##
##  GNU Pth - The GNU Portable Threads
##  Copyright (c) 1999-2006 Ralf S. Engelschall <rse@engelschall.com>
##
##  This file is part of GNU Pth, a non-preemptive thread scheduling
##  library which can be found at http://www.gnu.org/software/pth/.
##
##  This library is free software; you can redistribute it and/or
##  modify it under the terms of the GNU Lesser General Public
##  License as published by the Free Software Foundation; either
##  version 2.1 of the License, or (at your option) any later version.
##
##  This library is distributed in the hope that it will be useful,
##  but WITHOUT ANY WARRANTY; without even the implied warranty of
##  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
##  Lesser General Public License for more details.
##
##  You should have received a copy of the GNU Lesser General Public
##  License along with this library; if not, write to the Free Software
##  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
##  USA, or contact Ralf S. Engelschall <rse@engelschall.com>.
##
##  pth.pod: Pth manual page
##

#                            ``Real programmers don't document.
#                              Documentation is for wimps who can't
#                              read the listings of the object deck.''

=pod

=head1 NAME

B<pth> - GNU Portable Threads

=head1 VERSION

GNU Pth PTH_VERSION_STR

=head1 SYNOPSIS

=over 4

=item B<Global Library Management>

pth_init,
pth_kill,
pth_ctrl,
pth_version.

=item B<Thread Attribute Handling>

pth_attr_of,
pth_attr_new,
pth_attr_init,
pth_attr_set,
pth_attr_get,
pth_attr_destroy.

=item B<Thread Control>

pth_spawn,
pth_once,
pth_self,
pth_suspend,
pth_resume,
pth_yield,
pth_nap,
pth_wait,
pth_cancel,
pth_abort,
pth_raise,
pth_join,
pth_exit.

=item B<Utilities>

pth_fdmode,
pth_time,
pth_timeout,
pth_sfiodisc.

=item B<Cancellation Management>

pth_cancel_point,
pth_cancel_state.

=item B<Event Handling>

pth_event,
pth_event_typeof,
pth_event_extract,
pth_event_concat,
pth_event_isolate,
pth_event_walk,
pth_event_status,
pth_event_free.

=item B<Key-Based Storage>

pth_key_create,
pth_key_delete,
pth_key_setdata,
pth_key_getdata.

=item B<Message Port Communication>

pth_msgport_create,
pth_msgport_destroy,
pth_msgport_find,
pth_msgport_pending,
pth_msgport_put,
pth_msgport_get,
pth_msgport_reply.

=item B<Thread Cleanups>

pth_cleanup_push,
pth_cleanup_pop.

=item B<Process Forking>

pth_atfork_push,
pth_atfork_pop,
pth_fork.

=item B<Synchronization>

pth_mutex_init,
pth_mutex_acquire,
pth_mutex_release,
pth_rwlock_init,
pth_rwlock_acquire,
pth_rwlock_release,
pth_cond_init,
pth_cond_await,
pth_cond_notify,
pth_barrier_init,
pth_barrier_reach.

=item B<User-Space Context>

pth_uctx_create,
pth_uctx_make,
pth_uctx_switch,
pth_uctx_destroy.

=item B<Generalized POSIX Replacement API>

pth_sigwait_ev,
pth_accept_ev,
pth_connect_ev,
pth_select_ev,
pth_poll_ev,
pth_read_ev,
pth_readv_ev,
pth_write_ev,
pth_writev_ev,
pth_recv_ev,
pth_recvfrom_ev,
pth_send_ev,
pth_sendto_ev.

=item B<Standard POSIX Replacement API>

pth_nanosleep,
pth_usleep,
pth_sleep,
pth_waitpid,
pth_system,
pth_sigmask,
pth_sigwait,
pth_accept,
pth_connect,
pth_select,
pth_pselect,
pth_poll,
pth_read,
pth_readv,
pth_write,
pth_writev,
pth_pread,
pth_pwrite,
pth_recv,
pth_recvfrom,
pth_send,
pth_sendto.

=back

=head1 DESCRIPTION

  ____  _   _
 |  _ \| |_| |__
 | |_) | __| '_ \         ``Only those who attempt
 |  __/| |_| | | |          the absurd can achieve
 |_|    \__|_| |_|          the impossible.''

B<Pth> is a very portable POSIX/ANSI-C based library for Unix platforms which
provides non-preemptive priority-based scheduling for multiple threads of
execution (aka `multithreading') inside event-driven applications. All threads
run in the same address space of the application process, but each thread has
its own individual program counter, run-time stack, signal mask and C<errno>
variable.

The thread scheduling itself is done in a cooperative way, i.e., the threads
are managed and dispatched by a priority- and event-driven non-preemptive
scheduler. The intention is that this way both better portability and run-time
performance is achieved than with preemptive scheduling. The event facility
allows threads to wait until various types of internal and external events
occur, including pending I/O on file descriptors, asynchronous signals,
elapsed timers, pending I/O on message ports, thread and process termination,
and even results of customized callback functions.

B<Pth> also provides an optional emulation API for POSIX.1c threads
(`Pthreads') which can be used for backward compatibility to existing
multithreaded applications. See B<Pth>'s pthread(3) manual page for
details.

=head2 Threading Background

When programming event-driven applications, usually servers, lots of
regular jobs and one-shot requests have to be processed in parallel.
To efficiently simulate this parallel processing on uniprocessor
machines, we use `multitasking' -- that is, we have the application
ask the operating system to spawn multiple instances of itself. On
Unix, typically the kernel implements multitasking in a preemptive and
priority-based way through heavy-weight processes spawned with fork(2).
These processes usually do I<not> share a common address space. Instead
they are clearly separated from each other, and are created by direct
cloning a process address space (although modern kernels use memory
segment mapping and copy-on-write semantics to avoid unnecessary copying
of physical memory).

The drawbacks are obvious: Sharing data between the processes is
complicated, and can usually only be done efficiently through shared
memory (but which itself is not very portable). Synchronization is
complicated because of the preemptive nature of the Unix scheduler
(one has to use I<atomic> locks, etc). The machine's resources can be
exhausted very quickly when the server application has to serve too many
long-running requests (heavy-weight processes cost memory). And when
each request spawns a sub-process to handle it, the server performance
and responsiveness is horrible (heavy-weight processes cost time to
spawn). Finally, the server application doesn't scale very well with the
load because of these resource problems. In practice, lots of tricks
are usually used to overcome these problems - ranging from pre-forked
sub-process pools to semi-serialized processing, etc.

One of the most elegant ways to solve these resource- and data-sharing
problems is to have multiple I<light-weight> threads of execution
inside a single (heavy-weight) process, i.e., to use I<multithreading>.
Those I<threads> usually improve responsiveness and performance of the
application, often improve and simplify the internal program structure,
and most important, require less system resources than heavy-weight
processes. Threads are neither the optimal run-time facility for all
types of applications, nor can all applications benefit from them. But
at least event-driven server applications usually benefit greatly from
using threads.

=head2 The World of Threading

Even though lots of documents exists which describe and define the world
of threading, to understand B<Pth>, you need only basic knowledge about
threading. The following definitions of thread-related terms should at
least help you understand thread programming enough to allow you to use
B<Pth>.

=over 2

=item B<o> B<process> vs. B<thread>

A process on Unix systems consists of at least the following fundamental
ingredients: I<virtual memory table>, I<program code>, I<program
counter>, I<heap memory>, I<stack memory>, I<stack pointer>, I<file
descriptor set>, I<signal table>. On every process switch, the kernel
saves and restores these ingredients for the individual processes. On
the other hand, a thread consists of only a private program counter,
stack memory, stack pointer and signal table. All other ingredients, in
particular the virtual memory, it shares with the other threads of the
same process.

=item B<o> B<kernel-space> vs. B<user-space> threading

Threads on a Unix platform traditionally can be implemented either
inside kernel-space or user-space. When threads are implemented by the
kernel, the thread context switches are performed by the kernel without
the application's knowledge. Similarly, when threads are implemented in
user-space, the thread context switches are performed by an application
library, without the kernel's knowledge. There also are hybrid threading
approaches where, typically, a user-space library binds one or more
user-space threads to one or more kernel-space threads (there usually
called light-weight processes - or in short LWPs).

User-space threads are usually more portable and can perform faster
and cheaper context switches (for instance via swapcontext(2) or
setjmp(3)/longjmp(3)) than kernel based threads. On the other hand,
kernel-space threads can take advantage of multiprocessor machines and
don't have any inherent I/O blocking problems. Kernel-space threads are
usually scheduled in preemptive way side-by-side with the underlying
processes. User-space threads on the other hand use either preemptive or
non-preemptive scheduling.

=item B<o> B<preemptive> vs. B<non-preemptive> thread scheduling

In preemptive scheduling, the scheduler lets a thread execute until a
blocking situation occurs (usually a function call which would block)
or the assigned timeslice elapses. Then it detracts control from the
thread without a chance for the thread to object. This is usually
realized by interrupting the thread through a hardware interrupt
signal (for kernel-space threads) or a software interrupt signal (for
user-space threads), like C<SIGALRM> or C<SIGVTALRM>. In non-preemptive
scheduling, once a thread received control from the scheduler it keeps
it until either a blocking situation occurs (again a function call which
would block and instead switches back to the scheduler) or the thread
explicitly yields control back to the scheduler in a cooperative way.

=item B<o> B<concurrency> vs. B<parallelism>

Concurrency exists when at least two threads are I<in progress> at the
same time. Parallelism arises when at least two threads are I<executing>
simultaneously. Real parallelism can be only achieved on multiprocessor
machines, of course. But one also usually speaks of parallelism or
I<high concurrency> in the context of preemptive thread scheduling
and of I<low concurrency> in the context of non-preemptive thread
scheduling.

=item B<o> B<responsiveness>

The responsiveness of a system can be described by the user visible
delay until the system responses to an external request. When this delay
is small enough and the user doesn't recognize a noticeable delay,
the responsiveness of the system is considered good. When the user
recognizes or is even annoyed by the delay, the responsiveness of the
system is considered bad.

=item B<o> B<reentrant>, B<thread-safe> and B<asynchronous-safe> functions

A reentrant function is one that behaves correctly if it is called
simultaneously by several threads and then also executes simultaneously.
Functions that access global state, such as memory or files, of course,
need to be carefully designed in order to be reentrant. Two traditional
approaches to solve these problems are caller-supplied states and
thread-specific data.

Thread-safety is the avoidance of I<data races>, i.e., situations
in which data is set to either correct or incorrect value depending
upon the (unpredictable) order in which multiple threads access and
modify the data. So a function is thread-safe when it still behaves
semantically correct when called simultaneously by several threads (it
is not required that the functions also execute simultaneously). The
traditional approach to achieve thread-safety is to wrap a function body
with an internal mutual exclusion lock (aka `mutex'). As you should
recognize, reentrant is a stronger attribute than thread-safe, because
it is harder to achieve and results especially in no run-time contention
between threads. So, a reentrant function is always thread-safe, but not
vice versa.

Additionally there is a related attribute for functions named
asynchronous-safe, which comes into play in conjunction with signal
handlers. This is very related to the problem of reentrant functions. An
asynchronous-safe function is one that can be called safe and without
side-effects from within a signal handler context. Usually very few
functions are of this type, because an application is very restricted in
what it can perform from within a signal handler (especially what system
functions it is allowed to call). The reason mainly is, because only a
few system functions are officially declared by POSIX as guaranteed to
be asynchronous-safe. Asynchronous-safe functions usually have to be
already reentrant.

=back

=head2 User-Space Threads

User-space threads can be implemented in various way. The two
traditional approaches are:

=over 3

=item B<1.>

B<Matrix-based explicit dispatching between small units of execution:>

Here the global procedures of the application are split into small
execution units (each is required to not run for more than a few
milliseconds) and those units are implemented by separate functions.
Then a global matrix is defined which describes the execution (and
perhaps even dependency) order of these functions. The main server
procedure then just dispatches between these units by calling one
function after each other controlled by this matrix. The threads are
created by more than one jump-trail through this matrix and by switching
between these jump-trails controlled by corresponding occurred events.

This approach gives the best possible performance, because one can
fine-tune the threads of execution by adjusting the matrix, and the
scheduling is done explicitly by the application itself. It is also very
portable, because the matrix is just an ordinary data structure, and
functions are a standard feature of ANSI C.

The disadvantage of this approach is that it is complicated to write
large applications with this approach, because in those applications
one quickly gets hundreds(!) of execution units and the control flow
inside such an application is very hard to understand (because it is
interrupted by function borders and one always has to remember the
global dispatching matrix to follow it). Additionally, all threads
operate on the same execution stack. Although this saves memory, it is
often nasty, because one cannot switch between threads in the middle of
a function. Thus the scheduling borders are the function borders.

=item B<2.>

B<Context-based implicit scheduling between threads of execution:>

Here the idea is that one programs the application as with forked
processes, i.e., one spawns a thread of execution and this runs from the
begin to the end without an interrupted control flow. But the control
flow can be still interrupted - even in the middle of a function.
Actually in a preemptive way, similar to what the kernel does for the
heavy-weight processes, i.e., every few milliseconds the user-space
scheduler switches between the threads of execution. But the thread
itself doesn't recognize this and usually (except for synchronization
issues) doesn't have to care about this.

The advantage of this approach is that it's very easy to program,
because the control flow and context of a thread directly follows
a procedure without forced interrupts through function borders.
Additionally, the programming is very similar to a traditional and well
understood fork(2) based approach.

The disadvantage is that although the general performance is increased,
compared to using approaches based on heavy-weight processes, it is decreased
compared to the matrix-approach above. Because the implicit preemptive
scheduling does usually a lot more context switches (every user-space context
switch costs some overhead even when it is a lot cheaper than a kernel-level
context switch) than the explicit cooperative/non-preemptive scheduling.
Finally, there is no really portable POSIX/ANSI-C based way to implement
user-space preemptive threading. Either the platform already has threads,
or one has to hope that some semi-portable package exists for it. And
even those semi-portable packages usually have to deal with assembler
code and other nasty internals and are not easy to port to forthcoming
platforms.

=back

So, in short: the matrix-dispatching approach is portable and fast, but
nasty to program. The thread scheduling approach is easy to program,
but suffers from synchronization and portability problems caused by its
preemptive nature.

=head2 The Compromise of Pth

But why not combine the good aspects of both approaches while avoiding
their bad aspects? That's the goal of B<Pth>. B<Pth> implements
easy-to-program threads of execution, but avoids the problems of
preemptive scheduling by using non-preemptive scheduling instead.

This sounds like, and is, a useful approach. Nevertheless, one has to
keep the implications of non-preemptive thread scheduling in mind when
working with B<Pth>. The following list summarizes a few essential
points:

=over 2

=item B<o>

B<Pth provides maximum portability, but NOT the fanciest features>.

This is, because it uses a nifty and portable POSIX/ANSI-C approach for
thread creation (and this way doesn't require any platform dependent
assembler hacks) and schedules the threads in non-preemptive way (which
doesn't require unportable facilities like C<SIGVTALRM>). On the other
hand, this way not all fancy threading features can be implemented.
Nevertheless the available facilities are enough to provide a robust and
full-featured threading system.

=item B<o>

B<Pth increases the responsiveness and concurrency of an event-driven
application, but NOT the concurrency of number-crunching applications>.

The reason is the non-preemptive scheduling. Number-crunching
applications usually require preemptive scheduling to achieve
concurrency because of their long CPU bursts. For them, non-preemptive
scheduling (even together with explicit yielding) provides only the old
concept of `coroutines'. On the other hand, event driven applications
benefit greatly from non-preemptive scheduling. They have only short
CPU bursts and lots of events to wait on, and this way run faster under
non-preemptive scheduling because no unnecessary context switching
occurs, as it is the case for preemptive scheduling. That's why B<Pth>
is mainly intended for server type applications, although there is no
technical restriction.

=item B<o>

B<Pth requires thread-safe functions, but NOT reentrant functions>.

This nice fact exists again because of the nature of non-preemptive
scheduling, where a function isn't interrupted and this way cannot be
reentered before it returned. This is a great portability benefit,
because thread-safety can be achieved more easily than reentrance
possibility. Especially this means that under B<Pth> more existing
third-party libraries can be used without side-effects than it's the case
for other threading systems.

=item B<o>

B<Pth doesn't require any kernel support, but can NOT
benefit from multiprocessor machines>.

This means that B<Pth> runs on almost all Unix kernels, because the
kernel does not need to be aware of the B<Pth> threads (because they
are implemented entirely in user-space). On the other hand, it cannot
benefit from the existence of multiprocessors, because for this, kernel
support would be needed. In practice, this is no problem, because
multiprocessor systems are rare, and portability is almost more
important than highest concurrency.

=back

=head2 The life cycle of a thread

To understand the B<Pth> Application Programming Interface (API), it
helps to first understand the life cycle of a thread in the B<Pth>
threading system. It can be illustrated with the following directed
graph:

             NEW
              |
              V
      +---> READY ---+
      |       ^      |
      |       |      V
   WAITING <--+-- RUNNING
                     |
      :              V
   SUSPENDED       DEAD

When a new thread is created, it is moved into the B<NEW> queue of the
scheduler. On the next dispatching for this thread, the scheduler picks
it up from there and moves it to the B<READY> queue. This is a queue
containing all threads which want to perform a CPU burst. There they are
queued in priority order. On each dispatching step, the scheduler always
removes the thread with the highest priority only. It then increases the
priority of all remaining threads by 1, to prevent them from `starving'.

The thread which was removed from the B<READY> queue is the new
B<RUNNING> thread (there is always just one B<RUNNING> thread, of
course). The B<RUNNING> thread is assigned execution control. After
this thread yields execution (either explicitly by yielding execution
or implicitly by calling a function which would block) there are three
possibilities: Either it has terminated, then it is moved to the B<DEAD>
queue, or it has events on which it wants to wait, then it is moved into
the B<WAITING> queue. Else it is assumed it wants to perform more CPU
bursts and immediately enters the B<READY> queue again.

Before the next thread is taken out of the B<READY> queue, the
B<WAITING> queue is checked for pending events. If one or more events
occurred, the threads that are waiting on them are immediately moved to
the B<READY> queue.

The purpose of the B<NEW> queue has to do with the fact that in B<Pth>
a thread never directly switches to another thread. A thread always
yields execution to the scheduler and the scheduler dispatches to the
next thread. So a freshly spawned thread has to be kept somewhere until
the scheduler gets a chance to pick it up for scheduling. That is
what the B<NEW> queue is for.

The purpose of the B<DEAD> queue is to support thread joining. When a
thread is marked to be unjoinable, it is directly kicked out of the
system after it terminated. But when it is joinable, it enters the
B<DEAD> queue. There it remains until another thread joins it.

Finally, there is a special separated queue named B<SUSPENDED>, to where
threads can be manually moved from the B<NEW>, B<READY> or B<WAITING>
queues by the application. The purpose of this special queue is to
temporarily absorb suspended threads until they are again resumed by
the application. Suspended threads do not cost scheduling or event
handling resources, because they are temporarily completely out of the
scheduler's scope. If a thread is resumed, it is moved back to the queue
from where it originally came and this way again enters the schedulers
scope.

=head1 APPLICATION PROGRAMMING INTERFACE (API)

In the following the B<Pth> I<Application Programming Interface> (API)
is discussed in detail. With the knowledge given above, it should now
be easy to understand how to program threads with this API. In good
Unix tradition, B<Pth> functions use special return values (C<NULL>
in pointer context, C<FALSE> in boolean context and C<-1> in integer
context) to indicate an error condition and set (or pass through) the
C<errno> system variable to pass more details about the error to the
caller.

=head2 Global Library Management

The following functions act on the library as a whole.  They are used to
initialize and shutdown the scheduler and fetch information from it.

=over 4

=item int B<pth_init>(void);

This initializes the B<Pth> library. It has to be the first B<Pth> API
function call in an application, and is mandatory. It's usually done at
the begin of the main() function of the application. This implicitly
spawns the internal scheduler thread and transforms the single execution
unit of the current process into a thread (the `main' thread). It
returns C<TRUE> on success and C<FALSE> on error.

=item int B<pth_kill>(void);

This kills the B<Pth> library. It should be the last B<Pth> API function call
in an application, but is not really required. It's usually done at the end of
the main function of the application. At least, it has to be called from within
the main thread. It implicitly kills all threads and transforms back the
calling thread into the single execution unit of the underlying process.  The
usual way to terminate a B<Pth> application is either a simple
`C<pth_exit(0);>' in the main thread (which waits for all other threads to
terminate, kills the threading system and then terminates the process) or a
`C<pth_kill(); exit(0)>' (which immediately kills the threading system and
terminates the process). The pth_kill() return immediately with a return
code of C<FALSE> if it is not called from within the main thread. Else it
kills the threading system and returns C<TRUE>.

=item long B<pth_ctrl>(unsigned long I<query>, ...);

This is a generalized query/control function for the B<Pth> library.  The
argument I<query> is a bitmask formed out of one or more C<PTH_CTRL_>I<XXXX>
queries. Currently the following queries are supported:

=over 4

=item C<PTH_CTRL_GETTHREADS>

This returns the total number of threads currently in existence.  This query
actually is formed out of the combination of queries for threads in a
particular state, i.e., the C<PTH_CTRL_GETTHREADS> query is equal to the
OR-combination of all the following specialized queries:

C<PTH_CTRL_GETTHREADS_NEW> for the number of threads in the
new queue (threads created via pth_spawn(3) but still not
scheduled once), C<PTH_CTRL_GETTHREADS_READY> for the number of
threads in the ready queue (threads who want to do CPU bursts),
C<PTH_CTRL_GETTHREADS_RUNNING> for the number of running threads
(always just one thread!), C<PTH_CTRL_GETTHREADS_WAITING> for
the number of threads in the waiting queue (threads waiting for
events), C<PTH_CTRL_GETTHREADS_SUSPENDED> for the number of
threads in the suspended queue (threads waiting to be resumed) and
C<PTH_CTRL_GETTHREADS_DEAD> for the number of threads in the new queue
(terminated threads waiting for a join).

=item C<PTH_CTRL_GETAVLOAD>

This requires a second argument of type `C<float *>' (pointer to a floating
point variable).  It stores a floating point value describing the exponential
averaged load of the scheduler in this variable. The load is a function from
the number of threads in the ready queue of the schedulers dispatching unit.
So a load around 1.0 means there is only one ready thread (the standard
situation when the application has no high load). A higher load value means
there a more threads ready who want to do CPU bursts. The average load value
updates once per second only. The return value for this query is always 0.

=item C<PTH_CTRL_GETPRIO>

This requires a second argument of type `C<pth_t>' which identifies a
thread.  It returns the priority (ranging from C<PTH_PRIO_MIN> to
C<PTH_PRIO_MAX>) of the given thread.

=item C<PTH_CTRL_GETNAME>

This requires a second argument of type `C<pth_t>' which identifies a
thread. It returns the name of the given thread, i.e., the return value of
pth_ctrl(3) should be casted to a `C<char *>'.

=item C<PTH_CTRL_DUMPSTATE>

This requires a second argument of type `C<FILE *>' to which a summary
of the internal B<Pth> library state is written to. The main information
which is currently written out is the current state of the thread pool.

=item C<PTH_CTRL_FAVOURNEW>

This requires a second argument of type `C<int>' which specified whether
the B<GNU Pth> scheduler favours new threads on startup, i.e., whether
they are moved from the new queue to the top (argument is C<TRUE>) or
middle (argument is C<FALSE>) of the ready queue. The default is to
favour new threads to make sure they do not starve already at startup,
although this slightly violates the strict priority based scheduling.

=back

The function returns C<-1> on error.

=item long B<pth_version>(void);

This function returns a hex-value `0xI<V>I<RR>I<T>I<LL>' which describes the
current B<Pth> library version. I<V> is the version, I<RR> the revisions,
I<LL> the level and I<T> the type of the level (alphalevel=0, betalevel=1,
patchlevel=2, etc). For instance B<Pth> version 1.0b1 is encoded as 0x100101.
The reason for this unusual mapping is that this way the version number is
steadily I<increasing>. The same value is also available under compile time as
C<PTH_VERSION>.

=back

=head2 Thread Attribute Handling

Attribute objects are used in B<Pth> for two things: First stand-alone/unbound
attribute objects are used to store attributes for to be spawned threads.
Bounded attribute objects are used to modify attributes of already existing
threads. The following attribute fields exists in attribute objects:

=over 4

=item C<PTH_ATTR_PRIO> (read-write) [C<int>]

Thread Priority between C<PTH_PRIO_MIN> and C<PTH_PRIO_MAX>.
The default is C<PTH_PRIO_STD>.

=item C<PTH_ATTR_NAME> (read-write) [C<char *>]

Name of thread (up to 40 characters are stored only), mainly for debugging
purposes.

=item C<PTH_ATTR_DISPATCHES> (read-write) [C<int>]

In bounded attribute objects, this field is incremented every time the
context is switched to the associated thread.

=item C<PTH_ATTR_JOINABLE> (read-write> [C<int>]

The thread detachment type, C<TRUE> indicates a joinable thread,
C<FALSE> indicates a detached thread. When a thread is detached,
after termination it is immediately kicked out of the system instead of
inserted into the dead queue.

=item C<PTH_ATTR_CANCEL_STATE> (read-write) [C<unsigned int>]

The thread cancellation state, i.e., a combination of C<PTH_CANCEL_ENABLE> or
C<PTH_CANCEL_DISABLE> and C<PTH_CANCEL_DEFERRED> or
C<PTH_CANCEL_ASYNCHRONOUS>.

=item C<PTH_ATTR_STACK_SIZE> (read-write) [C<unsigned int>]

The thread stack size in bytes. Use lower values than 64 KB with great care!

=item C<PTH_ATTR_STACK_ADDR> (read-write) [C<char *>]

A pointer to the lower address of a chunk of malloc(3)'ed memory for the
stack.

=item C<PTH_ATTR_TIME_SPAWN> (read-only) [C<pth_time_t>]

The time when the thread was spawned.
This can be queried only when the attribute object is bound to a thread.

=item C<PTH_ATTR_TIME_LAST> (read-only) [C<pth_time_t>]

The time when the thread was last dispatched.
This can be queried only when the attribute object is bound to a thread.

=item C<PTH_ATTR_TIME_RAN> (read-only) [C<pth_time_t>]

The total time the thread was running.
This can be queried only when the attribute object is bound to a thread.

=item C<PTH_ATTR_START_FUNC> (read-only) [C<void *(*)(void *)>]

The thread start function.
This can be queried only when the attribute object is bound to a thread.

=item C<PTH_ATTR_START_ARG> (read-only) [C<void *>]

The thread start argument.
This can be queried only when the attribute object is bound to a thread.

=item C<PTH_ATTR_STATE> (read-only) [C<pth_state_t>]

The scheduling state of the thread, i.e., either C<PTH_STATE_NEW>,
C<PTH_STATE_READY>, C<PTH_STATE_WAITING>, or C<PTH_STATE_DEAD>
This can be queried only when the attribute object is bound to a thread.

=item C<PTH_ATTR_EVENTS> (read-only) [C<pth_event_t>]

The event ring the thread is waiting for.
This can be queried only when the attribute object is bound to a thread.

=item C<PTH_ATTR_BOUND> (read-only) [C<int>]

Whether the attribute object is bound (C<TRUE>) to a thread or not (C<FALSE>).

=back

The following API functions can be used to handle the attribute objects:

=over 4

=item pth_attr_t B<pth_attr_of>(pth_t I<tid>);

This returns a new attribute object I<bound> to thread I<tid>.  Any queries on
this object directly fetch attributes from I<tid>. And attribute modifications
directly change I<tid>. Use such attribute objects to modify existing threads.

=item pth_attr_t B<pth_attr_new>(void);

This returns a new I<unbound> attribute object. An implicit pth_attr_init() is
done on it. Any queries on this object just fetch stored attributes from it.
And attribute modifications just change the stored attributes.  Use such
attribute objects to pre-configure attributes for to be spawned threads.

=item int B<pth_attr_init>(pth_attr_t I<attr>);

This initializes an attribute object I<attr> to the default values:
C<PTH_ATTR_PRIO> := C<PTH_PRIO_STD>, C<PTH_ATTR_NAME> := `C<unknown>',
C<PTH_ATTR_DISPATCHES> := C<0>, C<PTH_ATTR_JOINABLE> := C<TRUE>,
C<PTH_ATTR_CANCELSTATE> := C<PTH_CANCEL_DEFAULT>,
C<PTH_ATTR_STACK_SIZE> := 64*1024 and
C<PTH_ATTR_STACK_ADDR> := C<NULL>. All other C<PTH_ATTR_*> attributes are
read-only attributes and don't receive default values in I<attr>, because they
exists only for bounded attribute objects.

=item int B<pth_attr_set>(pth_attr_t I<attr>, int I<field>, ...);

This sets the attribute field I<field> in I<attr> to a value
specified as an additional argument on the variable argument
list. The following attribute I<fields> and argument pairs can
be used:

 PTH_ATTR_PRIO           int
 PTH_ATTR_NAME           char *
 PTH_ATTR_DISPATCHES     int
 PTH_ATTR_JOINABLE       int
 PTH_ATTR_CANCEL_STATE   unsigned int
 PTH_ATTR_STACK_SIZE     unsigned int
 PTH_ATTR_STACK_ADDR     char *

=item int B<pth_attr_get>(pth_attr_t I<attr>, int I<field>, ...);

This retrieves the attribute field I<field> in I<attr> and stores its
value in the variable specified through a pointer in an additional
argument on the variable argument list. The following I<fields> and
argument pairs can be used:

 PTH_ATTR_PRIO           int *
 PTH_ATTR_NAME           char **
 PTH_ATTR_DISPATCHES     int *
 PTH_ATTR_JOINABLE       int *
 PTH_ATTR_CANCEL_STATE   unsigned int *
 PTH_ATTR_STACK_SIZE     unsigned int *
 PTH_ATTR_STACK_ADDR     char **
 PTH_ATTR_TIME_SPAWN     pth_time_t *
 PTH_ATTR_TIME_LAST      pth_time_t *
 PTH_ATTR_TIME_RAN       pth_time_t *
 PTH_ATTR_START_FUNC     void *(**)(void *)
 PTH_ATTR_START_ARG      void **
 PTH_ATTR_STATE          pth_state_t *
 PTH_ATTR_EVENTS         pth_event_t *
 PTH_ATTR_BOUND          int *

=item int B<pth_attr_destroy>(pth_attr_t I<attr>);

This destroys a attribute object I<attr>. After this I<attr> is no
longer a valid attribute object.

=back

=head2 Thread Control

The following functions control the threading itself and make up the main API
of the B<Pth> library.

=over 4

=item pth_t B<pth_spawn>(pth_attr_t I<attr>, void *(*I<entry>)(void *), void *I<arg>);

This spawns a new thread with the attributes given in I<attr> (or
C<PTH_ATTR_DEFAULT> for default attributes - which means that thread priority,
joinability and cancel state are inherited from the current thread) with the
starting point at routine I<entry>; the dispatch count is not inherited from
the current thread if I<attr> is not specified - rather, it is initialized
to zero.  This entry routine is called as `pth_exit(I<entry>(I<arg>))' inside
the new thread unit, i.e., I<entry>'s return value is fed to an implicit
pth_exit(3). So the thread can also exit by just returning. Nevertheless
the thread can also exit explicitly at any time by calling pth_exit(3). But
keep in mind that calling the POSIX function exit(3) still terminates the
complete process and not just the current thread.

There is no B<Pth>-internal limit on the number of threads one can spawn,
except the limit implied by the available virtual memory. B<Pth> internally
keeps track of thread in dynamic data structures. The function returns
C<NULL> on error.

=item int B<pth_once>(pth_once_t *I<ctrlvar>, void (*I<func>)(void *), void *I<arg>);

This is a convenience function which uses a control variable of type
C<pth_once_t> to make sure a constructor function I<func> is called only once
as `I<func>(I<arg>)' in the system. In other words: Only the first call to
pth_once(3) by any thread in the system succeeds. The variable referenced via
I<ctrlvar> should be declared as `C<pth_once_t> I<variable-name> =
C<PTH_ONCE_INIT>;' before calling this function.

=item pth_t B<pth_self>(void);

This just returns the unique thread handle of the currently running thread.
This handle itself has to be treated as an opaque entity by the application.
It's usually used as an argument to other functions who require an argument of
type C<pth_t>.

=item int B<pth_suspend>(pth_t I<tid>);

This suspends a thread I<tid> until it is manually resumed again via
pth_resume(3). For this, the thread is moved to the B<SUSPENDED> queue
and this way is completely out of the scheduler's event handling and
thread dispatching scope. Suspending the current thread is not allowed.
The function returns C<TRUE> on success and C<FALSE> on errors.

=item int B<pth_resume>(pth_t I<tid>);

This function resumes a previously suspended thread I<tid>, i.e. I<tid>
has to stay on the B<SUSPENDED> queue. The thread is moved to the
B<NEW>, B<READY> or B<WAITING> queue (dependent on what its state was
when the pth_suspend(3) call were made) and this way again enters the
event handling and thread dispatching scope of the scheduler. The
function returns C<TRUE> on success and C<FALSE> on errors.

=item int B<pth_raise>(pth_t I<tid>, int I<sig>)

This function raises a signal for delivery to thread I<tid> only.  When one
just raises a signal via raise(3) or kill(2), its delivered to an arbitrary
thread which has this signal not blocked.  With pth_raise(3) one can send a
signal to a thread and its guarantees that only this thread gets the signal
delivered. But keep in mind that nevertheless the signals I<action> is still
configured I<process>-wide.  When I<sig> is 0 plain thread checking is
performed, i.e., `C<pth_raise(tid, 0)>' returns C<TRUE> when thread I<tid>
still exists in the B<PTH> system but doesn't send any signal to it.

=item int B<pth_yield>(pth_t I<tid>);

This explicitly yields back the execution control to the scheduler thread.
Usually the execution is implicitly transferred back to the scheduler when a
thread waits for an event. But when a thread has to do larger CPU bursts, it
can be reasonable to interrupt it explicitly by doing a few pth_yield(3) calls
to give other threads a chance to execute, too.  This obviously is the
cooperating part of B<Pth>.  A thread I<has not> to yield execution, of
course. But when you want to program a server application with good response
times the threads should be cooperative, i.e., when they should split their CPU
bursts into smaller units with this call.

Usually one specifies I<tid> as C<NULL> to indicate to the scheduler that it
can freely decide which thread to dispatch next.  But if one wants to indicate
to the scheduler that a particular thread should be favored on the next
dispatching step, one can specify this thread explicitly. This allows the
usage of the old concept of I<coroutines> where a thread/routine switches to a
particular cooperating thread. If I<tid> is not C<NULL> and points to a I<new>
or I<ready> thread, it is guaranteed that this thread receives execution
control on the next dispatching step. If I<tid> is in a different state (that
is, not in C<PTH_STATE_NEW> or C<PTH_STATE_READY>) an error is reported.

The function usually returns C<TRUE> for success and only C<FALSE> (with
C<errno> set to C<EINVAL>) if I<tid> specified an invalid or still not
new or ready thread.

=item int B<pth_nap>(pth_time_t I<naptime>);

This functions suspends the execution of the current thread until I<naptime>
is elapsed. I<naptime> is of type C<pth_time_t> and this way has theoretically
a resolution of one microsecond. In practice you should neither rely on this
nor that the thread is awakened exactly after I<naptime> has elapsed. It's
only guarantees that the thread will sleep at least I<naptime>. But because
of the non-preemptive nature of B<Pth> it can last longer (when another thread
kept the CPU for a long time). Additionally the resolution is dependent of the
implementation of timers by the operating system and these usually have only a
resolution of 10 microseconds or larger. But usually this isn't important for
an application unless it tries to use this facility for real time tasks.

=item int B<pth_wait>(pth_event_t I<ev>);

This is the link between the scheduler and the event facility (see below for
the various pth_event_xxx() functions). It's modeled like select(2), i.e., one
gives this function one or more events (in the event ring specified by I<ev>)
on which the current thread wants to wait. The scheduler awakes the
thread when one ore more of them occurred or failed after tagging
them as such. The I<ev> argument is a I<pointer> to an event ring
which isn't changed except for the tagging. pth_wait(3) returns the
number of occurred or failed events and the application can use
pth_event_status(3) to test which events occurred or failed.

=item int B<pth_cancel>(pth_t I<tid>);

This cancels a thread I<tid>. How the cancellation is done depends on the
cancellation state of I<tid> which the thread can configure itself. When its
state is C<PTH_CANCEL_DISABLE> a cancellation request is just made pending.
When it is C<PTH_CANCEL_ENABLE> it depends on the cancellation type what is
performed. When its C<PTH_CANCEL_DEFERRED> again the cancellation request is
just made pending. But when its C<PTH_CANCEL_ASYNCHRONOUS> the thread is
immediately canceled before pth_cancel(3) returns. The effect of a thread
cancellation is equal to implicitly forcing the thread to call
`C<pth_exit(PTH_CANCELED)>' at one of his cancellation points.  In B<Pth>
thread enter a cancellation point either explicitly via pth_cancel_point(3) or
implicitly by waiting for an event.

=item int B<pth_abort>(pth_t I<tid>);

This is the cruel way to cancel a thread I<tid>. When it's already dead and
waits to be joined it just joins it (via `C<pth_join(>I<tid>C<, NULL)>') and
this way kicks it out of the system.  Else it forces the thread to be not
joinable and to allow asynchronous cancellation and then cancels it via
`C<pth_cancel(>I<tid>C<)>'.

=item int B<pth_join>(pth_t I<tid>, void **I<value>);

This joins the current thread with the thread specified via I<tid>.
It first suspends the current thread until the I<tid> thread has
terminated. Then it is awakened and stores the value of I<tid>'s
pth_exit(3) call into *I<value> (if I<value> and not C<NULL>) and
returns to the caller. A thread can be joined only when it has the
attribute C<PTH_ATTR_JOINABLE> set to C<TRUE> (the default). A thread
can only be joined once, i.e., after the pth_join(3) call the thread
I<tid> is completely removed from the system.

=item void B<pth_exit>(void *I<value>);

This terminates the current thread. Whether it's immediately removed
from the system or inserted into the dead queue of the scheduler depends
on its join type which was specified at spawning time. If it has the
attribute C<PTH_ATTR_JOINABLE> set to C<FALSE>, it's immediately removed
and I<value> is ignored. Else the thread is inserted into the dead queue
and I<value> remembered for a subsequent pth_join(3) call by another
thread.

=back

=head2 Utilities

Utility functions.

=over 4

=item int B<pth_fdmode>(int I<fd>, int I<mode>);

This switches the non-blocking mode flag on file descriptor I<fd>.  The
argument I<mode> can be C<PTH_FDMODE_BLOCK> for switching I<fd> into blocking
I/O mode, C<PTH_FDMODE_NONBLOCK> for switching I<fd> into non-blocking I/O
mode or C<PTH_FDMODE_POLL> for just polling the current mode. The current mode
is returned (either C<PTH_FDMODE_BLOCK> or C<PTH_FDMODE_NONBLOCK>) or
C<PTH_FDMODE_ERROR> on error. Keep in mind that since B<Pth> 1.1 there is no
longer a requirement to manually switch a file descriptor into non-blocking
mode in order to use it. This is automatically done temporarily inside B<Pth>.
Instead when you now switch a file descriptor explicitly into non-blocking
mode, pth_read(3) or pth_write(3) will never block the current thread.

=item pth_time_t B<pth_time>(long I<sec>, long I<usec>);

This is a constructor for a C<pth_time_t> structure which is a convenient
function to avoid temporary structure values. It returns a I<pth_time_t>
structure which holds the absolute time value specified by I<sec> and I<usec>.

=item pth_time_t B<pth_timeout>(long I<sec>, long I<usec>);

This is a constructor for a C<pth_time_t> structure which is a convenient
function to avoid temporary structure values.  It returns a I<pth_time_t>
structure which holds the absolute time value calculated by adding I<sec> and
I<usec> to the current time.

=item Sfdisc_t *B<pth_sfiodisc>(void);

This functions is always available, but only reasonably usable when B<Pth>
was built with B<Sfio> support (C<--with-sfio> option) and C<PTH_EXT_SFIO> is
then defined by C<pth.h>. It is useful for applications which want to use the
comprehensive B<Sfio> I/O library with the B<Pth> threading library. Then this
function can be used to get an B<Sfio> discipline structure (C<Sfdisc_t>)
which can be pushed onto B<Sfio> streams (C<Sfio_t>) in order to let this
stream use pth_read(3)/pth_write(2) instead of read(2)/write(2). The benefit
is that this way I/O on the B<Sfio> stream does only block the current thread
instead of the whole process. The application has to free(3) the C<Sfdisc_t>
structure when it is no longer needed. The Sfio package can be found at
http://www.research.att.com/sw/tools/sfio/.

=back

=head2 Cancellation Management

B<Pth> supports POSIX style thread cancellation via pth_cancel(3) and the
following two related functions:

=over 4

=item void B<pth_cancel_state>(int I<newstate>, int *I<oldstate>);

This manages the cancellation state of the current thread.  When I<oldstate>
is not C<NULL> the function stores the old cancellation state under the
variable pointed to by I<oldstate>. When I<newstate> is not C<0> it sets the
new cancellation state. I<oldstate> is created before I<newstate> is set.  A
state is a combination of C<PTH_CANCEL_ENABLE> or C<PTH_CANCEL_DISABLE> and
C<PTH_CANCEL_DEFERRED> or C<PTH_CANCEL_ASYNCHRONOUS>.
C<PTH_CANCEL_ENABLE|PTH_CANCEL_DEFERRED> (or C<PTH_CANCEL_DEFAULT>) is the
default state where cancellation is possible but only at cancellation points.
Use C<PTH_CANCEL_DISABLE> to complete disable cancellation for a thread and
C<PTH_CANCEL_ASYNCHRONOUS> for allowing asynchronous cancellations, i.e.,
cancellations which can happen at any time.

=item void B<pth_cancel_point>(void);

This explicitly enter a cancellation point. When the current cancellation
state is C<PTH_CANCEL_DISABLE> or no cancellation request is pending, this has
no side-effect and returns immediately. Else it calls
`C<pth_exit(PTH_CANCELED)>'.

=back

=head2 Event Handling

B<Pth> has a very flexible event facility which is linked into the scheduler
through the pth_wait(3) function. The following functions provide the handling
of event rings.

=over 4

=item pth_event_t B<pth_event>(unsigned long I<spec>, ...);

This creates a new event ring consisting of a single initial event.  The type
of the generated event is specified by I<spec>. The following types are
available:

=over 4

=item C<PTH_EVENT_FD>

This is a file descriptor event. One or more of C<PTH_UNTIL_FD_READABLE>,
C<PTH_UNTIL_FD_WRITEABLE> or C<PTH_UNTIL_FD_EXCEPTION> have to be OR-ed into
I<spec> to specify on which state of the file descriptor you want to wait.  The
file descriptor itself has to be given as an additional argument.  Example:
`C<pth_event(PTH_EVENT_FD|PTH_UNTIL_FD_READABLE, fd)>'.

=item C<PTH_EVENT_SELECT>

This is a multiple file descriptor event modeled directly after the select(2)
call (actually it is also used to implement pth_select(3) internally).  It's a
convenient way to wait for a large set of file descriptors at once and at each
file descriptor for a different type of state. Additionally as a nice
side-effect one receives the number of file descriptors which causes the event
to be occurred (using BSD semantics, i.e., when a file descriptor occurred in
two sets it's counted twice). The arguments correspond directly to the
select(2) function arguments except that there is no timeout argument (because
timeouts already can be handled via C<PTH_EVENT_TIME> events).

Example: `C<pth_event(PTH_EVENT_SELECT, &rc, nfd, rfds, wfds, efds)>' where
C<rc> has to be of type `C<int *>', C<nfd> has to be of type `C<int>' and
C<rfds>, C<wfds> and C<efds> have to be of type `C<fd_set *>' (see
select(2)). The number of occurred file descriptors are stored in C<rc>.

=item C<PTH_EVENT_SIGS>

This is a signal set event. The two additional arguments have to be a pointer
to a signal set (type `C<sigset_t *>') and a pointer to a signal number
variable (type `C<int *>').  This event waits until one of the signals in
the signal set occurred.  As a result the occurred signal number is stored in
the second additional argument. Keep in mind that the B<Pth> scheduler doesn't
block signals automatically.  So when you want to wait for a signal with this
event you've to block it via sigprocmask(2) or it will be delivered without
your notice. Example: `C<sigemptyset(&set); sigaddset(&set, SIGINT);
pth_event(PTH_EVENT_SIG, &set, &sig);>'.

=item C<PTH_EVENT_TIME>

This is a time point event. The additional argument has to be of type
C<pth_time_t> (usually on-the-fly generated via pth_time(3)). This events
waits until the specified time point has elapsed. Keep in mind that the value
is an absolute time point and not an offset. When you want to wait for a
specified amount of time, you've to add the current time to the offset
(usually on-the-fly achieved via pth_timeout(3)).  Example:
`C<pth_event(PTH_EVENT_TIME, pth_timeout(2,0))>'.

=item C<PTH_EVENT_MSG>

This is a message port event. The additional argument has to be of type
C<pth_msgport_t>. This events waits until one or more messages were received
on the specified message port.  Example: `C<pth_event(PTH_EVENT_MSG, mp)>'.

=item C<PTH_EVENT_TID>

This is a thread event. The additional argument has to be of type C<pth_t>.
One of C<PTH_UNTIL_TID_NEW>, C<PTH_UNTIL_TID_READY>, C<PTH_UNTIL_TID_WAITING>
or C<PTH_UNTIL_TID_DEAD> has to be OR-ed into I<spec> to specify on which
state of the thread you want to wait.  Example:
`C<pth_event(PTH_EVENT_TID|PTH_UNTIL_TID_DEAD, tid)>'.

=item C<PTH_EVENT_FUNC>

This is a custom callback function event. Three additional arguments
have to be given with the following types: `C<int (*)(void *)>',
`C<void *>' and `C<pth_time_t>'. The first is a function pointer to
a check function and the second argument is a user-supplied context
value which is passed to this function. The scheduler calls this
function on a regular basis (on his own scheduler stack, so be very
careful!) and the thread is kept sleeping while the function returns
C<FALSE>. Once it returned C<TRUE> the thread will be awakened. The
check interval is defined by the third argument, i.e., the check
function is polled again not until this amount of time elapsed. Example:
`C<pth_event(PTH_EVENT_FUNC, func, arg, pth_time(0,500000))>'.

=back

=item unsigned long B<pth_event_typeof>(pth_event_t I<ev>);

This returns the type of event I<ev>. It's a combination of the describing
C<PTH_EVENT_XX> and C<PTH_UNTIL_XX> value. This is especially useful to know
which arguments have to be supplied to the pth_event_extract(3) function.

=item int B<pth_event_extract>(pth_event_t I<ev>, ...);

When pth_event(3) is treated like sprintf(3), then this function is
sscanf(3), i.e., it is the inverse operation of pth_event(3). This means that
it can be used to extract the ingredients of an event.  The ingredients are
stored into variables which are given as pointers on the variable argument
list.  Which pointers have to be present depends on the event type and has to
be determined by the caller before via pth_event_typeof(3).

To make it clear, when you constructed I<ev> via `C<ev =
pth_event(PTH_EVENT_FD, fd);>' you have to extract it via
`C<pth_event_extract(ev, &fd)>', etc. For multiple arguments of an event the
order of the pointer arguments is the same as for pth_event(3). But always
keep in mind that you have to always supply I<pointers> to I<variables> and
these variables have to be of the same type as the argument of pth_event(3)
required.

=item pth_event_t B<pth_event_concat>(pth_event_t I<ev>, ...);

This concatenates one or more additional event rings to the event ring I<ev>
and returns I<ev>. The end of the argument list has to be marked with a
C<NULL> argument. Use this function to create real events rings out of the
single-event rings created by pth_event(3).

=item pth_event_t B<pth_event_isolate>(pth_event_t I<ev>);

This isolates the event I<ev> from possibly appended events in the event ring.
When in I<ev> only one event exists, this returns C<NULL>. When remaining
events exists, they form a new event ring which is returned.

=item pth_event_t B<pth_event_walk>(pth_event_t I<ev>, int I<direction>);

This walks to the next (when I<direction> is C<PTH_WALK_NEXT>) or previews
(when I<direction> is C<PTH_WALK_PREV>) event in the event ring I<ev> and
returns this new reached event. Additionally C<PTH_UNTIL_OCCURRED> can be
OR-ed into I<direction> to walk to the next/previous occurred event in the
ring I<ev>.

=item pth_status_t B<pth_event_status>(pth_event_t I<ev>);

This returns the status of event I<ev>. This is a fast operation
because only a tag on I<ev> is checked which was either set or still
not set by the scheduler. In other words: This doesn't check the
event itself, it just checks the last knowledge of the scheduler. The
possible returned status codes are: C<PTH_STATUS_PENDING> (event is
still pending), C<PTH_STATUS_OCCURRED> (event successfully occurred),
C<PTH_STATUS_FAILED> (event failed).

=item int B<pth_event_free>(pth_event_t I<ev>, int I<mode>);

This deallocates the event I<ev> (when I<mode> is C<PTH_FREE_THIS>) or all
events appended to the event ring under I<ev> (when I<mode> is
C<PTH_FREE_ALL>).

=back

=head2 Key-Based Storage

The following functions provide thread-local storage through unique keys
similar to the POSIX B<Pthread> API. Use this for thread specific global data.

=over 4

=item int B<pth_key_create>(pth_key_t *I<key>, void (*I<func>)(void *));

This created a new unique key and stores it in I<key>.  Additionally I<func>
can specify a destructor function which is called on the current threads
termination with the I<key>.

=item int B<pth_key_delete>(pth_key_t I<key>);

This explicitly destroys a key I<key>.

=item int B<pth_key_setdata>(pth_key_t I<key>, const void *I<value>);

This stores I<value> under I<key>.

=item void *B<pth_key_getdata>(pth_key_t I<key>);

This retrieves the value under I<key>.

=back

=head2 Message Port Communication

The following functions provide message ports which can be used for efficient
and flexible inter-thread communication.

=over 4

=item pth_msgport_t B<pth_msgport_create>(const char *I<name>);

This returns a pointer to a new message port. If name I<name>
is not C<NULL>, the I<name> can be used by other threads via
pth_msgport_find(3) to find the message port in case they do not know
directly the pointer to the message port.

=item void B<pth_msgport_destroy>(pth_msgport_t I<mp>);

This destroys a message port I<mp>. Before all pending messages on it are
replied to their origin message port.

=item pth_msgport_t B<pth_msgport_find>(const char *I<name>);

This finds a message port in the system by I<name> and returns the pointer to
it.

=item int B<pth_msgport_pending>(pth_msgport_t I<mp>);

This returns the number of pending messages on message port I<mp>.

=item int B<pth_msgport_put>(pth_msgport_t I<mp>, pth_message_t *I<m>);

This puts (or sends) a message I<m> to message port I<mp>.

=item pth_message_t *B<pth_msgport_get>(pth_msgport_t I<mp>);

This gets (or receives) the top message from message port I<mp>.  Incoming
messages are always kept in a queue, so there can be more pending messages, of
course.

=item int B<pth_msgport_reply>(pth_message_t *I<m>);

This replies a message I<m> to the message port of the sender.

=back

=head2 Thread Cleanups

Per-thread cleanup functions.

=over 4

=item int B<pth_cleanup_push>(void (*I<handler>)(void *), void *I<arg>);

This pushes the routine I<handler> onto the stack of cleanup routines for the
current thread.  These routines are called in LIFO order when the thread
terminates.

=item int B<pth_cleanup_pop>(int I<execute>);

This pops the top-most routine from the stack of cleanup routines for the
current thread. When I<execute> is C<TRUE> the routine is additionally called.

=back

=head2 Process Forking

The following functions provide some special support for process forking
situations inside the threading environment.

=over 4

=item int B<pth_atfork_push>(void (*I<prepare>)(void *), void (*)(void *I<parent>), void (*)(void *I<child>), void *I<arg>);

This function declares forking handlers to be called before and after
pth_fork(3), in the context of the thread that called pth_fork(3). The
I<prepare> handler is called before fork(2) processing commences. The
I<parent> handler is called   after fork(2) processing completes in the parent
process.  The I<child> handler is called after fork(2) processing completed in
the child process. If no handling is desired at one or more of these three
points, the corresponding handler can be given as C<NULL>.  Each handler is
called with I<arg> as the argument.

The order of calls to pth_atfork_push(3) is significant. The I<parent> and
I<child> handlers are called in the order in which they were established by
calls to pth_atfork_push(3), i.e., FIFO. The I<prepare> fork handlers are
called in the opposite order, i.e., LIFO.

=item int B<pth_atfork_pop>(void);

This removes the top-most handlers on the forking handler stack which were
established with the last pth_atfork_push(3) call. It returns C<FALSE> when no
more handlers couldn't be removed from the stack.

=item pid_t B<pth_fork>(void);

This is a variant of fork(2) with the difference that the current thread only
is forked into a separate process, i.e., in the parent process nothing changes
while in the child process all threads are gone except for the scheduler and
the calling thread. When you really want to duplicate all threads in the
current process you should use fork(2) directly. But this is usually not
reasonable. Additionally this function takes care of forking handlers as
established by pth_fork_push(3).

=back

=head2 Synchronization

The following functions provide synchronization support via mutual exclusion
locks (B<mutex>), read-write locks (B<rwlock>), condition variables (B<cond>)
and barriers (B<barrier>). Keep in mind that in a non-preemptive threading
system like B<Pth> this might sound unnecessary at the first look, because a
thread isn't interrupted by the system. Actually when you have a critical code
section which doesn't contain any pth_xxx() functions, you don't need any
mutex to protect it, of course.

But when your critical code section contains any pth_xxx() function the chance
is high that these temporarily switch to the scheduler. And this way other
threads can make progress and enter your critical code section, too.  This is
especially true for critical code sections which implicitly or explicitly use
the event mechanism.

=over 4

=item int B<pth_mutex_init>(pth_mutex_t *I<mutex>);

This dynamically initializes a mutex variable of type `C<pth_mutex_t>'.
Alternatively one can also use static initialization via `C<pth_mutex_t
mutex = PTH_MUTEX_INIT>'.

=item int B<pth_mutex_acquire>(pth_mutex_t *I<mutex>, int I<try>, pth_event_t I<ev>);

This acquires a mutex I<mutex>.  If the mutex is already locked by another
thread, the current threads execution is suspended until the mutex is unlocked
again or additionally the extra events in I<ev> occurred (when I<ev> is not
C<NULL>).  Recursive locking is explicitly supported, i.e., a thread is allowed
to acquire a mutex more than once before its released. But it then also has be
released the same number of times until the mutex is again lockable by others.
When I<try> is C<TRUE> this function never suspends execution. Instead it
returns C<FALSE> with C<errno> set to C<EBUSY>.

=item int B<pth_mutex_release>(pth_mutex_t *I<mutex>);

This decrements the recursion locking count on I<mutex> and when it is zero it
releases the mutex I<mutex>.

=item int B<pth_rwlock_init>(pth_rwlock_t *I<rwlock>);

This dynamically initializes a read-write lock variable of type
`C<pth_rwlock_t>'.  Alternatively one can also use static initialization
via `C<pth_rwlock_t rwlock = PTH_RWLOCK_INIT>'.

=item int B<pth_rwlock_acquire>(pth_rwlock_t *I<rwlock>, int I<op>, int I<try>, pth_event_t I<ev>);

This acquires a read-only (when I<op> is C<PTH_RWLOCK_RD>) or a read-write
(when I<op> is C<PTH_RWLOCK_RW>) lock I<rwlock>. When the lock is only locked
by other threads in read-only mode, the lock succeeds.  But when one thread
holds a read-write lock, all locking attempts suspend the current thread until
this lock is released again. Additionally in I<ev> events can be given to let
the locking timeout, etc. When I<try> is C<TRUE> this function never suspends
execution. Instead it returns C<FALSE> with C<errno> set to C<EBUSY>.

=item int B<pth_rwlock_release>(pth_rwlock_t *I<rwlock>);

This releases a previously acquired (read-only or read-write) lock.

=item int B<pth_cond_init>(pth_cond_t *I<cond>);

This dynamically initializes a condition variable variable of type
`C<pth_cond_t>'.  Alternatively one can also use static initialization via
`C<pth_cond_t cond = PTH_COND_INIT>'.

=item int B<pth_cond_await>(pth_cond_t *I<cond>, pth_mutex_t *I<mutex>, pth_event_t I<ev>);

This awaits a condition situation. The caller has to follow the semantics of
the POSIX condition variables: I<mutex> has to be acquired before this
function is called. The execution of the current thread is then suspended
either until the events in I<ev> occurred (when I<ev> is not C<NULL>) or
I<cond> was notified by another thread via pth_cond_notify(3).  While the
thread is waiting, I<mutex> is released. Before it returns I<mutex> is
reacquired.

=item int B<pth_cond_notify>(pth_cond_t *I<cond>, int I<broadcast>);

This notified one or all threads which are waiting on I<cond>.  When
I<broadcast> is C<TRUE> all thread are notified, else only a single
(unspecified) one.

=item int B<pth_barrier_init>(pth_barrier_t *I<barrier>, int I<threshold>);

This dynamically initializes a barrier variable of type `C<pth_barrier_t>'.
Alternatively one can also use static initialization via `C<pth_barrier_t
barrier = PTH_BARRIER_INIT(>I<threadhold>C<)>'.

=item int B<pth_barrier_reach>(pth_barrier_t *I<barrier>);

This function reaches a barrier I<barrier>. If this is the last thread (as
specified by I<threshold> on init of I<barrier>) all threads are awakened.
Else the current thread is suspended until the last thread reached the barrier
and this way awakes all threads. The function returns (beside C<FALSE> on
error) the value C<TRUE> for any thread which neither reached the barrier as
the first nor the last thread; C<PTH_BARRIER_HEADLIGHT> for the thread which
reached the barrier as the first thread and C<PTH_BARRIER_TAILLIGHT> for the
thread which reached the barrier as the last thread.

=back

=head2 User-Space Context

The following functions provide a stand-alone sub-API for user-space
context switching. It internally is based on the same underlying machine
context switching mechanism the threads in B<GNU Pth> are based on.
Hence these functions you can use for implementing your own simple
user-space threads. The C<pth_uctx_t> context is somewhat modeled after
POSIX ucontext(3).

The time required to create (via pth_uctx_make(3)) a user-space context
can range from just a few microseconds up to a more dramatical time
(depending on the machine context switching method which is available on
the platform). On the other hand, the raw performance in switching the
user-space contexts is always very good (nearly independent of the used
machine context switching method). For instance, on an Intel Pentium-III
CPU with 800Mhz running under FreeBSD 4 one usually achieves about
260,000 user-space context switches (via pth_uctx_switch(3)) per second.

=over 4

=item int B<pth_uctx_create>(pth_uctx_t *I<uctx>);

This function creates a user-space context and stores it into I<uctx>.
There is still no underlying user-space context configured. You still
have to do this with pth_uctx_make(3). On success, this function returns
C<TRUE>, else C<FALSE>.

=item int B<pth_uctx_make>(pth_uctx_t I<uctx>, char *I<sk_addr>, size_t I<sk_size>, const sigset_t *I<sigmask>, void (*I<start_func>)(void *), void *I<start_arg>, pth_uctx_t I<uctx_after>);

This function makes a new user-space context in I<uctx> which will
operate on the run-time stack I<sk_addr> (which is of maximum
size I<sk_size>), with the signals in I<sigmask> blocked (if
I<sigmask> is not C<NULL>) and starting to execute with the call
I<start_func>(I<start_arg>). If I<sk_addr> is C<NULL>, a stack
is dynamically allocated. The stack size I<sk_size> has to be at
least 16384 (16KB). If the start function I<start_func> returns and
I<uctx_after> is not C<NULL>, an implicit user-space context switch
to this context is performed. Else (if I<uctx_after> is C<NULL>) the
process is terminated with exit(3). This function is somewhat modeled
after POSIX makecontext(3). On success, this function returns C<TRUE>,
else C<FALSE>.

=item int B<pth_uctx_switch>(pth_uctx_t I<uctx_from>, pth_uctx_t I<uctx_to>);

This function saves the current user-space context in I<uctx_from> for
later restoring by another call to pth_uctx_switch(3) and restores
the new user-space context from I<uctx_to>, which previously had to
be set with either a previous call to pth_uctx_switch(3) or initially
by pth_uctx_make(3). This function is somewhat modeled after POSIX
swapcontext(3). If I<uctx_from> or I<uctx_to> are C<NULL> or if
I<uctx_to> contains no valid user-space context, C<FALSE> is returned
instead of C<TRUE>. These are the only errors possible.

=item int B<pth_uctx_destroy>(pth_uctx_t I<uctx>);

This function destroys the user-space context in I<uctx>. The run-time
stack associated with the user-space context is deallocated only if it
was not given by the application (see I<sk_addr> of pth_uctx_create(3)).
If I<uctx> is C<NULL>, C<FALSE> is returned instead of C<TRUE>. This is
the only error possible.

=back

=head2 Generalized POSIX Replacement API

The following functions are generalized replacements functions for the POSIX
API, i.e., they are similar to the functions under `B<Standard POSIX
Replacement API>' but all have an additional event argument which can be used
for timeouts, etc.

=over 4

=item int B<pth_sigwait_ev>(const sigset_t *I<set>, int *I<sig>, pth_event_t I<ev>);

This is equal to pth_sigwait(3) (see below), but has an additional event
argument I<ev>. When pth_sigwait(3) suspends the current threads execution it
usually only uses the signal event on I<set> to awake. With this function any
number of extra events can be used to awake the current thread (remember that
I<ev> actually is an event I<ring>).

=item int B<pth_connect_ev>(int I<s>, const struct sockaddr *I<addr>, socklen_t I<addrlen>, pth_event_t I<ev>);

This is equal to pth_connect(3) (see below), but has an additional event
argument I<ev>. When pth_connect(3) suspends the current threads execution it
usually only uses the I/O event on I<s> to awake. With this function any
number of extra events can be used to awake the current thread (remember that
I<ev> actually is an event I<ring>).

=item int B<pth_accept_ev>(int I<s>, struct sockaddr *I<addr>, socklen_t *I<addrlen>, pth_event_t I<ev>);

This is equal to pth_accept(3) (see below), but has an additional event
argument I<ev>. When pth_accept(3) suspends the current threads execution it
usually only uses the I/O event on I<s> to awake. With this function any
number of extra events can be used to awake the current thread (remember that
I<ev> actually is an event I<ring>).

=item int B<pth_select_ev>(int I<nfd>, fd_set *I<rfds>, fd_set *I<wfds>, fd_set *I<efds>, struct timeval *I<timeout>, pth_event_t I<ev>);

This is equal to pth_select(3) (see below), but has an additional event
argument I<ev>. When pth_select(3) suspends the current threads execution it
usually only uses the I/O event on I<rfds>, I<wfds> and I<efds> to awake. With
this function any number of extra events can be used to awake the current
thread (remember that I<ev> actually is an event I<ring>).

=item int B<pth_poll_ev>(struct pollfd *I<fds>, unsigned int I<nfd>, int I<timeout>, pth_event_t I<ev>);

This is equal to pth_poll(3) (see below), but has an additional event argument
I<ev>. When pth_poll(3) suspends the current threads execution it usually only
uses the I/O event on I<fds> to awake. With this function any number of extra
events can be used to awake the current thread (remember that I<ev> actually
is an event I<ring>).

=item ssize_t B<pth_read_ev>(int I<fd>, void *I<buf>, size_t I<nbytes>, pth_event_t I<ev>);

This is equal to pth_read(3) (see below), but has an additional event argument
I<ev>. When pth_read(3) suspends the current threads execution it usually only
uses the I/O event on I<fd> to awake. With this function any number of extra
events can be used to awake the current thread (remember that I<ev> actually
is an event I<ring>).

=item ssize_t B<pth_readv_ev>(int I<fd>, const struct iovec *I<iovec>, int I<iovcnt>, pth_event_t I<ev>);

This is equal to pth_readv(3) (see below), but has an additional event
argument I<ev>. When pth_readv(3) suspends the current threads execution it
usually only uses the I/O event on I<fd> to awake. With this function any
number of extra events can be used to awake the current thread (remember that
I<ev> actually is an event I<ring>).

=item ssize_t B<pth_write_ev>(int I<fd>, const void *I<buf>, size_t I<nbytes>, pth_event_t I<ev>);

This is equal to pth_write(3) (see below), but has an additional event argument
I<ev>. When pth_write(3) suspends the current threads execution it usually
only uses the I/O event on I<fd> to awake. With this function any number of
extra events can be used to awake the current thread (remember that I<ev>
actually is an event I<ring>).

=item ssize_t B<pth_writev_ev>(int I<fd>, const struct iovec *I<iovec>, int I<iovcnt>, pth_event_t I<ev>);

This is equal to pth_writev(3) (see below), but has an additional event
argument I<ev>. When pth_writev(3) suspends the current threads execution it
usually only uses the I/O event on I<fd> to awake. With this function any
number of extra events can be used to awake the current thread (remember that
I<ev> actually is an event I<ring>).

=item ssize_t B<pth_recv_ev>(int I<fd>, void *I<buf>, size_t I<nbytes>, int I<flags>, pth_event_t I<ev>);

This is equal to pth_recv(3) (see below), but has an additional event
argument I<ev>. When pth_recv(3) suspends the current threads execution it
usually only uses the I/O event on I<fd> to awake. With this function any
number of extra events can be used to awake the current thread (remember that
I<ev> actually is an event I<ring>).

=item ssize_t B<pth_recvfrom_ev>(int I<fd>, void *I<buf>, size_t I<nbytes>, int I<flags>, struct sockaddr *I<from>, socklen_t *I<fromlen>, pth_event_t I<ev>);

This is equal to pth_recvfrom(3) (see below), but has an additional event
argument I<ev>. When pth_recvfrom(3) suspends the current threads execution it
usually only uses the I/O event on I<fd> to awake. With this function any
number of extra events can be used to awake the current thread (remember that
I<ev> actually is an event I<ring>).

=item ssize_t B<pth_send_ev>(int I<fd>, const void *I<buf>, size_t I<nbytes>, int I<flags>, pth_event_t I<ev>);

This is equal to pth_send(3) (see below), but has an additional event
argument I<ev>. When pth_send(3) suspends the current threads execution it
usually only uses the I/O event on I<fd> to awake. With this function any
number of extra events can be used to awake the current thread (remember that
I<ev> actually is an event I<ring>).

=item ssize_t B<pth_sendto_ev>(int I<fd>, const void *I<buf>, size_t I<nbytes>, int I<flags>, const struct sockaddr *I<to>, socklen_t I<tolen>, pth_event_t I<ev>);

This is equal to pth_sendto(3) (see below), but has an additional event
argument I<ev>. When pth_sendto(3) suspends the current threads execution it
usually only uses the I/O event on I<fd> to awake. With this function any
number of extra events can be used to awake the current thread (remember that
I<ev> actually is an event I<ring>).

=back

=head2 Standard POSIX Replacement API

The following functions are standard replacements functions for the POSIX API.
The difference is mainly that they suspend the current thread only instead of
the whole process in case the file descriptors will block.

=over 4

=item int B<pth_nanosleep>(const struct timespec *I<rqtp>, struct timespec *I<rmtp>);

This is a variant of the POSIX nanosleep(3) function. It suspends the
current threads execution until the amount of time in I<rqtp> elapsed.
The thread is guaranteed to not wake up before this time, but because
of the non-preemptive scheduling nature of B<Pth>, it can be awakened
later, of course. If I<rmtp> is not C<NULL>, the C<timespec> structure
it references is updated to contain the unslept amount (the request time
minus the time actually slept time). The difference between nanosleep(3)
and pth_nanosleep(3) is that that pth_nanosleep(3) suspends only the
execution of the current thread and not the whole process.

=item int B<pth_usleep>(unsigned int I<usec>);

This is a variant of the 4.3BSD usleep(3) function. It suspends the current
threads execution until I<usec> microseconds (= I<usec>*1/1000000 sec)
elapsed.  The thread is guaranteed to not wake up before this time, but
because of the non-preemptive scheduling nature of B<Pth>, it can be awakened
later, of course.  The difference between usleep(3) and pth_usleep(3) is that
that pth_usleep(3) suspends only the execution of the current thread and not
the whole process.

=item unsigned int B<pth_sleep>(unsigned int I<sec>);

This is a variant of the POSIX sleep(3) function. It suspends the current
threads execution until I<sec> seconds elapsed.  The thread is guaranteed to
not wake up before this time, but because of the non-preemptive scheduling
nature of B<Pth>, it can be awakened later, of course.  The difference between
sleep(3) and pth_sleep(3) is that pth_sleep(3) suspends only the
execution of the current thread and not the whole process.

=item pid_t B<pth_waitpid>(pid_t I<pid>, int *I<status>, int I<options>);

This is a variant of the POSIX waitpid(2) function. It suspends the
current threads execution until I<status> information is available for a
terminated child process I<pid>.  The difference between waitpid(2) and
pth_waitpid(3) is that pth_waitpid(3) suspends only the execution of the
current thread and not the whole process.  For more details about the
arguments and return code semantics see waitpid(2).

=item int B<pth_system>(const char *I<cmd>);

This is a variant of the POSIX system(3) function. It executes the
shell command I<cmd> with Bourne Shell (C<sh>) and suspends the current
threads execution until this command terminates. The difference between
system(3) and pth_system(3) is that pth_system(3) suspends only
the execution of the current thread and not the whole process. For more
details about the arguments and return code semantics see system(3).

=item int B<pth_sigmask>(int I<how>, const sigset_t *I<set>, sigset_t *I<oset>)

This is the B<Pth> thread-related equivalent of POSIX sigprocmask(2) respectively
pthread_sigmask(3). The arguments I<how>, I<set> and I<oset> directly relate
to sigprocmask(2), because B<Pth> internally just uses sigprocmask(2) here. So
alternatively you can also directly call sigprocmask(2), but for consistency
reasons you should use this function pth_sigmask(3).

=item int B<pth_sigwait>(const sigset_t *I<set>, int *I<sig>);

This is a variant of the POSIX.1c sigwait(3) function. It suspends the current
threads execution until a signal in I<set> occurred and stores the signal
number in I<sig>. The important point is that the signal is not delivered to a
signal handler. Instead it's caught by the scheduler only in order to awake
the pth_sigwait() call. The trick and noticeable point here is that this way
you get an asynchronous aware application that is written completely
synchronously. When you think about the problem of I<asynchronous safe>
functions you should recognize that this is a great benefit.

=item int B<pth_connect>(int I<s>, const struct sockaddr *I<addr>, socklen_t I<addrlen>);

This is a variant of the 4.2BSD connect(2) function. It establishes a
connection on a socket I<s> to target specified in I<addr> and I<addrlen>.
The difference between connect(2) and pth_connect(3) is that
pth_connect(3) suspends only the execution of the current thread and not the
whole process.  For more details about the arguments and return code semantics
see connect(2).

=item int B<pth_accept>(int I<s>, struct sockaddr *I<addr>, socklen_t *I<addrlen>);

This is a variant of the 4.2BSD accept(2) function. It accepts a connection on
a socket by extracting the first connection request on the queue of pending
connections, creating a new socket with the same properties of I<s> and
allocates a new file descriptor for the socket (which is returned).  The
difference between accept(2) and pth_accept(3) is that pth_accept(3)
suspends only the execution of the current thread and not the whole process.
For more details about the arguments and return code semantics see accept(2).

=item int B<pth_select>(int I<nfd>, fd_set *I<rfds>, fd_set *I<wfds>, fd_set *I<efds>, struct timeval *I<timeout>);

This is a variant of the 4.2BSD select(2) function.  It examines the I/O
descriptor sets whose addresses are passed in I<rfds>, I<wfds>, and I<efds> to
see if some of their descriptors are ready for reading, are ready for writing,
or have an exceptional condition pending, respectively.  For more details
about the arguments and return code semantics see select(2).

=item int B<pth_pselect>(int I<nfd>, fd_set *I<rfds>, fd_set *I<wfds>, fd_set *I<efds>, const struct timespec *I<timeout>, const sigset_t *I<sigmask>);

This is a variant of the POSIX pselect(2) function, which in turn
is a stronger variant of 4.2BSD select(2). The difference is that
the higher-resolution C<struct timespec> is passed instead of the
lower-resolution C<struct timeval> and that a signal mask is specified
which is temporarily set while waiting for input. For more details about
the arguments and return code semantics see pselect(2) and select(2).

=item int B<pth_poll>(struct pollfd *I<fds>, unsigned int I<nfd>, int I<timeout>);

This is a variant of the SysV poll(2) function. It examines the I/O
descriptors which are passed in the array I<fds> to see if some of them are
ready for reading, are ready for writing, or have an exceptional condition
pending, respectively. For more details about the arguments and return code
semantics see poll(2).

=item ssize_t B<pth_read>(int I<fd>, void *I<buf>, size_t I<nbytes>);

This is a variant of the POSIX read(2) function. It reads up to I<nbytes>
bytes into I<buf> from file descriptor I<fd>.  The difference between read(2)
and pth_read(2) is that pth_read(2) suspends execution of the current
thread until the file descriptor is ready for reading. For more details about
the arguments and return code semantics see read(2).

=item ssize_t B<pth_readv>(int I<fd>, const struct iovec *I<iovec>, int I<iovcnt>);

This is a variant of the POSIX readv(2) function. It reads data from
file descriptor I<fd> into the first I<iovcnt> rows of the I<iov> vector.  The
difference between readv(2) and pth_readv(2) is that pth_readv(2)
suspends execution of the current thread until the file descriptor is ready for
reading. For more details about the arguments and return code semantics see
readv(2).

=item ssize_t B<pth_write>(int I<fd>, const void *I<buf>, size_t I<nbytes>);

This is a variant of the POSIX write(2) function. It writes I<nbytes> bytes
from I<buf> to file descriptor I<fd>.  The difference between write(2) and
pth_write(2) is that pth_write(2) suspends execution of the current
thread until the file descriptor is ready for writing.  For more details about
the arguments and return code semantics see write(2).

=item ssize_t B<pth_writev>(int I<fd>, const struct iovec *I<iovec>, int I<iovcnt>);

This is a variant of the POSIX writev(2) function. It writes data to
file descriptor I<fd> from the first I<iovcnt> rows of the I<iov> vector.  The
difference between writev(2) and pth_writev(2) is that pth_writev(2)
suspends execution of the current thread until the file descriptor is ready for
reading. For more details about the arguments and return code semantics see
writev(2).

=item ssize_t B<pth_pread>(int I<fd>, void *I<buf>, size_t I<nbytes>, off_t I<offset>);

This is a variant of the POSIX pread(3) function.  It performs the same action
as a regular read(2), except that it reads from a given position in the file
without changing the file pointer.  The first three arguments are the same as
for pth_read(3) with the addition of a fourth argument I<offset> for the
desired position inside the file.

=item ssize_t B<pth_pwrite>(int I<fd>, const void *I<buf>, size_t I<nbytes>, off_t I<offset>);

This is a variant of the POSIX pwrite(3) function.  It performs the same
action as a regular write(2), except that it writes to a given position in the
file without changing the file pointer. The first three arguments are the same
as for pth_write(3) with the addition of a fourth argument I<offset> for the
desired position inside the file.

=item ssize_t B<pth_recv>(int I<fd>, void *I<buf>, size_t I<nbytes>, int I<flags>);

This is a variant of the SUSv2 recv(2) function and equal to
``pth_recvfrom(fd, buf, nbytes, flags, NULL, 0)''.

=item ssize_t B<pth_recvfrom>(int I<fd>, void *I<buf>, size_t I<nbytes>, int I<flags>, struct sockaddr *I<from>, socklen_t *I<fromlen>);

This is a variant of the SUSv2 recvfrom(2) function. It reads up to
I<nbytes> bytes into I<buf> from file descriptor I<fd> while using
I<flags> and I<from>/I<fromlen>. The difference between recvfrom(2) and
pth_recvfrom(2) is that pth_recvfrom(2) suspends execution of the
current thread until the file descriptor is ready for reading. For more
details about the arguments and return code semantics see recvfrom(2).

=item ssize_t B<pth_send>(int I<fd>, const void *I<buf>, size_t I<nbytes>, int I<flags>);

This is a variant of the SUSv2 send(2) function and equal to
``pth_sendto(fd, buf, nbytes, flags, NULL, 0)''.

=item ssize_t B<pth_sendto>(int I<fd>, const void *I<buf>, size_t I<nbytes>, int I<flags>, const struct sockaddr *I<to>, socklen_t I<tolen>);

This is a variant of the SUSv2 sendto(2) function. It writes I<nbytes>
bytes from I<buf> to file descriptor I<fd> while using I<flags> and
I<to>/I<tolen>. The difference between sendto(2) and pth_sendto(2) is
that pth_sendto(2) suspends execution of the current thread until
the file descriptor is ready for writing. For more details about the
arguments and return code semantics see sendto(2).

=back

=head1 EXAMPLE

The following example is a useless server which does nothing more than
listening on TCP port 12345 and displaying the current time to the
socket when a connection was established. For each incoming connection a
thread is spawned. Additionally, to see more multithreading, a useless
ticker thread runs simultaneously which outputs the current time to
C<stderr> every 5 seconds. The example contains I<no> error checking and
is I<only> intended to show you the look and feel of B<Pth>.

 #include <stdio.h>
 #include <stdlib.h>
 #include <errno.h>
 #include <sys/types.h>
 #include <sys/socket.h>
 #include <netinet/in.h>
 #include <arpa/inet.h>
 #include <signal.h>
 #include <netdb.h>
 #include <unistd.h>
 #include "pth.h"

 #define PORT 12345

 /* the socket connection handler thread */
 static void *handler(void *_arg)
 {
     int fd = (int)_arg;
     time_t now;
     char *ct;

     now = time(NULL);
     ct = ctime(&now);
     pth_write(fd, ct, strlen(ct));
     close(fd);
     return NULL;
 }

 /* the stderr time ticker thread */
 static void *ticker(void *_arg)
 {
     time_t now;
     char *ct;
     float load;

     for (;;) {
         pth_sleep(5);
         now = time(NULL);
         ct = ctime(&now);
         ct[strlen(ct)-1] = '\0';
         pth_ctrl(PTH_CTRL_GETAVLOAD, &load);
         printf("ticker: time: %s, average load: %.2f\n", ct, load);
     }
 }

 /* the main thread/procedure */
 int main(int argc, char *argv[])
 {
     pth_attr_t attr;
     struct sockaddr_in sar;
     struct protoent *pe;
     struct sockaddr_in peer_addr;
     int peer_len;
     int sa, sw;
     int port;

     pth_init();
     signal(SIGPIPE, SIG_IGN);

     attr = pth_attr_new();
     pth_attr_set(attr, PTH_ATTR_NAME, "ticker");
     pth_attr_set(attr, PTH_ATTR_STACK_SIZE, 64*1024);
     pth_attr_set(attr, PTH_ATTR_JOINABLE, FALSE);
     pth_spawn(attr, ticker, NULL);

     pe = getprotobyname("tcp");
     sa = socket(AF_INET, SOCK_STREAM, pe->p_proto);
     sar.sin_family = AF_INET;
     sar.sin_addr.s_addr = INADDR_ANY;
     sar.sin_port = htons(PORT);
     bind(sa, (struct sockaddr *)&sar, sizeof(struct sockaddr_in));
     listen(sa, 10);

     pth_attr_set(attr, PTH_ATTR_NAME, "handler");
     for (;;) {
         peer_len = sizeof(peer_addr);
         sw = pth_accept(sa, (struct sockaddr *)&peer_addr, &peer_len);
         pth_spawn(attr, handler, (void *)sw);
     }
 }

=head1 BUILD ENVIRONMENTS

In this section we will discuss the canonical ways to establish the build
environment for a B<Pth> based program. The possibilities supported by B<Pth>
range from very simple environments to rather complex ones.

=head2 Manual Build Environment (Novice)

As a first example, assume we have the above test program staying in the
source file C<foo.c>. Then we can create a very simple build environment by
just adding the following C<Makefile>:

 $ vi Makefile
 | CC      = cc
 | CFLAGS  = `pth-config --cflags`
 | LDFLAGS = `pth-config --ldflags`
 | LIBS    = `pth-config --libs`
 |
 | all: foo
 | foo: foo.o
 |     $(CC) $(LDFLAGS) -o foo foo.o $(LIBS)
 | foo.o: foo.c
 |     $(CC) $(CFLAGS) -c foo.c
 | clean:
 |     rm -f foo foo.o

This imports the necessary compiler and linker flags on-the-fly from the
B<Pth> installation via its C<pth-config> program. This approach is
straight-forward and works fine for small projects.

=head2 Autoconf Build Environment (Advanced)

The previous approach is simple but inflexible. First, to speed up
building, it would be nice to not expand the compiler and linker flags
every time the compiler is started. Second, it would be useful to
also be able to build against uninstalled B<Pth>, that is, against
a B<Pth> source tree which was just configured and built, but not
installed. Third, it would be also useful to allow checking of the
B<Pth> version to make sure it is at least a minimum required version.
And finally, it would be also great to make sure B<Pth> works correctly
by first performing some sanity compile and run-time checks. All this
can be done if we use GNU B<autoconf> and the C<AC_CHECK_PTH> macro
provided by B<Pth>. For this, we establish the following three files:

First we again need the C<Makefile>, but this time it contains B<autoconf>
placeholders and additional cleanup targets. And we create it under the name
C<Makefile.in>, because it is now an input file for B<autoconf>:

 $ vi Makefile.in
 | CC      = @CC@
 | CFLAGS  = @CFLAGS@
 | LDFLAGS = @LDFLAGS@
 | LIBS    = @LIBS@
 |
 | all: foo
 | foo: foo.o
 |     $(CC) $(LDFLAGS) -o foo foo.o $(LIBS)
 | foo.o: foo.c
 |     $(CC) $(CFLAGS) -c foo.c
 | clean:
 |     rm -f foo foo.o
 | distclean:
 |     rm -f foo foo.o
 |     rm -f config.log config.status config.cache
 |     rm -f Makefile

Because B<autoconf> generates additional files, we added a canonical
C<distclean> target which cleans this up. Secondly, we wrote
C<configure.ac>, a (minimal) B<autoconf> script specification:

 $ vi configure.ac
 | AC_INIT(Makefile.in)
 | AC_CHECK_PTH(1.3.0)
 | AC_OUTPUT(Makefile)

Then we let B<autoconf>'s C<aclocal> program generate for us an C<aclocal.m4>
file containing B<Pth>'s C<AC_CHECK_PTH> macro. Then we generate the final
C<configure> script out of this C<aclocal.m4> file and the C<configure.ac>
file:

 $ aclocal --acdir=`pth-config --acdir`
 $ autoconf

After these steps, the working directory should look similar to this:

 $ ls -l
 -rw-r--r--  1 rse  users    176 Nov  3 11:11 Makefile.in
 -rw-r--r--  1 rse  users  15314 Nov  3 11:16 aclocal.m4
 -rwxr-xr-x  1 rse  users  52045 Nov  3 11:16 configure
 -rw-r--r--  1 rse  users     63 Nov  3 11:11 configure.ac
 -rw-r--r--  1 rse  users   4227 Nov  3 11:11 foo.c

If we now run C<configure> we get a correct C<Makefile> which
immediately can be used to build C<foo> (assuming that B<Pth> is already
installed somewhere, so that C<pth-config> is in C<$PATH>):

 $ ./configure
 creating cache ./config.cache
 checking for gcc... gcc
 checking whether the C compiler (gcc   ) works... yes
 checking whether the C compiler (gcc   ) is a cross-compiler... no
 checking whether we are using GNU C... yes
 checking whether gcc accepts -g... yes
 checking how to run the C preprocessor... gcc -E
 checking for GNU Pth... version 1.3.0, installed under /usr/local
 updating cache ./config.cache
 creating ./config.status
 creating Makefile
 rse@en1:/e/gnu/pth/ac
 $ make
 gcc -g -O2 -I/usr/local/include -c foo.c
 gcc -L/usr/local/lib -o foo foo.o -lpth

If B<Pth> is installed in non-standard locations or C<pth-config>
is not in C<$PATH>, one just has to drop the C<configure> script
a note about the location by running C<configure> with the option
C<--with-pth=>I<dir> (where I<dir> is the argument which was used with
the C<--prefix> option when B<Pth> was installed).

=head2 Autoconf Build Environment with Local Copy of Pth (Expert)

Finally let us assume the C<foo> program stays under either a I<GPL> or
I<LGPL> distribution license and we want to make it a stand-alone package for
easier distribution and installation.  That is, we don't want to oblige the
end-user to install B<Pth> just to allow our C<foo> package to
compile. For this, it is a convenient practice to include the required
libraries (here B<Pth>) into the source tree of the package (here C<foo>).
B<Pth> ships with all necessary support to allow us to easily achieve this
approach. Say, we want B<Pth> in a subdirectory named C<pth/> and this
directory should be seamlessly integrated into the configuration and build
process of C<foo>.

First we again start with the C<Makefile.in>, but this time it is a more
advanced version which supports subdirectory movement:

 $ vi Makefile.in
 | CC      = @CC@
 | CFLAGS  = @CFLAGS@
 | LDFLAGS = @LDFLAGS@
 | LIBS    = @LIBS@
 |
 | SUBDIRS = pth
 |
 | all: subdirs_all foo
 |
 | subdirs_all:
 |     @$(MAKE) $(MFLAGS) subdirs TARGET=all
 | subdirs_clean:
 |     @$(MAKE) $(MFLAGS) subdirs TARGET=clean
 | subdirs_distclean:
 |     @$(MAKE) $(MFLAGS) subdirs TARGET=distclean
 | subdirs:
 |     @for subdir in $(SUBDIRS); do \
 |         echo "===> $$subdir ($(TARGET))"; \
 |         (cd $$subdir; $(MAKE) $(MFLAGS) $(TARGET) || exit 1) || exit 1; \
 |         echo "<=== $$subdir"; \
 |     done
 |
 | foo: foo.o
 |     $(CC) $(LDFLAGS) -o foo foo.o $(LIBS)
 | foo.o: foo.c
 |     $(CC) $(CFLAGS) -c foo.c
 |
 | clean: subdirs_clean
 |     rm -f foo foo.o
 | distclean: subdirs_distclean
 |     rm -f foo foo.o
 |     rm -f config.log config.status config.cache
 |     rm -f Makefile

Then we create a slightly different B<autoconf> script C<configure.ac>:

 $ vi configure.ac
 | AC_INIT(Makefile.in)
 | AC_CONFIG_AUX_DIR(pth)
 | AC_CHECK_PTH(1.3.0, subdir:pth --disable-tests)
 | AC_CONFIG_SUBDIRS(pth)
 | AC_OUTPUT(Makefile)

Here we provided a default value for C<foo>'s C<--with-pth> option as the
second argument to C<AC_CHECK_PTH> which indicates that B<Pth> can be found in
the subdirectory named C<pth/>. Additionally we specified that the
C<--disable-tests> option of B<Pth> should be passed to the C<pth/>
subdirectory, because we need only to build the B<Pth> library itself. And we
added a C<AC_CONFIG_SUBDIR> call which indicates to B<autoconf> that it should
configure the C<pth/> subdirectory, too. The C<AC_CONFIG_AUX_DIR> directive
was added just to make B<autoconf> happy, because it wants to find a
C<install.sh> or C<shtool> script if C<AC_CONFIG_SUBDIRS> is used.

Now we let B<autoconf>'s C<aclocal> program again generate for us an
C<aclocal.m4> file with the contents of B<Pth>'s C<AC_CHECK_PTH> macro.
Finally we generate the C<configure> script out of this C<aclocal.m4>
file and the C<configure.ac> file.

 $ aclocal --acdir=`pth-config --acdir`
 $ autoconf

Now we have to create the C<pth/> subdirectory itself. For this, we extract the
B<Pth> distribution to the C<foo> source tree and just rename it to C<pth/>:

 $ gunzip <pth-X.Y.Z.tar.gz | tar xvf -
 $ mv pth-X.Y.Z pth

Optionally to reduce the size of the C<pth/> subdirectory, we can strip down
the B<Pth> sources to a minimum with the I<striptease> feature:

 $ cd pth
 $ ./configure
 $ make striptease
 $ cd ..

After this the source tree of C<foo> should look similar to this:

 $ ls -l
 -rw-r--r--  1 rse  users    709 Nov  3 11:51 Makefile.in
 -rw-r--r--  1 rse  users  16431 Nov  3 12:20 aclocal.m4
 -rwxr-xr-x  1 rse  users  57403 Nov  3 12:21 configure
 -rw-r--r--  1 rse  users    129 Nov  3 12:21 configure.ac
 -rw-r--r--  1 rse  users   4227 Nov  3 11:11 foo.c
 drwxr-xr-x  2 rse  users   3584 Nov  3 12:36 pth
 $ ls -l pth/
 -rw-rw-r--  1 rse  users   26344 Nov  1 20:12 COPYING
 -rw-rw-r--  1 rse  users    2042 Nov  3 12:36 Makefile.in
 -rw-rw-r--  1 rse  users    3967 Nov  1 19:48 README
 -rw-rw-r--  1 rse  users     340 Nov  3 12:36 README.1st
 -rw-rw-r--  1 rse  users   28719 Oct 31 17:06 config.guess
 -rw-rw-r--  1 rse  users   24274 Aug 18 13:31 config.sub
 -rwxrwxr-x  1 rse  users  155141 Nov  3 12:36 configure
 -rw-rw-r--  1 rse  users  162021 Nov  3 12:36 pth.c
 -rw-rw-r--  1 rse  users   18687 Nov  2 15:19 pth.h.in
 -rw-rw-r--  1 rse  users    5251 Oct 31 12:46 pth_acdef.h.in
 -rw-rw-r--  1 rse  users    2120 Nov  1 11:27 pth_acmac.h.in
 -rw-rw-r--  1 rse  users    2323 Nov  1 11:27 pth_p.h.in
 -rw-rw-r--  1 rse  users     946 Nov  1 11:27 pth_vers.c
 -rw-rw-r--  1 rse  users   26848 Nov  1 11:27 pthread.c
 -rw-rw-r--  1 rse  users   18772 Nov  1 11:27 pthread.h.in
 -rwxrwxr-x  1 rse  users   26188 Nov  3 12:36 shtool

Now when we configure and build the C<foo> package it looks similar to this:

 $ ./configure
 creating cache ./config.cache
 checking for gcc... gcc
 checking whether the C compiler (gcc   ) works... yes
 checking whether the C compiler (gcc   ) is a cross-compiler... no
 checking whether we are using GNU C... yes
 checking whether gcc accepts -g... yes
 checking how to run the C preprocessor... gcc -E
 checking for GNU Pth... version 1.3.0, local under pth
 updating cache ./config.cache
 creating ./config.status
 creating Makefile
 configuring in pth
 running /bin/sh ./configure  --enable-subdir --enable-batch
 --disable-tests --cache-file=.././config.cache --srcdir=.
 loading cache .././config.cache
 checking for gcc... (cached) gcc
 checking whether the C compiler (gcc   ) works... yes
 checking whether the C compiler (gcc   ) is a cross-compiler... no
 [...]
 $ make
 ===> pth (all)
 ./shtool scpp -o pth_p.h -t pth_p.h.in -Dcpp -Cintern -M '==#==' pth.c
 pth_vers.c
 gcc -c -I. -O2 -pipe pth.c
 gcc -c -I. -O2 -pipe pth_vers.c
 ar rc libpth.a pth.o pth_vers.o
 ranlib libpth.a
 <=== pth
 gcc -g -O2 -Ipth -c foo.c
 gcc -Lpth -o foo foo.o -lpth

As you can see, B<autoconf> now automatically configures the local
(stripped down) copy of B<Pth> in the subdirectory C<pth/> and the
C<Makefile> automatically builds the subdirectory, too.

=head1 SYSTEM CALL WRAPPER FACILITY

B<Pth> per default uses an explicit API, including the system calls. For
instance you've to explicitly use pth_read(3) when you need a thread-aware
read(3) and cannot expect that by just calling read(3) only the current thread
is blocked. Instead with the standard read(3) call the whole process will be
blocked. But because for some applications (mainly those consisting of lots of
third-party stuff) this can be inconvenient.  Here it's required that a call
to read(3) `magically' means pth_read(3). The problem here is that such
magic B<Pth> cannot provide per default because it's not really portable.
Nevertheless B<Pth> provides a two step approach to solve this problem:

=head2 Soft System Call Mapping

This variant is available on all platforms and can I<always> be enabled by
building B<Pth> with C<--enable-syscall-soft>. This then triggers some
C<#define>'s in the C<pth.h> header which map for instance read(3) to
pth_read(3), etc. Currently the following functions are mapped: fork(2),
nanosleep(3), usleep(3), sleep(3), sigwait(3), waitpid(2), system(3),
select(2), poll(2), connect(2), accept(2), read(2), write(2), recv(2),
send(2), recvfrom(2), sendto(2).

The drawback of this approach is just that really all source files
of the application where these function calls occur have to include
C<pth.h>, of course. And this also means that existing libraries,
including the vendor's B<stdio>, usually will still block the whole
process if one of its I/O functions block.

=head2 Hard System Call Mapping

This variant is available only on those platforms where the syscall(2)
function exists and there it can be enabled by building B<Pth> with
C<--enable-syscall-hard>. This then builds wrapper functions (for instances
read(3)) into the B<Pth> library which internally call the real B<Pth>
replacement functions (pth_read(3)). Currently the following functions
are mapped: fork(2), nanosleep(3), usleep(3), sleep(3), waitpid(2),
system(3), select(2), poll(2), connect(2), accept(2), read(2), write(2).

The drawback of this approach is that it depends on syscall(2) interface
and prototype conflicts can occur while building the wrapper functions
due to different function signatures in the vendor C header files.
But the advantage of this mapping variant is that the source files of
the application where these function calls occur have not to include
C<pth.h> and that existing libraries, including the vendor's B<stdio>,
magically become thread-aware (and then block only the current thread).

=head1 IMPLEMENTATION NOTES

B<Pth> is very portable because it has only one part which perhaps has
to be ported to new platforms (the machine context initialization). But
it is written in a way which works on mostly all Unix platforms which
support makecontext(2) or at least sigstack(2) or sigaltstack(2) [see
C<pth_mctx.c> for details]. Any other B<Pth> code is POSIX and ANSI C
based only.

The context switching is done via either SUSv2 makecontext(2) or POSIX
make[sig]setjmp(3) and [sig]longjmp(3). Here all CPU registers, the
program counter and the stack pointer are switched. Additionally the
B<Pth> dispatcher switches also the global Unix C<errno> variable [see
C<pth_mctx.c> for details] and the signal mask (either implicitly via
sigsetjmp(3) or in an emulated way via explicit setprocmask(2) calls).

The B<Pth> event manager is mainly select(2) and gettimeofday(2) based,
i.e., the current time is fetched via gettimeofday(2) once per context
switch for time calculations and all I/O events are implemented via a
single central select(2) call [see C<pth_sched.c> for details].

The thread control block management is done via virtual priority
queues without any additional data structure overhead. For this, the
queue linkage attributes are part of the thread control blocks and the
queues are actually implemented as rings with a selected element as the
entry point [see C<pth_tcb.h> and C<pth_pqueue.c> for details].

Most time critical code sections (especially the dispatcher and event
manager) are speeded up by inline functions (implemented as ANSI C
pre-processor macros). Additionally any debugging code is I<completely>
removed from the source when not built with C<-DPTH_DEBUG> (see Autoconf
C<--enable-debug> option), i.e., not only stub functions remain [see
C<pth_debug.c> for details].

=head1 RESTRICTIONS

B<Pth> (intentionally) provides no replacements for non-thread-safe
functions (like strtok(3) which uses a static internal buffer) or
synchronous system functions (like gethostbyname(3) which doesn't
provide an asynchronous mode where it doesn't block). When you want to
use those functions in your server application together with threads,
you've to either link the application against special third-party
libraries (or for thread-safe/reentrant functions possibly against an
existing C<libc_r> of the platform vendor). For an asynchronous DNS
resolver library use the GNU B<adns> package from Ian Jackson ( see
http://www.gnu.org/software/adns/adns.html ).

=head1 HISTORY

The B<Pth> library was designed and implemented between February and
July 1999 by I<Ralf S. Engelschall> after evaluating numerous (mostly
preemptive) thread libraries and after intensive discussions with
I<Peter Simons>, I<Martin Kraemer>, I<Lars Eilebrecht> and I<Ralph
Babel> related to an experimental (matrix based) non-preemptive C++
scheduler class written by I<Peter Simons>.

B<Pth> was then implemented in order to combine the I<non-preemptive>
approach of multithreading (which provides better portability and
performance) with an API similar to the popular one found in B<Pthread>
libraries (which provides easy programming).

So the essential idea of the non-preemptive approach was taken over from
I<Peter Simons> scheduler. The priority based scheduling algorithm was
suggested by I<Martin Kraemer>. Some code inspiration also came from
an experimental threading library (B<rsthreads>) written by I<Robert
S. Thau> for an ancient internal test version of the Apache webserver.
The concept and API of message ports was borrowed from AmigaOS' B<Exec>
subsystem. The concept and idea for the flexible event mechanism came
from I<Paul Vixie>'s B<eventlib> (which can be found as a part of
B<BIND> v8).

=head1 BUG REPORTS AND SUPPORT

If you think you have found a bug in B<Pth>, you should send a report as
complete as possible to I<bug-pth@gnu.org>. If you can, please try to
fix the problem and include a patch, made with 'C<diff -u3>', in your
report. Always, at least, include a reasonable amount of description in
your report to allow the author to deterministically reproduce the bug.

For further support you additionally can subscribe to the
I<pth-users@gnu.org> mailing list by sending an Email to
I<pth-users-request@gnu.org> with `C<subscribe pth-users>' (or
`C<subscribe pth-users> I<address>' if you want to subscribe
from a particular Email I<address>) in the body. Then you can
discuss your issues with other B<Pth> users by sending messages to
I<pth-users@gnu.org>. Currently (as of August 2000) you can reach about
110 Pth users on this mailing list. Old postings you can find at
I<http://www.mail-archive.com/pth-users@gnu.org/>.

=head1 SEE ALSO

=head2 Related Web Locations

`comp.programming.threads Newsgroup Archive',
http://www.deja.com/topics_if.xp?
search=topic&group=comp.programming.threads

`comp.programming.threads Frequently Asked Questions (F.A.Q.)',
http://www.lambdacs.com/newsgroup/FAQ.html

`I<Multithreading - Definitions and Guidelines>',
Numeric Quest Inc 1998;
http://www.numeric-quest.com/lang/multi-frame.html

`I<The Single UNIX Specification, Version 2 - Threads>',
The Open Group 1997;
http://www.opengroup.org/onlinepubs /007908799/xsh/threads.html

SMI Thread Resources,
Sun Microsystems Inc;
http://www.sun.com/workshop/threads/

Bibliography on threads and multithreading,
Torsten Amundsen;
http://liinwww.ira.uka.de/bibliography/Os/threads.html

=head2 Related Books

B. Nichols, D. Buttlar, J.P. Farrel:
`I<Pthreads Programming - A POSIX Standard for Better Multiprocessing>',
O'Reilly 1996;
ISBN 1-56592-115-1

B. Lewis, D. J. Berg:
`I<Multithreaded Programming with Pthreads>',
Sun Microsystems Press, Prentice Hall 1998;
ISBN 0-13-680729-1

B. Lewis, D. J. Berg:
`I<Threads Primer - A Guide To Multithreaded Programming>',
Prentice Hall 1996;
ISBN 0-13-443698-9

S. J. Norton, M. D. Dipasquale:
`I<Thread Time - The Multithreaded Programming Guide>',
Prentice Hall 1997;
ISBN 0-13-190067-6

D. R. Butenhof:
`I<Programming with POSIX Threads>',
Addison Wesley 1997;
ISBN 0-201-63392-2

=head2 Related Manpages

pth-config(1), pthread(3).

getcontext(2), setcontext(2), makecontext(2), swapcontext(2),
sigstack(2), sigaltstack(2), sigaction(2), sigemptyset(2), sigaddset(2),
sigprocmask(2), sigsuspend(2), sigsetjmp(3), siglongjmp(3), setjmp(3),
longjmp(3), select(2), gettimeofday(2).

=head1 AUTHOR

 Ralf S. Engelschall
 rse@engelschall.com
 www.engelschall.com

=cut