NAME

thread - Extension for script access to Tcl threading

SYNOPSIS

package require Tcl 8.4

package require Thread ?2.8?

thread::create ?-joinable? ?-preserved? ?script?

thread::preserve ?id?

thread::release ?-wait? ?id?

thread::id

thread::errorproc ?procname?

thread::cancel ?-unwind? id ?result?

thread::unwind

thread::exit ?status?

thread::names

thread::exists id

thread::send ?-async? ?-head? id script ?varname?

thread::broadcast script

thread::wait

thread::eval ?-lock mutex? arg ?arg ...?

thread::join id

thread::configure id ?option? ?value? ?...?

thread::transfer id channel

thread::detach channel

thread::attach channel

thread::mutex

thread::mutex create ?-recursive?

thread::mutex destroy mutex

thread::mutex lock mutex

thread::mutex unlock mutex

thread::rwmutex

thread::rwmutex create

thread::rwmutex destroy mutex

thread::rwmutex rlock mutex

thread::rwmutex wlock mutex

thread::rwmutex unlock mutex

thread::cond

thread::cond create

thread::cond destroy cond

thread::cond notify cond

thread::cond wait cond mutex ?ms?

DESCRIPTION

The thread extension creates threads that contain Tcl interpreters, and it lets you send scripts to those threads for evaluation. Additionally, it provides script-level access to basic thread synchronization primitives, like mutexes and condition variables.

COMMANDS

This section describes commands for creating and destroying threads and sending scripts to threads for evaluation.

thread::create ?-joinable? ?-preserved? ?script?

This command creates a thread that contains a Tcl interpreter. The Tcl interpreter either evaluates the optional script, if specified, or it waits in the event loop for scripts that arrive via the thread::send command. Both of them would take place simultaneously with the return of command thread::create to the caller thread. Neither the caller is waiting for the finishing of optional script, nor the result, if any, of the script is returned to the caller. The result of thread::create is the ID of the thread. This is the opaque handle which identifies the newly created thread for all other package commands. The handle of the thread goes out of scope automatically when thread is marked for exit (see the thread::release command below).

If the optional script argument contains the thread::wait command the thread will enter into the event loop. If such command is not found in the script the thread will run the script to the end and exit. In that case, the handle may be safely ignored since it refers to a thread which does not exists any more at the time when the command returns.

Using flag -joinable it is possible to create a joinable thread, i.e. one upon whose exit can be waited upon by using thread::join command. Note that failure to join a thread created with -joinable flag results in resource and memory leaks.

Threads created by the thread::create cannot be destroyed forcefully. Consequently, there is no corresponding thread destroy command. A thread may only be released using the thread::release and if its internal reference count drops to zero, the thread is marked for exit. This kicks the thread out of the event loop servicing and the thread continues to execute commands passed in the script argument, following the thread::wait command. If this was the last command in the script, as usually the case, the thread will exit.

It is possible to create a situation in which it may be impossible to terminate the thread, for example by putting some endless loop after the thread::wait or entering the event loop again by doing an vwait-type of command. In such cases, the thread may never exit. This is considered to be a bad practice and should be avoided if possible. This is best illustrated by the example below:

    # You should never do ...
    set tid [thread::create {
        package require Http
        thread::wait
        vwait forever ; # <-- this!
    }]

: The thread created in the above example will never be able to exit. After it has been released with the last matching thread::release call, the thread will jump out of the thread::wait and continue to execute commands following. It will enter vwait command and wait endlessly for events. There is no way one can terminate such thread, so you wouldn't want to do this!
Each newly created has its internal reference counter set to 0 (zero), i.e. it is unreserved. This counter gets incremented by a call to thread::preserve and decremented by a call to thread::release command. These two commands implement simple but effective thread reservation system and offer predictable and controllable thread termination capabilities. It is however possible to create initially preserved threads by using flag -preserved of the thread::create command. Threads created with this flag have the initial value of the reference counter of 1 (one), and are thus initially marked reserved.

thread::preserve ?id?: This command increments the thread reference counter. Each call to this command increments the reference counter by one (1). Command returns the value of the reference counter after the increment. If called with the optional thread id, the command preserves the given thread. Otherwise the current thread is preserved.
With reference counting, one can implement controlled access to a shared Tcl thread. By incrementing the reference counter, the caller signalizes that he/she wishes to use the thread for a longer period of time. By decrementing the counter, caller signalizes that he/she has finished using the thread.
thread::release ?-wait? ?id?: This command decrements the thread reference counter. Each call to this command decrements the reference counter by one (1). If called with the optional thread id, the command releases the given thread. Otherwise, the current thread is released. Command returns the value of the reference counter after the decrement. When the reference counter reaches zero (0), the target thread is marked for termination. You should not reference the thread after the thread::release command returns zero or negative integer. The handle of the thread goes out of scope and should not be used any more. Any following reference to the same thread handle will result in Tcl error.
Optional flag -wait instructs the caller thread to wait for the target thread to exit, if the effect of the command would result in termination of the target thread, i.e. if the return result would be zero (0). Without the flag, the caller thread does not wait for the target thread to exit. Care must be taken when using the -wait, since this may block the caller thread indefinitely. This option has been implemented for some special uses of the extension and is deprecated for regular use. Regular users should create joinable threads by using the -joinable option of the thread::create command and the thread::join to wait for thread to exit.
thread::id: This command returns the ID of the current thread.
thread::errorproc ?procname?: This command sets a handler for errors that occur in scripts sent asynchronously, using the -async flag of the thread::send command, to other threads. If no handler is specified, the current handler is returned. The empty string resets the handler to default (unspecified) value. An uncaught error in a thread causes an error message to be sent to the standard error channel. This default reporting scheme can be changed by registering a procedure which is called to report the error. The procname is called in the interpreter that invoked the thread::errorproc command. The procname is called like this:

    myerrorproc thread_id errorInfo

thread::cancel ?-unwind? id ?result?: This command requires Tcl version 8.6 or higher.
Cancels the script being evaluated in the thread given by the id parameter. Without the -unwind switch the evaluation stack for the interpreter is unwound until an enclosing catch command is found or there are no further invocations of the interpreter left on the call stack. With the -unwind switch the evaluation stack for the interpreter is unwound without regard to any intervening catch command until there are no further invocations of the interpreter left on the call stack. If result is present, it will be used as the error message string; otherwise, a default error message string will be used.
thread::unwind: Use of this command is deprecated in favour of more advanced thread reservation system implemented with thread::preserve and thread::release commands. Support for thread::unwind command will disappear in some future major release of the extension.
This command stops a prior thread::wait command. Execution of the script passed to newly created thread will continue from the thread::wait command. If thread::wait was the last command in the script, the thread will exit. The command returns empty result but may trigger Tcl error with the message "target thread died" in some situations.
thread::exit ?status?: Use of this command is deprecated in favour of more advanced thread reservation system implemented with thread::preserve and thread::release commands. Support for thread::exit command will disappear in some future major release of the extension.
This command forces a thread stuck in the thread::wait command to unconditionally exit. The thread's exit status defaults to 666 and can be specified using the optional status argument. The execution of thread::exit command is guaranteed to leave the program memory in the inconsistent state, produce memory leaks and otherwise affect other subsystem(s) of the Tcl application in an unpredictable manner. The command returns empty result but may trigger Tcl error with the message "target thread died" in some situations.
thread::names: This command returns a list of thread IDs. These are only for threads that have been created via thread::create command. If your application creates other threads at the C level, they are not reported by this command.
thread::exists id: Returns true (1) if thread given by the id parameter exists, false (0) otherwise. This applies only for threads that have been created via thread::create command.
thread::send ?-async? ?-head? id script ?varname?: This command passes a script to another thread and, optionally, waits for the result. If the -async flag is specified, the command does not wait for the result and it returns empty string. The target thread must enter it's event loop in order to receive scripts sent via this command. This is done by default for threads created without a startup script. Threads can enter the event loop explicitly by calling thread::wait or any other relevant Tcl/Tk command, like update, vwait, etc.
Optional varname specifies name of the variable to store the result of the script. Without the -async flag, the command returns the evaluation code, similarly to the standard Tcl catch command. If, however, the -async flag is specified, the command returns immediately and caller can later vwait on ?varname? to get the result of the passed script

    set t1 [thread::create]
    set t2 [thread::create]
    thread::send -async $t1 "set a 1" result
    thread::send -async $t2 "set b 2" result
    for {set i 0} {$i < 2} {incr i} {
        vwait result
    }

: In the above example, two threads were fed work and both of them were instructed to signalize the same variable "result" in the calling thread. The caller entered the event loop twice to get both results. Note, however, that the order of the received results may vary, depending on the current system load, type of work done, etc, etc.
Many threads can simultaneously send scripts to the target thread for execution. All of them are entered into the event queue of the target thread and executed on the FIFO basis, intermingled with optional other events pending in the event queue of the target thread. Using the optional ?-head? switch, scripts posted to the thread's event queue can be placed on the head, instead on the tail of the queue, thus being executed in the LIFO fashion.

thread::broadcast script: This command passes a script to all threads created by the package for execution. It does not wait for response from any of the threads.
thread::wait: This enters the event loop so a thread can receive messages from the thread::send command. This command should only be used within the script passed to the thread::create. It should be the very last command in the script. If this is not the case, the exiting thread will continue executing the script lines past the thread::wait which is usually not what you want and/or expect.

    set t1 [thread::create {
        #
        # Do some initialization work here
        #
        thread::wait ; # Enter the event loop
    }]

thread::eval ?-lock mutex? arg ?arg ...?

This command concatenates passed arguments and evaluates the resulting script under the mutex protection. If no mutex is specified by using the ?-lock mutex? optional argument, the internal static mutex is used.

thread::join id

This command waits for the thread with ID id to exit and then returns it's exit code. Errors will be returned for threads which are not joinable or already waited upon by another thread. Upon the join the handle of the thread has gone out of scope and should not be used any more.

thread::configure id ?option? ?value? ?...?

This command configures various low-level aspects of the thread with ID id in the similar way as the standard Tcl command fconfigure configures some Tcl channel options. Options currently supported are: -eventmark and -unwindonerror.

When -eventmark is provided with a value greater than 0 (zero), that value is the maximum number of asynchronously posted scripts that may be pending for the thread. thread::send -async blocks until the number of pending scripts in the event loop drops below the -eventmark value.

When -unwindonerror is provided with a value of true, an error result in a script causes the thread to unwind, making it unavailable to evaluate additional scripts.

thread::transfer id channel

This moves the specified channel from the current thread and interpreter to the main interpreter of the thread with the given id. After the move the current interpreter has no access to the channel any more, but the main interpreter of the target thread will be able to use it from now on. The command waits until the other thread has incorporated the channel. Because of this it is possible to deadlock the participating threads by commanding the other through a synchronous thread::send to transfer a channel to us. This easily extends into longer loops of threads waiting for each other. Other restrictions: the channel in question must not be shared among multiple interpreters running in the sending thread. This automatically excludes the special channels for standard input, output and error.

Due to the internal Tcl core implementation and the restriction on transferring shared channels, one has to take extra measures when transferring socket channels created by accepting the connection out of the socket commands callback procedures:

    socket -server _Accept 2200
    proc _Accept {s ipaddr port} {
        after idle [list Accept $s $ipaddr $port]
    }
    proc Accept {s ipaddr port} {
        set tid [thread::create]
        thread::transfer $tid $s
    }

thread::detach channel

This detaches the specified channel from the current thread and interpreter. After that, the current interpreter has no access to the channel any more. The channel is in the parked state until some other (or the same) thread attaches the channel again with thread::attach. Restrictions: same as for transferring shared channels with the thread::transfer command.

thread::attach channel

This attaches the previously detached channel in the current thread/interpreter. For already existing channels, the command does nothing, i.e. it is not an error to attach the same channel more than once. The first operation will actually perform the operation, while all subsequent operation will just do nothing. Command throws error if the channel cannot be found in the list of detached channels and/or in the current interpreter.

thread::mutex

Mutexes are most common thread synchronization primitives. They are used to synchronize access from two or more threads to one or more shared resources. This command provides script-level access to exclusive and/or recursive mutexes. Exclusive mutexes can be locked only once by one thread, while recursive mutexes can be locked many times by the same thread. For recursive mutexes, number of lock and unlock operations must match, otherwise, the mutex will never be released, which would lead to various deadlock situations.

Care has to be taken when using mutexes in an multithreading program. Improper use of mutexes may lead to various deadlock situations, especially when using exclusive mutexes.

The thread::mutex command supports following subcommands and options:

thread::mutex create ?-recursive?: Creates the mutex and returns it's opaque handle. This handle should be used for any future reference to the newly created mutex. If no optional ?-recursive? argument was specified, the command creates the exclusive mutex. With the ?-recursive? argument, the command creates a recursive mutex.
thread::mutex destroy mutex: Destroys the mutex. Mutex should be in unlocked state before the destroy attempt. If the mutex is locked, the command will throw Tcl error.
thread::mutex lock mutex: Locks the mutex. Locking the exclusive mutex may throw Tcl error if on attempt to lock the same mutex twice from the same thread. If your program logic forces you to lock the same mutex twice or more from the same thread (this may happen in recursive procedure invocations) you should consider using the recursive mutexes.
thread::mutex unlock mutex: Unlocks the mutex so some other thread may lock it again. Attempt to unlock the already unlocked mutex will throw Tcl error.

thread::rwmutex

This command creates many-readers/single-writer mutexes. Reader/writer mutexes allow you to serialize access to a shared resource more optimally. In situations where a shared resource gets mostly read and seldom modified, you might gain some performance by using reader/writer mutexes instead of exclusive or recursive mutexes.

For reading the resource, thread should obtain a read lock on the resource. Read lock is non-exclusive, meaning that more than one thread can obtain a read lock to the same resource, without waiting on other readers. For changing the resource, however, a thread must obtain a exclusive write lock. This lock effectively blocks all threads from gaining the read-lock while the resource is been modified by the writer thread. Only after the write lock has been released, the resource may be read-locked again.

The thread::rwmutex command supports following subcommands and options:

thread::rwmutex create: Creates the reader/writer mutex and returns it's opaque handle. This handle should be used for any future reference to the newly created mutex.
thread::rwmutex destroy mutex: Destroys the reader/writer mutex. If the mutex is already locked, attempt to destroy it will throw Tcl error.
thread::rwmutex rlock mutex: Locks the mutex for reading. More than one thread may read-lock the same mutex at the same time.
thread::rwmutex wlock mutex: Locks the mutex for writing. Only one thread may write-lock the same mutex at the same time. Attempt to write-lock same mutex twice from the same thread will throw Tcl error.
thread::rwmutex unlock mutex: Unlocks the mutex so some other thread may lock it again. Attempt to unlock already unlocked mutex will throw Tcl error.

thread::cond: This command provides script-level access to condition variables. A condition variable creates a safe environment for the program to test some condition, sleep on it when false and be awakened when it might have become true. A condition variable is always used in the conjunction with an exclusive mutex. If you attempt to use other type of mutex in conjunction with the condition variable, a Tcl error will be thrown.
The command supports following subcommands and options:

thread::cond create

Creates the condition variable and returns it's opaque handle. This handle should be used for any future reference to newly created condition variable.

thread::cond destroy cond

Destroys condition variable cond. Extreme care has to be taken that nobody is using (i.e. waiting on) the condition variable, otherwise unexpected errors may happen.

thread::cond notify cond

Wakes up all threads waiting on the condition variable cond.

thread::cond wait cond mutex ?ms?

This command is used to suspend program execution until the condition variable cond has been signalled or the optional timer has expired. The exclusive mutex must be locked by the calling thread on entrance to this command. If the mutex is not locked, Tcl error is thrown. While waiting on the cond, the command releases mutex. Before returning to the calling thread, the command re-acquires the mutex again. Unlocking the mutex and waiting on the condition variable cond is done atomically.

The ms command option, if given, must be an integer specifying time interval in milliseconds the command waits to be signalled. Otherwise the command waits on condition notify forever.

In multithreading programs, there are many situations where a thread has to wait for some event to happen until it is allowed to proceed. This is usually accomplished by repeatedly testing a condition under the mutex protection and waiting on the condition variable until the condition evaluates to true:

    set mutex [thread::mutex create]
    set cond  [thread::cond  create]
    thread::mutex lock $mutex
    while {<some_condition_is_true>} {
        thread::cond wait $cond $mutex
    }
    # Do some work under mutex protection
    thread::mutex unlock $mutex

: Repeated testing of the condition is needed since the condition variable may get signalled without the condition being actually changed (spurious thread wake-ups, for example).

DISCUSSION

The fundamental threading model in Tcl is that there can be one or more Tcl interpreters per thread, but each Tcl interpreter should only be used by a single thread which created it. A "shared memory" abstraction is awkward to provide in Tcl because Tcl makes assumptions about variable and data ownership. Therefore this extension supports a simple form of threading where the main thread can manage several background, or "worker" threads. For example, an event-driven server can pass requests to worker threads, and then await responses from worker threads or new client requests. Everything goes through the common Tcl event loop, so message passing between threads works naturally with event-driven I/O, vwait on variables, and so forth. For the transfer of bulk information it is possible to move channels between the threads.

For advanced multithreading scripts, script-level access to two basic synchronization primitives, mutex and condition variables, is also supported.

KEYWORDS