OLE Client Specification


 * ''There are still many points unclear about this protocol. You can discuss them on the Talk page.

This page is intended as a full description of the Computer Concepts OLE protocol as seen from the client side. A programmer writing an application should be able to find here all the information required to let the application work as an OLE client. To let the application work as an OLE server, see the OLE Server Specification.

The client
A client application (such as a desktop publishing program, or a mail handler) may wish to edit data it is capable of loading and rendering (such as drawfiles or plain text files). Such an application could provide facilities to edit these files itself, or use an already resident editor by sharing the file with it.

It seems sensible and easier to choose the second option, in which case the client needs to ask a 'compliant' server to engage in a two-way data sharing session.

An OLE session
An OLE session should be initiated by a client application.

Checking the environment variable
The client should check to see if an OLEServer$Type_XXX variable exists for its file type. If it cannot find such a variable then there is no server known for that file type at the moment. The client may then choose to use the OLESupport module. It can simulate the response of a server for a chosen file type. If the client does not want to use the OLESupport module, then the OLE attempt has failed at this point. The client should inform the user and clean up any internal data.

Note that if the OLESupport module has been used earlier for a particular file type, then there will already be an environment variable, which points to the module. If that is so, the client will find this variable in the first step, and transparently use the module, using the normal message protocol.

It is advisable to check for this variable every time an OLE session is requested, and not, e.g. only once when the client application starts or when the first OLE session is started. During the runtime of the client the variable may well change, for example because a server was seen by the Filer.

Using the OLESupport module

 * ''See OLESuppDoc for the original document.

First the client must check if the module is loaded, by calling the OLE_Version SWI in the X variant: SWI XOLE_Version (&67B00) On entry - On exit - R0 = Major revision number R1 = Minor revision number R2 = Task handle (0 if task not running) Errors  - Overflow set if SWI is not known On normal exit (overflow clear) the module is already loaded. The server should proceed by calling the OLE_SimulateSession SWI as described below.

If an error is returned the module is not yet loaded. The client should then try to start it with: SWI XWimp_StartTask (&600DE) On entry - R0 = ^ ".OLESupport" On exit - R0 = Task handle of module task (0 if task quit) Errors  - Overflow set if command failed
 * Note: the command string in R0 was taken from OLESuppDoc. It should be replaced with the correct string to start a module task. Should probably be "System:Modules.OLESupport".

If an error is returned, the module could not be found and the OLE session will have failed.

Now that the module is running, the client can ask it to handle sessions for the desired file type by calling OLE_SimulateSession: SWI OLE_SimulateSession (&47B05) On entry - R0 = file type On exit - The module will create an environment variable of the form OLEServer$Type_XXX : -N OLESupport -R /Desktop_OLESupport and will be ready to receive and send all OLE messages relevant to this file type. In effect the client can communicate transparently with the module as if it were a real OLE server.

The next step for the client is to start from the top: read the environment variable, send a broadcast, etc. Care must be taken not to fall into an endless loop if the environment variable the client looks for is not found. It should only attempt to start the module at most once for each OLE session.

Start the session
Having found a server variable, the client should save its data to disc in a suitable temporary location (see Programming Conventions) and send an OpenSession message with format 0 as a User_Message_Recorded (18) broadcast using the unique name specified in the variable.

Message_OLEOpenSession (&80E21) ---   SWI Wimp_SendMessage (&400E7) On entry - R0 = User message recorded (18) - R1 = ^Block +0 = length of block +4 = not used on entry +8 = not used on entry +12 = your_ref (=0, this is an original message) +16 = message number (&80E21) +20 = the 16 byte unique name from the -N tag, padded with zero bytes +36 = window handle of display holding file +40 = x offset of data in window +44 = y offset of data in window +48 = format number format = 0 or 1 (edit file) +52 = Session number (24 bit number invented by the client) +56 = file type +60 = full pathname of the data file, zero terminated format = 2 (re-edit file) +52 = Session number (24 bit number invented by the client) format > 2 (reserved for future expansion) - R2 = 0 (broadcast) On exit - R0 = corrupted Message block is updated: R1+4 = task handle of sender (i.e. us, the client) R1+8 = my_ref (unique Wimp-generated word > 0)

If the server is already running it will respond with an acknowledge.

If there is no acknowledgement, the client will get its message back as a User_Message_Acknowledge (19) message. It should then try to start the server using the -R token and again broadcast a recorded OpenSession message, this time with format 1. If again the message is not acknowledged, the session failed. The client should inform the user, delete the data file it created and tidy up.

In the case that all is well, the server will have acknowledged the OpenSession message by sending an OLEOpenSessionAck message to inform the client a session is truely open.


 * Note: the following information on format 2 messages is dubious. See the talk page

If a client knows it already has a link to a server, it should not attempt to send a format 0 open session message. It can send a format 2 message which will inform a server that the user has tried to perform an OLE action on the same data a second time. This gives those applications which allow documents to be closed, but not lost from memory (eg ArtWorks) a chance to reopen an edit window on the data.

If a format 2 message comes back unacknowledged, the server has presumably died in the meantime. The client should start from scratch with a format 0 message.

The server replies
Message_OLEOpenSessionAck (&80E22) --   The server returns the same block as OpenSession but copies my_ref (+8) to your_ref (+12) and puts the OLEOpenSessionAck code (&80E22) in the message number field (+16). This message tells the client that a session was opened successfully and may expect OLEFileChanged messages. The sending of this message by the server has several effects:
 * Because it sets the your_ref field to the my_ref field of the OLEOpenSession message, it tells the WIMP that it has acknowledged that message and the client will not get that OLEOpenSession message bounced back to it.
 * A new message (OLEOpenSessionAck) is sent to the client and can be used by the client to mark the session as open.
 * It tells the client what the task handle of the server is. Though the client might have gotten a task handle from the Wimp_StartTask call it used to start the server, in the most common cases the server will already be running, so this is the first point at which the client is informed of the task handle of the server.

The data has changed
Whenever the server saves data back to file, it sends an OLEFileChanged User message to the client (using the task handle passed in OLEOpenSession). This message format is as follows

Message_OLEFileChanged (&80E1E) ---                +0 = length of block +4 = task handle of sender +8 = my ref +12 = 0               +16 = message number (&80E1E) +20 = format number format = 0 (saved to a different file) then +24 = Session number +28 = full pathname of data, zero terminated format = 1 (saved to the same file) then +24 = Session number (format used by OLESupport) format > 1 then +24... reserved for future extensions

Note that the server should not feel it owns the file and thus should not attempt to delete the file during emergencies. However, when the OLESupport module is used the editing program invoked by the module will have no such limitations and might delete the file.

If the server corrupts the file, the client should be capable of working out that the file format has been compromised when it receives an OLEFileChanged and reads back the file.

The server will send a format 1 message when it saved the data to the same file. If the data was saved to a different file, it will send a format 0 message, providing the new pathname at +28.

The client should now read the changed file from disk.

Closing the session
Whenever the server throws data away through user action (in particular when the editing window is closed), it sends a User message back to the client informing it that the session has been terminated.

Message_OLECloseSession (&80E23) +0 = length of block +4 = task handle of sender +8 = my ref +12 = 0               +16 = message number (&80E23) +20 = format number format = 0 then +24 = Session number (-1 means all sessions are closing) format > 0 reserved for future extensions

This message tells the client that the session has ended and the file was not modified any further. It should remove the file and drop this session from its list.

Note that not only can a server send this message to a client, the client can send this message to a server or broadcast it to all servers when sessions are being closed from the client's end. (i.e. the user is closing the application down or removing one of its documents).

When a document is removed from the client (e.g. a Draw picture is deleted from a DTP application document) and there was an OLE session for that document, the client should send this message to the server with the appropriate Session number filled in. The server can then abandon the edit of this document.

When a client is closing down, it should broadcast this message with the Session number set to -1. The servers that are handling its OLE editing sessions then know that sessions for this client can be abandoned.

Similarly, when a server closes down, it will also send a broadcast of this message with the Session number set to -1. Clients who have sessions open with this server (recognised by the server task handle) should drop those sessions.

Starting a session
When a client wants to start an OLE session, it should write the file to disk and check for the environment variable for the proper file type. If no environment variable (and thus no server) is found, it could try to use the OLESupport module, which can be made to create an appropriate variable and operate as a server. The client should make sure it can handle the messages as described in, i.e. have the appropriate message handlers installed. It should generate a Session number and remember it belongs to this particular session in its internal data. It should then broadcast an OLEOpenSession format 0 User_Message_Recorded (type 18) message.

The handler for the OLEOpenSession message will deal with any problems in case the server was not running. The other handlers will deal with the rest of the session.

Discarding a file
When the user takes an action in the client application that discards an object for which an OLE session is in progress, the client should send an OLECloseSession User message to the server. It should then clean up (remove the file, etc).

Quitting the client
If the client is closing down and it still has sessions open, it should either send each of the servers it knows about an OLECLoseSession User message with the Session number set to -1 ( = close all sessions), or broadcast such a message.

Actually, the client could simply broadcast the message, regardless of whether it has, or ever had, any sessions open.

OLEOpenSession
Message_OLEOpenSession (&80E21) ---     +0 = length of block +4 = task handle of the sender (a client) +8 = my_ref +12 = 0    +16 = message number (&80E21) +20 = 16 byte unique name padded with zeros +36 = window handle of display holding file +40 = x offset of data in window +44 = y offset of data in window +48 = format number format = 0 or 1 (edit file) +52 = Session number +56 = file type +60 = full pathname of data, zero terminated format = 2 (re-edit file) +52 = Session number format > 2 (reserved for future expansion)

The client receives one of these messages as a type 19 message (User_Message_Acknowledge) when the broadcast of the message by the client itself was not acknowledged. It should take further action to try to get the server to respond.

Note that this message may also arrive due to a broadcast by another client who is looking for a server, or even receive its own message back as part of the broadcast. In that case it will be a type 17 (User_Message) or, more likely, type 18 (User_Message_Recorded) message. These should be ignored by the client.

Format 0
If a format 0 message comes back, this means the server is not running. The client should try to start the server using the value in the -R field of the environment variable to Wimp_StartTask. If that fails (it returned an error or a task handle of 0) the client should consider the attempted OLE session failed, display an appropriate message to the user, and clean up (remove the temporary file, remove the session from its internal list, etc).

If the server was started successfully, the client should send a new OLEOpenSession User_Message_Recorded (18) message directly to the server, using the task handle returned by Wimp_StartTask, this time with format set to 1.
 * Note: experiments have shown that existing clients often broadcast this message. See Talk:OLE Client Specification

Format 1
If a format 1 message comes back, this means that we tried to start the server but it either died or does not acknowledge the message (i.e. it refuses to handle the file for some reason). The client should assume the session failed, inform the user with a suitable message, and clean up (remove the temporary file, remove the session from its internal list, etc).

Format 2
If a format 2 message comes back, the server apparently has died since the last contact. The client should clean up the old session and start a new OLE session from scratch, i.e. read the environment variable, send an OLEOpenSession message with format 0 and continue the same way as for the original session.

OLEOpenSessionAck
Message_OLEOpenSessionAck (&80E22) --     +0 = length of block +4 = task handle of the sender (the server) +8 = my_ref +12 = your_ref (copy of my_ref from the OLEOpenSession sent by the client) +16 = message number (&80E22) +20 = 16 byte unique name padded with zeros +36 = window handle of display holding file +40 = x offset of data in window +44 = y offset of data in window +48 = format number format = 0 or 1 (edit file) +52 = Session number +56 = file type +60 = full pathname of data, zero terminated format = 2 (re-edit file) +52 = Session number format > 2 (reserved for future expansion)

This message means we are in business. The server is working on the document and the client may expect OLEFileChanged messages for it.

It is a copy of the OLEOpenSession message the client sent to the server, with the your_ref field filled with the value of the my_ref of the client's original message, and the message code replaced with the code for OLEOpenSessionAck (&80E22).

The client should check the your_ref value to see if it matches the my_ref of an OLEOpenSession message it sent, and check the Session number. If there is no match, the message should be ignored.

If the message is recognised the client should make a note of the task handle of the sender for this Session number. It will need this task handle to recognise messages coming from this server and when sending a message to the server.

OLEFileChanged
Message_OLEFileChanged (&80E1E) ---     +0 = length of block +4 = task handle of sender (the server) +8 = my ref +12 = 0    +16 = message number (&80E1E) +20 = format number format = 0 (saved to a different file) then +24 = Session number +28 = full pathname of data, zero terminated format = 1 (saved to the same file) then +24 = Session number (format used by OLESupport) format > 1 then +24... reserved for future extensions

This is what OLE is all about. The client is informed that the file has been changed by the server.

The client must check if it knows the task handle of the server as given in the message, and has a session with the given Session number open with it. If not, it should ignore the message.

If there is a session, it can read the modified file (from the original path for format 1, the given path for format 0). The client may want to display the modified file in its own window.

Note that this message does not imply that the session is over. The server may issue more OLEFileChanged messages for this file.

OLECloseSession
Message_OLECloseSession (&80E23) +0 = length of block +4 = task handle of sender (a server or client) +8 = my ref +12 = 0    +16 = message number (&80E23) +20 = format number format = 0 then +24 = Session number (-1 means all sessions are closing) format > 0 reserved for future extensions

This message is sent by a server when it wants to close a session (probably because the user has finished editing the object) or when it is shutting down, in which case all its sessions should be closed (Session number = -1).

The client must check if it knows the task handle of the sender as a server it has open sessions associated with. If not, it should ignore the message.

For the given session number the client should remove the file on disk (if it still exists), and remove the session from its list of active sessions. If the session number was -1, it should do so for all open sessions which were handled by the server identified by the given task handle.

The server variable
Any application which provides its own file type and is capable of editing such files may set itself up to be an OLE server. To do so it needs to create a system variable, outlining the file type it can edit.

The syntax of this variable is as follows:

Variable name = OLEServer$Type_XXX Variable value = -N  -R  X         = 0..9 | A..F AlphaChar  = 0..9 | A..Z | a..z UniqueName = [1..16]* (i.e. one to sixteen AlphaChars) RunCommand = A * command string that will start the server

Spaces must be used as separators. The -N token should always be first and the -R token last. Both tokens must be present.

Typical examples are

OLEServer$Type_AFF : -N OLESupport -R /Desktop_OLESupport OLEServer$Type_FFF : -N StrongED -R /ADFS::HardDisk4.$.Apps.!StrongED

Tokens

 * -N 
 * This token specifies a unique name to identify the server in an OpenSession message. This message is broadcast so it is up to the server who recognises the name to respond. This string can be up to 16 characters long. When passed in messages it should be specified as a 16 byte string with all unused bytes zeroed.
 * Note the server name should be modelled on the application name such as 'OLESupport' used by the support module or 'StrongED' as used by the StrongED text editor.


 * -R 
 * This token allows a potential client to run the server. It should provide a string which can be passed directly to Wimp_StartTask. This should start with a run command and be followed by the location of the server. This location could be an expanded pathname or more usually a system variable. Note that the RunCommand can contain spaces, and for convenience -R should be the last token in the variable. Examples:

run  /

Session numbers & task handles
To provide context for OLE sessions a session number and task handle should be kept by the client and server for each session opened. Session numbers must be allocated by the client task in a way which makes them unique for the run time of the program. The client should also keep a copy of the server's task handle which it will receive via the OLEOpenSessionAck message. This way it can inform the server whenever it closes down and recognise messages sent by the server.

Note that the combination of client task handle and session number uniquely identifies a session.

A session number is limited to 24 bits (it is unknown why this is), so using pointers to internal data structures as session numbers is not practical. A recommended strategy for generating session numbers by a client is to simply start at 1 when the application starts and increment by one for each new session. Assuming the user starts a new OLE session every 5 seconds for 12 hours a day, seven days a week it will take more than five years for the counter to run out.