1
(* $Id: xdr_mstring.mli 1558 2011-03-04 17:15:46Z gerd $ *)
5
(** Managed strings are used in XDR context for constant strings that
6
are stored either as string or as memory (bigarray of char).
8
A managed string [ms] is declared in the XDR file as in
11
typedef _managed string ms<>;
14
In the encoded XDR stream there is no difference between strings and
15
managed strings, i.e. the wire representation is identical. Only
16
the Ocaml type differs to which the managed string is mapped. This
17
type is {!Xdr_mstring.mstring} (below).
19
In the RPC context there is often the problem that the I/O backend
20
would profit from a different string representation than the user of
21
the RPC layer. To bridge this gap, managed strings have been invented.
22
Generally, the user can determine how to represent strings (usually
23
either as an Ocaml string, or as memory), and the I/O backend
24
can request to transform to a different representation when this
25
leads to an improvement (i.e. copy operations can be saved).
27
Only large managed strings result in a speedup of the program
30
{2 How to practically use managed strings}
32
There are two cases: The encoding case, and the decoding case.
33
In the encoding case the [mstring] object is created by the user
34
and passed to the RPC library. This happens when a client prepares
35
an argument for calling a remote procedure, or when the server
36
sends a response back to the caller. In the decoding case the client
37
analyzes the response from an RPC call, or the server looks at the
38
arguments of an RPC invocation. The difference here is that in the
39
encoding case user code can directly create [mstring] objects by
40
calling functions of this module, whereas in the decoding case the
41
RPC library creates the [mstring] objects.
43
For simplicity, let us only look at this problem from the perspective
46
{b Encoding.} Image a client wants to call an RPC, and one of the
47
arguments is a managed string. This means we finally need an [mstring]
48
object that can be put into the argument list of the call.
50
This library supports two string representation specially: The normal
51
Ocaml [string] type, and {!Netsys_mem.memory} which is actually just
52
a bigarray of char's. There are two factories [fac],
54
- {!Xdr_mstring.string_based_mstrings}, and
55
- {!Xdr_mstring.memory_based_mstrings},
57
and both can be used to create the [mstring] to pass to the
58
RPC layer. It should be noted that this layer can process the
59
[memory] representation a bit better. So, if the original [data]
60
value is a string, the factory for [string] should be used, and
61
if it is a char bigarray, the factory for [memory] should be used.
62
Now, the [mstring] object is created by
64
- [let mstring = fac # create_from_string data pos len copy_flag], or by
65
- [let mstring = fac # create_from_memory data pos len copy_flag].
67
Of course, if [fac] is the factory for strings, the [create_from_string]
68
method works better, and if [fac] is for [memory], the [create_from_memory]
69
method works better. [pos] and [len] can select a substring of [data].
70
If [copy_flag] is [false], the [mstring] object does not copy the data
71
if possible, but just keeps a reference to [data] until it is accessed;
72
otherwise if [copy_flag] is [true], a copy is made immediately.
73
Of couse, delaying the copy is better, but this requires that [data]
74
is not modified until the RPC call is completed.
76
{b Decoding.} Now, the call is done, and the client looks at the
77
result. There is also an [mstring] object in the result. As noted
78
above, this [mstring] object was already created by the RPC library
79
(and currently this library prefers string-based objects if not
80
told otherwise). The user code can now access this [mstring]
81
object with the access methods of the [mstring] class (see below).
82
As these methods are quite limited, it makes normally only sense
83
to output the [mstring] contents to a file descriptor.
85
The user can request a different factory for managed strings. The
86
function {!Rpc_client.set_mstring_factories} can be used for this
87
purpose. (Similar ways exist for managed clients, and for RPC servers.)
89
{b Potential.} Before introducing managed strings, a clean analysis
90
was done how many copy operations can be avoided by using this
91
technique. Example: The first N bytes of a file are taken as
92
argument of an RPC call. Instead of reading these bytes into a
93
normal Ocaml string, an optimal implementation uses now a [memory]
94
buffer for this purpose. This gives:
96
- Old implementation with strings and ocamlnet-2:
97
Data is copied {b six} times from reading it from the file until
98
writing it to the socket.
99
- New implementation with memory-based mstrings:
100
Data is copied only {b twice}! The first copy reads it from the
101
file into the input buffer (a [memory] value), and the second copy
102
writes the data into the socket.
104
Part of the optimization is that [Unix.read] and [Unix.write]
105
do a completely avoidable copy of the data which is prevented by
106
switching to {!Netsys_mem.mem_read} and {!Netsys_mem.mem_write},
107
respectively. The latter two functions exploit an optimization
108
that is only possible when the data is [memory]-typed.
110
The possible optimizations for the decoding side of the problem
111
are slightly less impressive, but still worth doing it.
118
(** The object holding the string value *)
122
(** The length of the managed string *)
124
method blit_to_string : int -> string -> int -> int -> unit
125
(** [blit_to_string mpos s spos len]: Copies the substring of the
126
managed string from [mpos] to [mpos+len-1] to the substring of
127
[s] from [spos] to [spos+len-1]
130
method blit_to_memory : int -> memory -> int -> int -> unit
131
(** [blit_to_string mpos mem mempos len]: Copies the substring of the
132
managed string from [mpos] to [mpos+len-1] to the substring of
133
[mem] from [mempos] to [mempos+len-1]
136
method as_string : string * int
137
(** Returns the contents as string. It is undefined whether the returned
138
string is a copy or the underlying buffer. The int is the position
139
where the contents start
142
method as_memory : memory * int
143
(** Returns the contents as memory. It is undefined whether the returned
144
memory is a copy or the underlying buffer. The int is the position
145
where the contents start
148
method preferred : [ `Memory | `String ]
149
(** Whether [as_memory] or [as_string] is cheaper *)
154
(** The object creating new [mstring] objects *)
155
class type mstring_factory =
157
method create_from_string : string -> int -> int -> bool -> mstring
158
(** [create_from_string s pos len must_copy]: Creates the [mstring] from the
159
sub string of s starting at [pos] with length [len]
161
If [must_copy] the mstring object must create a copy. Otherwise
162
it can just keep the string passed in.
165
method create_from_memory : memory -> int -> int -> bool -> mstring
166
(** [create_from_memory m pos len must_copy]: Creates the [mstring] from the
167
sub string of m starting at [pos] with length [len]
169
If [must_copy] the mstring object must create a copy. Otherwise
170
it can just keep the memory passed in.
175
val string_based_mstrings : mstring_factory
176
(** Uses strings to represent mstrings *)
178
val string_to_mstring : ?pos:int -> ?len:int -> string -> mstring
179
(** Represent a string as mstring (no copy) *)
181
val memory_based_mstrings : mstring_factory
182
(** Uses memory to represent mstrings. The memory bigarrays are allocated
183
with [Bigarray.Array1.create]
186
val memory_to_mstring : ?pos:int -> ?len:int -> memory -> mstring
187
(** Represent memory as mstring (no copy) *)
189
val paligned_memory_based_mstrings : mstring_factory
190
(** Uses memory to represent mstrings. The memory bigarrays are allocated
191
with {!Netsys_mem.alloc_memory_pages} if available, and
192
[Bigarray.Array1.create] if not.
195
val memory_pool_based_mstrings : Netsys_mem.memory_pool -> mstring_factory
196
(** Uses memory to represent mstrings. The memory bigarrays are obtained
197
from the pool. The length of these mstrings is limited by the
198
blocksize of the pool.
201
val length_mstrings : mstring list -> int
202
(** returns the sum of the lengths of the mstrings *)
204
val concat_mstrings : mstring list -> string
205
(** concatenates the mstrings and return them as single string. The returned
206
string may be shared with one of the mstrings passed in.
209
val prefix_mstrings : mstring list -> int -> string
210
(** [prefix_mstrings l n]: returns the first [n] chars of the
211
concatenated mstrings [l] as single string
214
val blit_mstrings_to_memory : mstring list -> memory -> unit
215
(** blits the mstrings one after the other to the memory, so that
216
they appear there concatenated
219
val shared_sub_mstring : mstring -> int -> int -> mstring
220
(** [shared_sub_mstring ms pos len]: returns an mstring that includes
221
a substring of [ms], starting at [pos], and with [len] bytes.
222
The returned mstring shares the buffer with the original mstring [ms]
225
val shared_sub_mstrings : mstring list -> int -> int -> mstring list
226
(** Same for a list of mstrings *)
228
val copy_mstring : mstring -> mstring
231
val copy_mstrings : mstring list -> mstring list
235
type named_mstring_factories =
236
(string, mstring_factory) Hashtbl.t