Skip to content


Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
tag: v948
Fetching contributors…

Cannot retrieve contributors at this time

565 lines (404 sloc) 22.592 kb
This is a program which takes as input a file containing a (limited) set of
OCaml type definitions plus some annotations and generates code in OCaml and OPA
to transmit these values between networked instances. In theory, the encoded
types should be compatible with the current JSON types and could be used to send
data to javascript endpoints but this has not been tested.
The mlidl program
Recognised types
This takes as input the following subset of OCaml types, defined in pseudo-ML
open MLIDL_Module
external external_ty : Mod.external_type = "<complex set of implementation functions>"
type ty =
| int
| float
| bool
| string
| ty option
| ty list
| (ty * ty * ... )
| { lab1:ty; lab2:ty; ... }
| (Cons1 of ty | Cons2 of ty | Cons3)
| MLIDL_Module.ty
| external_ty
Note that there are no:
- Type variables: 'a ty
- References: ty ref
- Functional types: ty -> ty -> ...
- Type labels: ~label:ty
Although some support is planned for a simplified form of type variables which
will require a complex set of definitions as used by the external type syntax.
You can import types from other IDL files with the syntax indicated above. You
have to "open" the name of the generated module by the included IDL file and
then prefix *all* of the types you import from other files.
There is also provision for importing externally-defined types from OCaml
modules but the syntax for this is highly complex. Simply put, you have to do
all the work done for you by the IDL code generator and indicate to the IDL
function names which are the equivalent of its own generated functions for these
types. This is a lot of work so you should maybe consider using the IDL program
to generate these types and then use the import mechanism if you have access to
the sources of these types. One point, however, is that this mechanism allows
you to define a completely separate type representation between the OCaml and
OPA types. See the "External Types" section below for more details.
Basic functionality generated
The types are translated more-or-less literally into OCaml types on the ML side
and equivalent types on the OPA side. The OCaml output code includes an
interface file. The following functions are generated:
String input/output
Firstly, for OCaml this functionality is wrapped in functors (MakeInput and
MakeOutput) where the input and output functions are abstract. However, for
convenience, an instantiation of these modules is provided generating the
- <type>_of_string
- string_of_<type>
for each <type> in the IDL file. These are the functions used by the
network layer code. If you wish to output to Buffer.t or whatever then
instantiate the functors.
For OPA the output functions are called:
- output_<type>
and are wrapped in a module called STR_<name>, where <name> is the base name
of the IDL file. There is currently no input_<type> functionality for
native OPA (due to the slowness of OPA parsers).
For BSL output, the <type>_of_string and string_of_<type> functions are made
available to OPA where the types remain external OCaml types. Note,
however, that the wrap_<type> and unwrap_<type> functions used by these
routines are also made available to OPA.
Generation of these functions can be prevented with the option:
--string-functions false
JSON input/output
These are functions to convert the types into JsonTypes types in OCaml and
to RPC.Json.json types for OPA:
- tojson_<type>
- fromjson_<type>
These functions are available for both OCaml and Native OPA. The OPA
functions are included in a module called "JSON_<name>".
Currently, the BSL generated code does not include JSON conversion.
There used to be support for low-level JSON types in OPA (OCaml JsonTypes as
external types in the OPA file) but this has fallen out of date since the
Json types were updated. It could be resurrected on request.
You can individually defeat the tojson_<type> and fromjson_<type> functions
--tojson-functions false
--fromjson-functions false
Create types
There is primitive support for generating types from constant types in the
IDL file: type date = { year : int; month : int; day : int; }
MLI file: val create_date : int -> int -> int -> date
Application: let date = create_date 1961 8 24;;
These can be switched off by:
--create-functions false
Network layer code generated
The IDL code generator also understands a small set of annotations (disguised as
OCaml "val" statements) which indicate that it should generate code for the
network layer communications which use the basic conversion functions:
val sendreceive_profile : profile -> profile
val send_date : hare
val receive_gender : gender
val protocol_cat : cat -> string
val responder_dog : dog -> string
Don't be fooled by the ML-like syntax of these annotations, there are no such
values actually generated by the IDL code generator.
The most important of these is the "responder" annotation which generates all
the code needed to implement a server/client combination between two (Hlnet)
endpoints. The other routines provide low-level network operations which may be
useful where the responder pattern is not what is required. A brief summary of
these operations:
- "val sendreceive_<name> : <type1> -> <type2>"
This is an Hlnet.sendreceive call where the types are encoded according to
the encoding defined in the IDL file (currently there is only one encoding).
- "val send_<name> : <type>"
Generates an Hlnet.send call with the encoded type.
- "val receive_<name> : <type>"
Generates an Hlnet.receive call with the encoded type.
- "val protocol_<name> : <type1> -> <type2>"
Generates an Hlnet.Aux.easy_protocol value with query type <type1> and
response type <type2>. Note that you can set the name and version number for
the protocol by annotations in the IDL file. Note also that if you set
either type to "string" then no type encoding is applied to that type.
- "val responder_<name> : <type1> -> <type2>"
This simple statement generates all the apparatus needed to implement a
server/client pair using the indicated types. See the "Responders" section
below for details.
Command line and in-built arguments
The output of the IDL file can be controlled by either command-line arguments or
with annotations built into the IDL files themselves.
One important point is that the in-built parameters take precedence. If you
wish to control the output from the command line, the in-built argument
equivalents must not be present. Here are the command line arguments:
-output-suffix <string> Output suffix (default: "types").
The generated files (ml, mli, opa, bsl) will all have this suffix appended
to the base name of the IDL file (eg. test.mlidl ->
-bsl-prefix <string> Bsl prefix (default: "bsl").
This prefix will be prepended to the base name of generated BSL files. This
is to prevent overwriting the ml files of the same name. Note that if your
code is not destined for the OPA BSL mechanism, a more suitable prefix might
be "plugin".
-ocaml-wrap-options <bool> Wrap option around OCaml input (default: "true").
-opa-wrap-options <bool> Wrap option around Opa input (default: "true").
Redundant, do not use.
-native-parser <bool> Use native parser instead of TRX wrappers (default: "true").
Redundant, do not use.
-hlnet-logging <bool> Add logger statements to Hlnet wrappers (default: "false").
Network-layer code can optionally generate debug logging message, this flag
switches this functionality on and off.
-logger-function <string> Logger function (default: "Logger.log").
The logger function to use for the --hlnet-logging feature.
-protocol-version <int> Protocol version number (default: 1).
The protocol number to install in any protocols generated for the network
-default-port <int> Default port number (default: 49152).
The default port to be built into the network layer endpoint description.
-default-addr <string> Default inet address (default: "Unix.inet_addr_loopback").
The default address for the endpoint.
-create-functions <bool> Output create value from type functions (default: true).
Add the create functions to the output.
-tojson-functions <bool> Output type to json functions (default: true).
-fromjson-functions <bool> Output type from json functions (default: true).
Enable/disable JSON output/input functions.
-string-functions <bool> Output type to/from string functions (default: true).
Enable/disable to/from string functions.
-bsl-file <bool> Output BSL file for OCaml functions (default: true).
Generate BSL wrapper code. Note: you need -no-ocaml to be false since only
the wrappers are generated, not the referenced OCaml code.
-no-ocaml <bool> Don't generate OCaml output (default: false).
Do not generate OCaml output.
-no-opa <bool> Don't generate OPA output (default: false).
Do not generate native OPA output.
The in-built option have the following syntax:
let module_name = "test2"
let protocol_version = 2
let verbose = true
The value must be of the correct type. Mostly these are exactly the same as the
command-line arguments:
In-built argument Type Command-line option
----------------- ------ -------------------
module_name String <none>
output_suffix String --output-suffix
bsl_prefix String --bsl-prefix
encoding_number Int --encoding-number
ocaml_wrap_opt Bool --ocaml-wrap-opt
opa_wrap_opt Bool --opa-wrap-opt
native_parser Bool --native-parser
hlnet_logging Bool --hlnet-logging
logger_function String --logger-function
protocol_version Int --protocol-version
default_port Int --default-port
default_addr String --default-addr
create_functions Bool --create-functions
tojson_functions Bool --tojson-functions
fromjson_functions Bool --fromjson-functions
string_functions Bool --string-functions
bsl_file Bool --bsl-file
no_ocaml Bool --no-ocaml
no_opa Bool --no-opa
verbose Bool --v
debug Bool --g
The "module_name" argument does not appear on the command line since the name of
the output module will apply to each IDL file, it would not make sense to set
the same module name for all included files as well.
External Types
This is a difficult feature to use but is potentially rewarding in that, once
you have generated all the necessary support code by hand, you can include
external types in your IDL files with the same status as IDL types.
To define an external type, use:
external ip : Ip.ip = "[(<name>,<value>);...]"
This defines an IDL type called "ip" with OCaml type Ip.ip. The string
definition is an assoc list of name-value pairs which define the names of
functions to be used as replacement functions for those not generated by the IDL
code generator. Note that the type of the OPA-side value is also defined in the
assoc list. Here is the current list of names:
Name Type of value
------------- -------------
ocamlstringof <type> -> string
ocamlofstring string -> <type>
ocamltojson <type> -> JsonTypes.json
ocamlfromjson JsonTypes.json -> <type>
bslwrap <type> -> <opatype>
bslunwrap <opatype> -> <type>
opatype name of the OPA type
opastringof <opatype> -> string
opaofstring string -> <opatype>
opatojson <opatype> -> RPC.Json.json
opafromjson RPC.Json.json -> <opatype>
opatojsonll <opatype> -> (external) JsonTypes.json
opafromjsonll (external) JsonTypes.json -> <opatype>
1) The text provided by the <value> string is technically OCaml code but note
that you can't currently use complex code here because the string is parsed
very simply with String.slice, so you can't have any semicolons or commas
in your ML code! The above table indicates the types of the resulting
OCaml text.
2) The <opatype> type need in no way correspond to the <type> type. This
allows matching of unrelated types (the Ip.ip type is a good example of
3) For an example of the implementation of this code, see the and
ipopa.opa files in libnet/tests. (Note: don't call your .ml and .opa files
by the same name!).
4) It is unlikely that you will need all of the above values for any
particular application. The IDL code generator will inform you if you are
missing any of the functions for a particular output code.
Compiling and running the IDL code generator
The IDL code generator executable is called mlidl.native and is installed by
build_tools into _build/protocols/mlidl.native.
The command itself simply takes the name of a single IDL file (the extension
should be .mlidl) but note that it may start reading included IDL files. These
should already have been processed by the IDL code generator before processing
the outer IDL file.
Compiling the output is somewhat problematical since the dependencies in the
generated code are quite complex. Note that you need to include any file
dependencies on the command line, particularly, the ML and BSL files for any
included types plus have the libraries for any external types. Hopefully, this
should be handled by opalang/bld but this is untested.
This is currently the main pay-off for the IDL code generator. From a very
simple description file it is possible to generate code which reduces the
complex business of managing the transfer of typed data from one network
endpoint to another. The IDL annotation:
val responder_kind : kind -> string
generates the following values in the OCaml output file:
val protocol_kind : (kind,string) Hlnet.protocol
val entrypoint_protocol_kind : (unit option, (kind,string) Hlnet.protocol
val port_kind : int ref
val addr_kind : Unix.inet_addr ref
val endpoint_kind : Hlnet.endpoint ref
val scheduler_kind : Scheduler.t ref
val init_responder_kind : int -> Unix.inet_addr -> Scheduler.t -> unit
These allow the specification of endpoints for the resulting network layer. The
port number and network address are actually used to define the endpoint. They
are included separately so they can be read back more easily. Use the
init_responder_kind function to set all the endpoint values in one go, for
init_responder_kind 12345 Unix.inet_addr_loopback Scheduler.default
The endpoint and protocol values can then be used by your own communications
code but the IDL also generates server and client code which uses these values:
val respond_server_kind :
('a * 'b) -> (('a * 'b) -> kind -> 'a * 'b * string option * bool) -> unit
val respond_client_kind :
'a -> ('a -> ('a -> kind option * ('a -> string -> ('a -> bool -> unit) -> unit) option -> unit) -> unit)
-> ('a -> unit) -> unit
These are a generic server/client pair and the rather complex types need some
explanation. For the server, the actual call to the server looks like:
respond_server_kind (server_data, connection_data) responder
This initialises a server. The user-supplied data is divided up into two
portions, server_data which persists while the server is active and
connection_data which is reinitialised to the value given here at each new
The "responder" function is a callback function which is called by the server
when a value of the given type is received, for example:
let responder (count,ud_conn) (msg:K.kind) =
match msg with
| K.Low str -> (count, ud_conn, Some (String.uppercase str), true)
| K.Up str -> (count, ud_conn, Some (String.lowercase str), true)
| K.Kill -> (sleep 2 @> fun _ -> ()); (count+1, ud_conn, None, count < 2)
This function is passed the user data (server,connection) plus the type received
from the client. It is expected to return a quadruple:
(<updated server data>,
<updated connection data>,
<response type>,
<continue flag>)
Most of these are self-explanatory, the response type is the data sent back to
the client. In this case the return type was just a string so there is no type
conversion done in this case. The continue flag should be "true" if the server
is to continue. If "false" then the server terminates.
The semantics of the client is more complex, the client call looks like:
respond_client_kind userdata client_cont termination_cont
Here, the user data is a single value. The client continuation parameter allows
control over the connection:
let client_cont (list_message,_) k =
match list_message with
(*data*) (*msg*) (*handler*)
| [] -> k ([],true) ((Some K.Kill), None) (* Send message and close connection *)
| (msg::t) -> k (t,false) ((Some msg), (Some client_handler)) (* Send-receive message *)
(*| <whatever> -> k userdata (None, (Some client_handler)) (* Receive message *)*)
(*| <whatever> -> k userdata (None, None) (* Close connection immediately *)*)
This is a continuation function and is passed the user data and the following
continuation ("k"). Based on the user data, the client continuation can decide
to do various things, the following continuation is passed the updated user data
(in this case a list of messages and a status code) plus a pair of an optional
query type message for the server and a continuation handler for replies from
the server. Based on these values the client code will:
Message Handler Action
------- ------- ------
Some msg Some handler call Hlnet.sendreceive with the message and handler
Some msg None call Hlnet.send but does not wait for a reply
None Some handler do not send anything but wait with Hlnet.receive
None None terminate the connection
The receive handler has the following semantics:
let client_handler ((_,last) as userdata) str k =
k userdata (not last) (* true=>continue, false=>close connection *)
It is passed the user data, the message received from the server and a
continuation function. It should pass the continuation function the updated
user data plus a continuation flag, "false" means terminate the connection.
Although quite complex, this scheme allows fine control over the server and
client by the application, without having to deal with low-level network issues
although server/client coordination is still the responsibility of the
application code.
The abstract client model allows the implementation of the following auxiliary
client functions:
val respond_client_single_kind : kind -> (string option -> unit) -> unit
val respond_client_send_kind : kind -> (unit -> unit) -> unit
val respond_client_receive_kind : (string option -> unit) -> unit
These are simple one-shot client operations:
respond_client_single_kind: make a connection to the server, send a single
message and handle the reply with the handler function before terminating the
respond_client_send_kind: connect, send message and close connection.
respond_client_receive_kind: connect, handle received message and close
Note that the values received by the handlers are always options. A "None"
value means an error in receiving the value.
Note also that there are significant differences between the BSL implementation
of these functions and the OCaml version. This was because I couldn't get OCaml
to call the OPA continuation functions without segfaults. The OPA interface is
therefore only "pseudo-cps" and works by callback functions instead. This may
change if I can ever get OCaml to call an OPA continuation function. See the
multi_clients_bsl.opa file for an example.
Test code and examples
The standard test script
In libnet/tests there is a script "" which allows testing of various
features and is intended to be run as a reftester test one day. This has
various options for testing specific phases and includes two applications
mentioned below.
Note that this test script attempts to compile using the bare ocaml command line
and does not work via opalang/bld. You need a successfully installed
opalang/mkinstall for this to work.
This is a bare-bones test which defines one type for each of the types currently
handled by the IDL code generator. Note, however, that there are actually a
couple of special cases which look like duplicated types but are actually there
to test the peculiarities of the OCaml type system (eg. Cons (int * bool) is
handled slightly strangely by OCaml). The phase flags are absent from this file
to allow the test script to control the phases from the command-line.
This is a trivial IDL file but it is used in the multi_clients_bsl.opa and applications which are IDL implementations of an old Hlnet
test program. They use the responder mechanism and illustrate the differences
between the two implementations.
N. Scaife
Jump to Line
Something went wrong with that request. Please try again.