Plan 9 from Bell Labs’s /usr/web/sources/contrib/fgb/root/sys/src/cmd/tcl/doc/Notifier.3

Copyright © 2021 Plan 9 Foundation.
Distributed under the MIT License.
Download the Plan 9 distribution.


'\"
'\" Copyright (c) 1998-1999 Scriptics Corporation
'\" Copyright (c) 1995-1997 Sun Microsystems, Inc.
'\"
'\" See the file "license.terms" for information on usage and redistribution
'\" of this file, and for a DISCLAIMER OF ALL WARRANTIES.
'\" 
'\" RCS: @(#) $Id: Notifier.3,v 1.21 2007/12/13 15:22:31 dgp Exp $
'\" 
.so man.macros
.TH Notifier 3 8.1 Tcl "Tcl Library Procedures"
.BS
.SH NAME
Tcl_CreateEventSource, Tcl_DeleteEventSource, Tcl_SetMaxBlockTime, Tcl_QueueEvent, Tcl_ThreadQueueEvent, Tcl_ThreadAlert, Tcl_GetCurrentThread, Tcl_DeleteEvents, Tcl_InitNotifier, Tcl_FinalizeNotifier, Tcl_WaitForEvent, Tcl_AlertNotifier, Tcl_SetTimer, Tcl_ServiceAll, Tcl_ServiceEvent, Tcl_GetServiceMode, Tcl_SetServiceMode \- the event queue and notifier interfaces
.SH SYNOPSIS
.nf
\fB#include <tcl.h>\fR
.sp
void
\fBTcl_CreateEventSource\fR(\fIsetupProc, checkProc, clientData\fR)
.sp
void
\fBTcl_DeleteEventSource\fR(\fIsetupProc, checkProc, clientData\fR)
.sp
void
\fBTcl_SetMaxBlockTime\fR(\fItimePtr\fR)
.sp
void
\fBTcl_QueueEvent\fR(\fIevPtr, position\fR)
.sp
void
\fBTcl_ThreadQueueEvent\fR(\fIthreadId, evPtr, position\fR)
.sp
void
\fBTcl_ThreadAlert\fR(\fIthreadId\fR)
.sp
Tcl_ThreadId
\fBTcl_GetCurrentThread\fR()
.sp
void
\fBTcl_DeleteEvents\fR(\fIdeleteProc, clientData\fR)
.sp
ClientData
\fBTcl_InitNotifier\fR()
.sp
void
\fBTcl_FinalizeNotifier\fR(\fIclientData\fR)
.sp
int
\fBTcl_WaitForEvent\fR(\fItimePtr\fR)
.sp
void
\fBTcl_AlertNotifier\fR(\fIclientData\fR)
.sp
void
\fBTcl_SetTimer\fR(\fItimePtr\fR)
.sp
int
\fBTcl_ServiceAll\fR()
.sp
int
\fBTcl_ServiceEvent\fR(\fIflags\fR)
.sp
int
\fBTcl_GetServiceMode\fR()
.sp
int
\fBTcl_SetServiceMode\fR(\fImode\fR)
.sp
void
\fBTcl_ServiceModeHook\fR(\fImode\fR)
.sp
void
\fBTcl_SetNotifier\fR(\fInotifierProcPtr\fR)
.SH ARGUMENTS
.AS Tcl_EventDeleteProc *notifierProcPtr
.AP Tcl_EventSetupProc *setupProc in
Procedure to invoke to prepare for event wait in \fBTcl_DoOneEvent\fR.
.AP Tcl_EventCheckProc *checkProc in
Procedure for \fBTcl_DoOneEvent\fR to invoke after waiting for
events.  Checks to see if any events have occurred and, if so,
queues them.
.AP ClientData clientData in
Arbitrary one-word value to pass to \fIsetupProc\fR, \fIcheckProc\fR, or
\fIdeleteProc\fR.
.AP Tcl_Time *timePtr in
Indicates the maximum amount of time to wait for an event.  This
is specified as an interval (how long to wait), not an absolute
time (when to wakeup).  If the pointer passed to \fBTcl_WaitForEvent\fR
is NULL, it means there is no maximum wait time:  wait forever if
necessary.
.AP Tcl_Event *evPtr in
An event to add to the event queue.  The storage for the event must
have been allocated by the caller using \fBTcl_Alloc\fR or \fBckalloc\fR.
.AP Tcl_QueuePosition position in
Where to add the new event in the queue:  \fBTCL_QUEUE_TAIL\fR,
\fBTCL_QUEUE_HEAD\fR, or \fBTCL_QUEUE_MARK\fR.
.AP Tcl_ThreadId threadId in
A unique identifier for a thread.
.AP Tcl_EventDeleteProc *deleteProc in
Procedure to invoke for each queued event in \fBTcl_DeleteEvents\fR.
.AP int flags in
What types of events to service.  These flags are the same as those
passed to \fBTcl_DoOneEvent\fR.
.AP int mode in
Indicates whether events should be serviced by \fBTcl_ServiceAll\fR.
Must be one of \fBTCL_SERVICE_NONE\fR or \fBTCL_SERVICE_ALL\fR.
.AP Tcl_NotifierProcs* notifierProcPtr in
Structure of function pointers describing notifier procedures that are
to replace the ones installed in the executable.  See
\fBREPLACING THE NOTIFIER\fR for details.
.BE

.SH INTRODUCTION
.PP
The interfaces described here are used to customize the Tcl event
loop.  The two most common customizations are to add new sources of
events and to merge Tcl's event loop with some other event loop, such
as one provided by an application in which Tcl is embedded.  Each of
these tasks is described in a separate section below.
.PP
The procedures in this manual entry are the building blocks out of which
the Tcl event notifier is constructed.  The event notifier is the lowest
layer in the Tcl event mechanism.  It consists of three things:
.IP [1]
Event sources: these represent the ways in which events can be
generated.  For example, there is a timer event source that implements
the \fBTcl_CreateTimerHandler\fR procedure and the \fBafter\fR
command, and there is a file event source that implements the
\fBTcl_CreateFileHandler\fR procedure on Unix systems.  An event
source must work with the notifier to detect events at the right
times, record them on the event queue, and eventually notify
higher-level software that they have occurred.  The procedures
\fBTcl_CreateEventSource\fR, \fBTcl_DeleteEventSource\fR,
and \fBTcl_SetMaxBlockTime\fR, \fBTcl_QueueEvent\fR, and
\fBTcl_DeleteEvents\fR are used primarily by event sources.
.IP [2]
The event queue: for non-threaded applications,
there is a single queue for the whole application,
containing events that have been detected but not yet serviced.  Event
sources place events onto the queue so that they may be processed in
order at appropriate times during the event loop. The event queue
guarantees a fair discipline of event handling, so that no event
source can starve the others.  It also allows events to be saved for
servicing at a future time.  Threaded applications work in a
similar manner, except that there is a separate event queue for
each thread containing a Tcl interpreter.
\fBTcl_QueueEvent\fR is used (primarily
by event sources) to add events to the event queue and 
\fBTcl_DeleteEvents\fR is used to remove events from the queue without
processing them.  In a threaded application, \fBTcl_QueueEvent\fR adds
an event to the current thread's queue, and \fBTcl_ThreadQueueEvent\fR
adds an event to a queue in a specific thread.
.IP [3]
The event loop: in order to detect and process events, the application
enters a loop that waits for events to occur, places them on the event
queue, and then processes them.  Most applications will do this by
calling the procedure \fBTcl_DoOneEvent\fR, which is described in a
separate manual entry.
.PP
Most Tcl applications need not worry about any of the internals of
the Tcl notifier.  However, the notifier now has enough flexibility
to be retargeted either for a new platform or to use an external event
loop (such as the Motif event loop, when Tcl is embedded in a Motif
application).  The procedures \fBTcl_WaitForEvent\fR and
\fBTcl_SetTimer\fR are normally implemented by Tcl, but may be
replaced with new versions to retarget the notifier (the
\fBTcl_InitNotifier\fR, \fBTcl_AlertNotifier\fR,
\fBTcl_FinalizeNotifier\fR, \fBTcl_Sleep\fR,
\fBTcl_CreateFileHandler\fR, and \fBTcl_DeleteFileHandler\fR must
also be replaced; see CREATING A NEW NOTIFIER below for details).
The procedures \fBTcl_ServiceAll\fR, \fBTcl_ServiceEvent\fR,
\fBTcl_GetServiceMode\fR, and \fBTcl_SetServiceMode\fR are provided
to help connect Tcl's event loop to an external event loop such as
Motif's.
.SH "NOTIFIER BASICS"
.PP
The easiest way to understand how the notifier works is to consider
what happens when \fBTcl_DoOneEvent\fR is called.
\fBTcl_DoOneEvent\fR is passed a \fIflags\fR argument that indicates
what sort of events it is OK to process and also whether or not to
block if no events are ready.  \fBTcl_DoOneEvent\fR does the following
things:
.IP [1]
Check the event queue to see if it contains any events that can
be serviced.  If so, service the first possible event, remove it
from the queue, and return.  It does this by calling
\fBTcl_ServiceEvent\fR and passing in the \fIflags\fR argument.
.IP [2]
Prepare to block for an event.  To do this, \fBTcl_DoOneEvent\fR
invokes a \fIsetup procedure\fR in each event source.
The event source will perform event-source specific initialization and
possibly call \fBTcl_SetMaxBlockTime\fR to limit how long
\fBTcl_WaitForEvent\fR will block if no new events occur.
.IP [3]
Call \fBTcl_WaitForEvent\fR.  This procedure is implemented differently
on different platforms;  it waits for an event to occur, based on the
information provided by the event sources.
It may cause the application to block if \fItimePtr\fR specifies
an interval other than 0.
\fBTcl_WaitForEvent\fR returns when something has happened,
such as a file becoming readable or the interval given by \fItimePtr\fR
expiring.  If there are no events for \fBTcl_WaitForEvent\fR to
wait for, so that it would block forever, then it returns immediately
and \fBTcl_DoOneEvent\fR returns 0.
.IP [4]
Call a \fIcheck procedure\fR in each event source.  The check
procedure determines whether any events of interest to this source
occurred.  If so, the events are added to the event queue.
.IP [5]
Check the event queue to see if it contains any events that can
be serviced.  If so, service the first possible event, remove it
from the queue, and return.
.IP [6]
See if there are idle callbacks pending. If so, invoke all of them and
return.
.IP [7]
Either return 0 to indicate that no events were ready, or go back to
step [2] if blocking was requested by the caller.

.SH "CREATING A NEW EVENT SOURCE"
.PP
An event source consists of three procedures invoked by the notifier,
plus additional C procedures that are invoked by higher-level code
to arrange for event-driven callbacks.  The three procedures called
by the notifier consist of the setup and check procedures described
above, plus an additional procedure that is invoked when an event
is removed from the event queue for servicing.
.PP
The procedure \fBTcl_CreateEventSource\fR creates a new event source.
Its arguments specify the setup procedure and check procedure for
the event source.
\fISetupProc\fR should match the following prototype:
.CS
typedef void Tcl_EventSetupProc(
        ClientData \fIclientData\fR,
        int \fIflags\fR);
.CE
The \fIclientData\fR argument will be the same as the \fIclientData\fR
argument to \fBTcl_CreateEventSource\fR;  it is typically used to
point to private information managed by the event source.
The \fIflags\fR argument will be the same as the \fIflags\fR
argument passed to \fBTcl_DoOneEvent\fR except that it will never
be 0 (\fBTcl_DoOneEvent\fR replaces 0 with \fBTCL_ALL_EVENTS\fR).
\fIFlags\fR indicates what kinds of events should be considered;
if the bit corresponding to this event source is not set, the event
source should return immediately without doing anything.  For
example, the file event source checks for the \fBTCL_FILE_EVENTS\fR
bit.
.PP
\fISetupProc\fR's job is to make sure that the application wakes up
when events of the desired type occur.  This is typically done in a
platform-dependent fashion.  For example, under Unix an event source
might call \fBTcl_CreateFileHandler\fR; under Windows it might
request notification with a Windows event.  For timer-driven event
sources such as timer events or any polled event, the event source
can call \fBTcl_SetMaxBlockTime\fR to force the application to wake
up after a specified time even if no events have occurred.
If no event source calls \fBTcl_SetMaxBlockTime\fR
then \fBTcl_WaitForEvent\fR will wait as long as necessary for an
event to occur; otherwise, it will only wait as long as the shortest
interval passed to \fBTcl_SetMaxBlockTime\fR by one of the event
sources.  If an event source knows that it already has events ready to
report, it can request a zero maximum block time.  For example, the
setup procedure for the X event source looks to see if there are
events already queued.  If there are, it calls
\fBTcl_SetMaxBlockTime\fR with a 0 block time so that
\fBTcl_WaitForEvent\fR does not block if there is no new data on the X
connection.
The \fItimePtr\fR argument to \fBTcl_WaitForEvent\fR points to
a structure that describes a time interval in seconds and
microseconds:
.CS
typedef struct Tcl_Time {
        long \fIsec\fR;
        long \fIusec\fR;
} Tcl_Time;
.CE
The \fIusec\fR field should be less than 1000000.
.PP
Information provided to \fBTcl_SetMaxBlockTime\fR
is only used for the next call to \fBTcl_WaitForEvent\fR; it is
discarded after \fBTcl_WaitForEvent\fR returns.
The next time an event wait is done each of the event sources'
setup procedures will be called again, and they can specify new
information for that event wait.
.PP
If the application uses an external event loop rather than
\fBTcl_DoOneEvent\fR, the event sources may need to call
\fBTcl_SetMaxBlockTime\fR at other times.  For example, if a new event
handler is registered that needs to poll for events, the event source
may call \fBTcl_SetMaxBlockTime\fR to set the block time to zero to
force the external event loop to call Tcl.  In this case,
\fBTcl_SetMaxBlockTime\fR invokes \fBTcl_SetTimer\fR with the shortest
interval seen since the last call to \fBTcl_DoOneEvent\fR or
\fBTcl_ServiceAll\fR.
.PP
In addition to the generic procedure \fBTcl_SetMaxBlockTime\fR, other
platform-specific procedures may also be available for
\fIsetupProc\fR, if there is additional information needed by
\fBTcl_WaitForEvent\fR on that platform.  For example, on Unix systems
the \fBTcl_CreateFileHandler\fR interface can be used to wait for file events.
.PP
The second procedure provided by each event source is its check
procedure, indicated by the \fIcheckProc\fR argument to
\fBTcl_CreateEventSource\fR.  \fICheckProc\fR must match the
following prototype:
.CS
typedef void Tcl_EventCheckProc(
        ClientData \fIclientData\fR,
        int \fIflags\fR);
.CE
The arguments to this procedure are the same as those for \fIsetupProc\fR.
\fBCheckProc\fR is invoked by \fBTcl_DoOneEvent\fR after it has waited
for events.  Presumably at least one event source is now prepared to
queue an event.  \fBTcl_DoOneEvent\fR calls each of the event sources
in turn, so they all have a chance to queue any events that are ready.
The check procedure does two things.  First, it must see if any events
have triggered.  Different event sources do this in different ways.
.PP
If an event source's check procedure detects an interesting event, it
must add the event to Tcl's event queue.  To do this, the event source
calls \fBTcl_QueueEvent\fR.  The \fIevPtr\fR argument is a pointer to
a dynamically allocated structure containing the event (see below for
more information on memory management issues).  Each event source can
define its own event structure with whatever information is relevant
to that event source.  However, the first element of the structure
must be a structure of type \fBTcl_Event\fR, and the address of this
structure is used when communicating between the event source and the
rest of the notifier.  A \fBTcl_Event\fR has the following definition:
.CS
typedef struct {
    Tcl_EventProc *\fIproc\fR;
    struct Tcl_Event *\fInextPtr\fR;
} Tcl_Event;
.CE
The event source must fill in the \fIproc\fR field of
the event before calling \fBTcl_QueueEvent\fR.
The \fInextPtr\fR is used to link together the events in the queue
and should not be modified by the event source.
.PP
An event may be added to the queue at any of three positions, depending
on the \fIposition\fR argument to \fBTcl_QueueEvent\fR:
.IP \fBTCL_QUEUE_TAIL\fR 24
Add the event at the back of the queue, so that all other pending
events will be serviced first.  This is almost always the right
place for new events.
.IP \fBTCL_QUEUE_HEAD\fR 24
Add the event at the front of the queue, so that it will be serviced
before all other queued events.
.IP \fBTCL_QUEUE_MARK\fR 24
Add the event at the front of the queue, unless there are other
events at the front whose position is \fBTCL_QUEUE_MARK\fR;  if so,
add the new event just after all other \fBTCL_QUEUE_MARK\fR events.
This value of \fIposition\fR is used to insert an ordered sequence of
events at the front of the queue, such as a series of
Enter and Leave events synthesized during a grab or ungrab operation
in Tk.
.PP
When it is time to handle an event from the queue (steps 1 and 4
above) \fBTcl_ServiceEvent\fR will invoke the \fIproc\fR specified
in the first queued \fBTcl_Event\fR structure.
\fIProc\fR must match the following prototype:
.CS
typedef int Tcl_EventProc(
        Tcl_Event *\fIevPtr\fR,
        int \fIflags\fR);
.CE
The first argument to \fIproc\fR is a pointer to the event, which will
be the same as the first argument to the \fBTcl_QueueEvent\fR call that
added the event to the queue.
The second argument to \fIproc\fR is the \fIflags\fR argument for the
current call to \fBTcl_ServiceEvent\fR;  this is used by the event source
to return immediately if its events are not relevant.
.PP
It is up to \fIproc\fR to handle the event, typically by invoking
one or more Tcl commands or C-level callbacks.
Once the event source has finished handling the event it returns 1
to indicate that the event can be removed from the queue.
If for some reason the event source decides that the event cannot
be handled at this time, it may return 0 to indicate that the event
should be deferred for processing later;  in this case \fBTcl_ServiceEvent\fR
will go on to the next event in the queue and attempt to service it.
There are several reasons why an event source might defer an event.
One possibility is that events of this type are excluded by the
\fIflags\fR argument.
For example, the file event source will always return 0 if the
\fBTCL_FILE_EVENTS\fR bit is not set in \fIflags\fR.
Another example of deferring events happens in Tk if
\fBTk_RestrictEvents\fR has been invoked to defer certain kinds
of window events.
.PP
When \fIproc\fR returns 1, \fBTcl_ServiceEvent\fR will remove the
event from the event queue and free its storage.
Note that the storage for an event must be allocated by
the event source (using \fBTcl_Alloc\fR or the Tcl macro \fBckalloc\fR)
before calling \fBTcl_QueueEvent\fR, but it
will be freed by \fBTcl_ServiceEvent\fR, not by the event source.
.PP
Threaded applications work in a
similar manner, except that there is a separate event queue for
each thread containing a Tcl interpreter.
Calling \fBTcl_QueueEvent\fR in a multithreaded application adds
an event to the current thread's queue.
To add an event to another thread's queue, use \fBTcl_ThreadQueueEvent\fR.
\fBTcl_ThreadQueueEvent\fR accepts as an argument a Tcl_ThreadId argument,
which uniquely identifies a thread in a Tcl application.  To obtain the
Tcl_ThreadID for the current thread, use the \fBTcl_GetCurrentThread\fR
procedure.  (A thread would then need to pass this identifier to other
threads for those threads to be able to add events to its queue.)
After adding an event to another thread's queue, you then typically
need to call \fBTcl_ThreadAlert\fR to
.QW "wake up"
that thread's notifier to alert it to the new event.
.PP
\fBTcl_DeleteEvents\fR can be used to explicitly remove one or more
events from the event queue.  \fBTcl_DeleteEvents\fR calls \fIproc\fR
for each event in the queue, deleting those for with the procedure
returns 1.  Events for which the procedure returns 0 are left in the
queue.  \fIProc\fR should match the following prototype:
.CS
typedef int Tcl_EventDeleteProc(
        Tcl_Event *\fIevPtr\fR,
        ClientData \fIclientData\fR);
.CE
The \fIclientData\fR argument will be the same as the \fIclientData\fR
argument to \fBTcl_DeleteEvents\fR; it is typically used to point to
private information managed by the event source.  The \fIevPtr\fR will
point to the next event in the queue.
.PP
\fBTcl_DeleteEventSource\fR deletes an event source.  The \fIsetupProc\fR,
\fIcheckProc\fR, and \fIclientData\fR arguments must exactly match those
provided to the \fBTcl_CreateEventSource\fR for the event source to be deleted.
If no such source exists, \fBTcl_DeleteEventSource\fR has no effect.

.SH "CREATING A NEW NOTIFIER"
.PP
The notifier consists of all the procedures described in this manual
entry, plus \fBTcl_DoOneEvent\fR and \fBTcl_Sleep\fR, which are
available on all platforms, and \fBTcl_CreateFileHandler\fR and
\fBTcl_DeleteFileHandler\fR, which are Unix-specific.  Most of these
procedures are generic, in that they are the same for all notifiers.
However, none of the procedures are notifier-dependent:
\fBTcl_InitNotifier\fR, \fBTcl_AlertNotifier\fR,
\fBTcl_FinalizeNotifier\fR, \fBTcl_SetTimer\fR, \fBTcl_Sleep\fR,
\fBTcl_WaitForEvent\fR, \fBTcl_CreateFileHandler\fR,
\fBTcl_DeleteFileHandler\fR and \fBTcl_ServiceModeHook\fR.  To support a
new platform or to integrate Tcl with an application-specific event loop,
you must write new versions of these procedures.
.PP
\fBTcl_InitNotifier\fR initializes the notifier state and returns
a handle to the notifier state.  Tcl calls this
procedure when initializing a Tcl interpreter.  Similarly,
\fBTcl_FinalizeNotifier\fR shuts down the notifier, and is
called by \fBTcl_Finalize\fR when shutting down a Tcl interpreter.
.PP
\fBTcl_WaitForEvent\fR is the lowest-level procedure in the notifier;
it is responsible for waiting for an
.QW interesting
event to occur or
for a given time to elapse.  Before \fBTcl_WaitForEvent\fR is invoked,
each of the event sources' setup procedure will have been invoked.
The \fItimePtr\fR argument to
\fBTcl_WaitForEvent\fR gives the maximum time to block for an event,
based on calls to \fBTcl_SetMaxBlockTime\fR made by setup procedures
and on other information (such as the \fBTCL_DONT_WAIT\fR bit in
\fIflags\fR).
.PP
Ideally, \fBTcl_WaitForEvent\fR should only wait for an event
to occur; it should not actually process the event in any way.
Later on, the
event sources will process the raw events and create Tcl_Events on
the event queue in their \fIcheckProc\fR procedures.
However, on some platforms (such as Windows) this is not possible;
events may be processed in \fBTcl_WaitForEvent\fR, including queuing
Tcl_Events and more (for example, callbacks for native widgets may be
invoked).  The return value from \fBTcl_WaitForEvent\fR must be either
0, 1, or \-1.  On platforms such as Windows where events get processed in
\fBTcl_WaitForEvent\fR, a return value of 1 means that there may be more
events still pending that have not been processed.  This is a sign to the
caller that it must call \fBTcl_WaitForEvent\fR again if it wants all
pending events to be processed. A 0 return value means that calling
\fBTcl_WaitForEvent\fR again will not have any effect: either this is a
platform where \fBTcl_WaitForEvent\fR only waits without doing any event
processing, or \fBTcl_WaitForEvent\fR knows for sure that there are no
additional events to process (e.g. it returned because the time
elapsed).  Finally, a return value of \-1 means that the event loop is
no longer operational and the application should probably unwind and
terminate.  Under Windows this happens when a WM_QUIT message is received;
under Unix it happens when \fBTcl_WaitForEvent\fR would have waited
forever because there were no active event sources and the timeout was
infinite.
.PP
\fBTcl_AlertNotifier\fR is used in multithreaded applications to allow
any thread to
.QW "wake up"
the notifier to alert it to new events on its
queue.  \fBTcl_AlertNotifier\fR requires as an argument the notifier
handle returned by \fBTcl_InitNotifier\fR.
.PP
If the notifier will be used with an external event loop, then it must
also support the \fBTcl_SetTimer\fR interface.  \fBTcl_SetTimer\fR is
invoked by \fBTcl_SetMaxBlockTime\fR whenever the maximum blocking
time has been reduced.  \fBTcl_SetTimer\fR should arrange for the
external event loop to invoke \fBTcl_ServiceAll\fR after the specified
interval even if no events have occurred.  This interface is needed
because \fBTcl_WaitForEvent\fR is not invoked when there is an external
event loop.  If the
notifier will only be used from \fBTcl_DoOneEvent\fR, then
\fBTcl_SetTimer\fR need not do anything.
.PP
\fBTcl_ServiceModeHook\fR is called by the platform-independent portion
of the notifier when client code makes a call to
\fBTcl_SetServiceMode\fR. This hook is provided to support operating
systems that require special event handling when the application is in
a modal loop (the Windows notifier, for instance, uses this hook to
create a communication window).
.PP
On Unix systems, the file event source also needs support from the
notifier.  The file event source consists of the
\fBTcl_CreateFileHandler\fR and \fBTcl_DeleteFileHandler\fR
procedures, which are described in the \fBTcl_CreateFileHandler\fR
manual page.
.PP
The \fBTcl_Sleep\fR and \fBTcl_DoOneEvent\fR interfaces are described
in their respective manual pages.
.PP
The easiest way to create a new notifier is to look at the code
for an existing notifier, such as the files \fBunix/tclUnixNotfy.c\fR
or \fBwin/tclWinNotify.c\fR in the Tcl source distribution.

.SH "REPLACING THE NOTIFIER"
.PP
A notifier that has been written according to the conventions above
can also be installed in a running process in place of the standard
notifier.  This mechanism is used so that a single executable can be
used (with the standard notifier) as a stand-alone program and reused
(with a replacement notifier in a loadable extension) as an extension
to another program, such as a Web browser plugin.
.PP
To do this, the extension makes a call to \fBTcl_SetNotifier\fR
passing a pointer to a \fBTcl_NotifierProcs\fR data structure.  The
structure has the following layout:
.CS
typedef struct Tcl_NotifierProcs {
    Tcl_SetTimerProc *setTimerProc;
    Tcl_WaitForEventProc *waitForEventProc;
    Tcl_CreateFileHandlerProc *createFileHandlerProc;
    Tcl_DeleteFileHandlerProc *deleteFileHandlerProc;
    Tcl_InitNotifierProc *initNotifierProc;
    Tcl_FinalizeNotifierProc *finalizeNotifierProc;
    Tcl_AlertNotifierProc *alertNotifierProc;
    Tcl_ServiceModeHookProc *serviceModeHookProc;
} Tcl_NotifierProcs;
.CE
Following the call to \fBTcl_SetNotifier\fR, the pointers given in
the \fBTcl_NotifierProcs\fR structure replace whatever notifier had
been installed in the process.
.PP
It is extraordinarily unwise to replace a running notifier. Normally,
\fBTcl_SetNotifier\fR should be called at process initialization time
before the first call to \fBTcl_InitNotifier\fR.

.SH "EXTERNAL EVENT LOOPS"
.PP
The notifier interfaces are designed so that Tcl can be embedded into
applications that have their own private event loops.  In this case,
the application does not call \fBTcl_DoOneEvent\fR except in the case
of recursive event loops such as calls to the Tcl commands \fBupdate\fR
or \fBvwait\fR.  Most of the time is spent in the external event loop
of the application.  In this case the notifier must arrange for the
external event loop to call back into Tcl when something
happens on the various Tcl event sources.  These callbacks should
arrange for appropriate Tcl events to be placed on the Tcl event queue.
.PP
Because the external event loop is not calling \fBTcl_DoOneEvent\fR on
a regular basis, it is up to the notifier to arrange for
\fBTcl_ServiceEvent\fR to be called whenever events are pending on the
Tcl event queue.  The easiest way to do this is to invoke
\fBTcl_ServiceAll\fR at the end of each callback from the external
event loop.  This will ensure that all of the event sources are
polled, any queued events are serviced, and any pending idle handlers
are processed before returning control to the application.  In
addition, event sources that need to poll for events can call
\fBTcl_SetMaxBlockTime\fR to force the external event loop to call
Tcl even if no events are available on the system event queue.
.PP
As a side effect of processing events detected in the main external
event loop, Tcl may invoke \fBTcl_DoOneEvent\fR to start a recursive event
loop in commands like \fBvwait\fR.  \fBTcl_DoOneEvent\fR will invoke
the external event loop, which will result in callbacks as described
in the preceding paragraph, which will result in calls to
\fBTcl_ServiceAll\fR.  However, in these cases it is undesirable to
service events in \fBTcl_ServiceAll\fR.  Servicing events there is
unnecessary because control will immediately return to the
external event loop and hence to \fBTcl_DoOneEvent\fR, which can
service the events itself.  Furthermore, \fBTcl_DoOneEvent\fR is
supposed to service only a single event, whereas \fBTcl_ServiceAll\fR
normally services all pending events.  To handle this situation,
\fBTcl_DoOneEvent\fR sets a flag for \fBTcl_ServiceAll\fR
that causes it to return without servicing any events.
This flag is called the \fIservice mode\fR;
\fBTcl_DoOneEvent\fR restores it to its previous value before it returns.
.PP
In some cases, however, it may be necessary for \fBTcl_ServiceAll\fR
to service events
even when it has been invoked from \fBTcl_DoOneEvent\fR.  This happens
when there is yet another recursive event loop invoked via an
event handler called by \fBTcl_DoOneEvent\fR (such as one that is
part of a native widget).  In this case, \fBTcl_DoOneEvent\fR may not
have a chance to service events so \fBTcl_ServiceAll\fR must service
them all.  Any recursive event loop that calls an external event
loop rather than \fBTcl_DoOneEvent\fR must reset the service mode so
that all events get processed in \fBTcl_ServiceAll\fR.  This is done
by invoking the \fBTcl_SetServiceMode\fR procedure.  If
\fBTcl_SetServiceMode\fR is passed \fBTCL_SERVICE_NONE\fR, then calls
to \fBTcl_ServiceAll\fR will return immediately without processing any
events.  If \fBTcl_SetServiceMode\fR is passed \fBTCL_SERVICE_ALL\fR,
then calls to \fBTcl_ServiceAll\fR will behave normally.
\fBTcl_SetServiceMode\fR returns the previous value of the service
mode, which should be restored when the recursive loop exits.
\fBTcl_GetServiceMode\fR returns the current value of the service
mode.

.SH "SEE ALSO"
\fBTcl_CreateFileHandler\fR, \fBTcl_DeleteFileHandler\fR, \fBTcl_Sleep\fR,
\fBTcl_DoOneEvent\fR, \fBThread(3)\fR
.SH KEYWORDS
event, notifier, event queue, event sources, file events, timer, idle, service mode, threads

Bell Labs OSI certified Powered by Plan 9

(Return to Plan 9 Home Page)

Copyright © 2021 Plan 9 Foundation. All Rights Reserved.
Comments to webmaster@9p.io.