set_uint16_ne b i v sets b's native-endian unsigned 16-bit integer starting at byte index i to v.
set_uint16_be b i v sets b's big-endian unsigned 16-bit integer starting at byte index i to v.
set_uint16_le b i v sets b's little-endian unsigned 16-bit integer starting at byte index i to v.
set_int16_ne b i v sets b's native-endian signed 16-bit integer starting at byte index i to v.
set_int16_be b i v sets b's big-endian signed 16-bit integer starting at byte index i to v.
set_int16_le b i v sets b's little-endian signed 16-bit integer starting at byte index i to v.
set_int32_ne b i v sets b's native-endian 32-bit integer starting at byte index i to v.
set_int32_be b i v sets b's big-endian 32-bit integer starting at byte index i to v.
set_int32_le b i v sets b's little-endian 32-bit integer starting at byte index i to v.
set_int64_ne b i v sets b's native-endian 64-bit integer starting at byte index i to v.
set_int64_be b i v sets b's big-endian 64-bit integer starting at byte index i to v.
set_int64_le b i v sets b's little-endian 64-bit integer starting at byte index i to v.
Care must be taken when concurrently accessing byte sequences from multiple domains: accessing a byte sequence will never crash a program, but unsynchronized accesses might yield surprising (non-sequentially-consistent) results.
Every byte sequence operation that accesses more than one byte is not atomic. This includes iteration and scanning.
For example, consider the following program:
let size = 100_000_000
let b = Bytes.make size ' '
let update b f () =
Bytes.iteri (fun i x -> Bytes.set b i (Char.chr (f (Char.code x)))) b
let d1 = Domain.spawn (update b (fun x -> x + 1))
let d2 = Domain.spawn (update b (fun x -> 2 * x + 1))
let () = Domain.join d1; Domain.join d2the bytes sequence b may contain a non-deterministic mixture of '!', 'A', 'B', and 'C' values.
After executing this code, each byte of the sequence b is either '!', 'A', 'B', or 'C'. If atomicity is required, then the user must implement their own synchronization (for example, using Mutex.t).
If two domains only access disjoint parts of a byte sequence, then the observed behaviour is the equivalent to some sequential interleaving of the operations from the two domains.
A data race is said to occur when two domains access the same byte without synchronization and at least one of the accesses is a write. In the absence of data races, the observed behaviour is equivalent to some sequential interleaving of the operations from different domains.
Whenever possible, data races should be avoided by using synchronization to mediate the accesses to the elements of the sequence.
Indeed, in the presence of data races, programs will not crash but the observed behaviour may not be equivalent to any sequential interleaving of operations from different domains. Nevertheless, even in the presence of data races, a read operation will return the value of some prior write to that location.
Another subtle point is that if a data race involves mixed-size writes and reads to the same location, the order in which those writes and reads are observed by domains is not specified. For instance, the following code write sequentially a 32-bit integer and a char to the same index
let b = Bytes.make 10 '\000'
let d1 = Domain.spawn (fun () -> Bytes.set_int32_ne b 0 100; b.[0] <- 'd' )In this situation, a domain that observes the write of 'd' to b.0 is not guaranteed to also observe the write to indices 1, 2, or 3.
val get_unsafe_16 : t -> int -> intval get_unsafe_32 : t -> int -> int32val get_unsafe_64 : t -> int -> int64val get_16 : t -> int -> intval get_32 : t -> int -> int32val get_64 : t -> int -> int64val set_unsafe_16 : t -> int -> int -> unitval set_unsafe_32 : t -> int -> int32 -> unitval set_unsafe_64 : t -> int -> int64 -> unitval set_16 : t -> int -> int -> unitval set_32 : t -> int -> int32 -> unitval set_64 : t -> int -> int64 -> unitval pp : Format.formatter -> bytes -> unit