VERSION

\s-1GNU\s0 Pth \s-12.0.7 (08-Jun-2006)\s0

SYNOPSIS

Global Library Management

pth_init, pth_kill, pth_ctrl, pth_version.

Thread Attribute Handling

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

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.

Utilities

pth_fdmode, pth_time, pth_timeout, pth_sfiodisc.

Cancellation Management

pth_cancel_point, pth_cancel_state.

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.

Key-Based Storage

pth_key_create, pth_key_delete, pth_key_setdata, pth_key_getdata.

Message Port Communication

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

Thread Cleanups

pth_cleanup_push, pth_cleanup_pop.

Process Forking

pth_atfork_push, pth_atfork_pop, pth_fork.

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.

User-Space Context

pth_uctx_create, pth_uctx_make, pth_uctx_switch, pth_uctx_destroy.

Generalized \s-1POSIX\s0 Replacement \s-1API\s0

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.

Standard \s-1POSIX\s0 Replacement \s-1API\s0

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.

DESCRIPTION

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

Pth is a very portable \s-1POSIX/ANSI-C\s0 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\*(C' 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.

Pth also provides an optional emulation \s-1API\s0 for \s-1POSIX\s0.1c threads (`Pthreads') which can be used for backward compatibility to existing multithreaded applications. See Pth's pthread\|(3) manual page for details.

\$1

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 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 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 light-weight threads of execution inside a single (heavy-weight) process, i.e., to use multithreading. Those 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. Even though lots of documents exists which describe and define the world of threading, to understand 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 Pth.

o process vs. thread

A process on Unix systems consists of at least the following fundamental ingredients: virtual memory table, program code, program counter, heap memory, stack memory, stack pointer, file descriptor set, 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.

o kernel-space vs. 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.

o preemptive vs. 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\*(C' or \*(C`SIGVTALRM\*(C'. 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.

o concurrency vs. parallelism

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

o 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.

o reentrant, thread-safe and 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 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 \s-1POSIX\s0 as guaranteed to be asynchronous-safe. Asynchronous-safe functions usually have to be already reentrant. User-space threads can be implemented in various way. The two traditional approaches are:

1.

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 \s-1ANSI\s0 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.

2.

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 \s-1POSIX/ANSI-C\s0 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.

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. But why not combine the good aspects of both approaches while avoiding their bad aspects? That's the goal of Pth. 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 Pth. The following list summarizes a few essential points:

o

Pth provides maximum portability, but \s-1NOT\s0 the fanciest features. This is, because it uses a nifty and portable \s-1POSIX/ANSI-C\s0 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\*(C'). 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.

o

Pth increases the responsiveness and concurrency of an event-driven application, but \s-1NOT\s0 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 \s-1CPU\s0 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 \s-1CPU\s0 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 Pth is mainly intended for server type applications, although there is no technical restriction.

o

Pth requires thread-safe functions, but \s-1NOT\s0 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 Pth more existing third-party libraries can be used without side-effects than it's the case for other threading systems.

o

Pth doesn't require any kernel support, but can \s-1NOT\s0 benefit from multiprocessor machines. This means that Pth runs on almost all Unix kernels, because the kernel does not need to be aware of the 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. To understand the Pth Application Programming Interface (\s-1API\s0), it helps to first understand the life cycle of a thread in the 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 \s-1NEW\s0 queue of the scheduler. On the next dispatching for this thread, the scheduler picks it up from there and moves it to the \s-1READY\s0 queue. This is a queue containing all threads which want to perform a \s-1CPU\s0 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 \s-1READY\s0 queue is the new \s-1RUNNING\s0 thread (there is always just one \s-1RUNNING\s0 thread, of course). The \s-1RUNNING\s0 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 \s-1DEAD\s0 queue, or it has events on which it wants to wait, then it is moved into the \s-1WAITING\s0 queue. Else it is assumed it wants to perform more \s-1CPU\s0 bursts and immediately enters the \s-1READY\s0 queue again.

Before the next thread is taken out of the \s-1READY\s0 queue, the \s-1WAITING\s0 queue is checked for pending events. If one or more events occurred, the threads that are waiting on them are immediately moved to the \s-1READY\s0 queue.

The purpose of the \s-1NEW\s0 queue has to do with the fact that in 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 \s-1NEW\s0 queue is for.

The purpose of the \s-1DEAD\s0 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 \s-1DEAD\s0 queue. There it remains until another thread joins it.

Finally, there is a special separated queue named \s-1SUSPENDED\s0, to where threads can be manually moved from the \s-1NEW\s0, \s-1READY\s0 or \s-1WAITING\s0 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.

APPLICATION PROGRAMMING INTERFACE (API)

In the following the Pth Application Programming Interface (\s-1API\s0) is discussed in detail. With the knowledge given above, it should now be easy to understand how to program threads with this \s-1API\s0. In good Unix tradition, Pth functions use special return values (\*(C`NULL\*(C' in pointer context, \*(C`FALSE\*(C' in boolean context and \*(C`-1\*(C' in integer context) to indicate an error condition and set (or pass through) the \*(C`errno\*(C' system variable to pass more details about the error to the caller. The following functions act on the library as a whole. They are used to initialize and shutdown the scheduler and fetch information from it.

int pth_init(void);

This initializes the Pth library. It has to be the first Pth \s-1API\s0 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\*(C' on success and \*(C`FALSE\*(C' on error.

int pth_kill(void);

This kills the Pth library. It should be the last Pth \s-1API\s0 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 Pth application is either a simple `\*(C`pth_exit(0);\*(C'' 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)\*(C'' (which immediately kills the threading system and terminates the process). The pth_kill() return immediately with a return code of \*(C`FALSE\*(C' if it is not called from within the main thread. Else it kills the threading system and returns \*(C`TRUE\*(C'.

long pth_ctrl(unsigned long query, ...);

This is a generalized query/control function for the Pth library. The argument query is a bitmask formed out of one or more \*(C`PTH_CTRL_\*(C'\s-1XXXX\s0 queries. Currently the following queries are supported:

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\*(C' query is equal to the OR-combination of all the following specialized queries: \*(C`PTH_CTRL_GETTHREADS_NEW\*(C' 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\*(C' for the number of threads in the ready queue (threads who want to do \s-1CPU\s0 bursts), \*(C`PTH_CTRL_GETTHREADS_RUNNING\*(C' for the number of running threads (always just one thread!), \*(C`PTH_CTRL_GETTHREADS_WAITING\*(C' for the number of threads in the waiting queue (threads waiting for events), \*(C`PTH_CTRL_GETTHREADS_SUSPENDED\*(C' for the number of threads in the suspended queue (threads waiting to be resumed) and \*(C`PTH_CTRL_GETTHREADS_DEAD\*(C' for the number of threads in the new queue (terminated threads waiting for a join). This requires a second argument of type `\*(C`float *\*(C'' (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 \s-1CPU\s0 bursts. The average load value updates once per second only. The return value for this query is always 0. This requires a second argument of type `\*(C`pth_t\*(C'' which identifies a thread. It returns the priority (ranging from \*(C`PTH_PRIO_MIN\*(C' to \*(C`PTH_PRIO_MAX\*(C') of the given thread. This requires a second argument of type `\*(C`pth_t\*(C'' 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 *\*(C''. This requires a second argument of type `\*(C`FILE *\*(C'' to which a summary of the internal Pth library state is written to. The main information which is currently written out is the current state of the thread pool. This requires a second argument of type `\*(C`int\*(C'' which specified whether the \s-1GNU\s0 Pth scheduler favours new threads on startup, i.e., whether they are moved from the new queue to the top (argument is \*(C`TRUE\*(C') or middle (argument is \*(C`FALSE\*(C') 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.

The function returns \*(C`-1\*(C' on error.

long pth_version(void);

This function returns a hex-value `0xV\s-1RR\s0T\s-1LL\s0' which describes the current Pth library version. V is the version, \s-1RR\s0 the revisions, \s-1LL\s0 the level and T the type of the level (alphalevel=0, betalevel=1, patchlevel=2, etc). For instance Pth version 1.0b1 is encoded as 0x100101. The reason for this unusual mapping is that this way the version number is steadily increasing. The same value is also available under compile time as \*(C`PTH_VERSION\*(C'. Attribute objects are used in 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: Thread Priority between \*(C`PTH_PRIO_MIN\*(C' and \*(C`PTH_PRIO_MAX\*(C'. The default is \*(C`PTH_PRIO_STD\*(C'. Name of thread (up to 40 characters are stored only), mainly for debugging purposes. In bounded attribute objects, this field is incremented every time the context is switched to the associated thread. The thread detachment type, \*(C`TRUE\*(C' indicates a joinable thread, \*(C`FALSE\*(C' 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. The thread cancellation state, i.e., a combination of \*(C`PTH_CANCEL_ENABLE\*(C' or \*(C`PTH_CANCEL_DISABLE\*(C' and \*(C`PTH_CANCEL_DEFERRED\*(C' or \*(C`PTH_CANCEL_ASYNCHRONOUS\*(C'. The thread stack size in bytes. Use lower values than 64 \s-1KB\s0 with great care! A pointer to the lower address of a chunk of malloc\|(3)'ed memory for the stack. The time when the thread was spawned. This can be queried only when the attribute object is bound to a thread. The time when the thread was last dispatched. This can be queried only when the attribute object is bound to a thread. The total time the thread was running. This can be queried only when the attribute object is bound to a thread. The thread start function. This can be queried only when the attribute object is bound to a thread. The thread start argument. This can be queried only when the attribute object is bound to a thread. The scheduling state of the thread, i.e., either \*(C`PTH_STATE_NEW\*(C', \*(C`PTH_STATE_READY\*(C', \*(C`PTH_STATE_WAITING\*(C', or \*(C`PTH_STATE_DEAD\*(C' This can be queried only when the attribute object is bound to a thread. The event ring the thread is waiting for. This can be queried only when the attribute object is bound to a thread. Whether the attribute object is bound (\*(C`TRUE\*(C') to a thread or not (\*(C`FALSE\*(C').

The following \s-1API\s0 functions can be used to handle the attribute objects:

pth_attr_t pth_attr_of(pth_t tid);

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

pth_attr_t pth_attr_new(void);

This returns a new 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.

int pth_attr_init(pth_attr_t attr);

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

int pth_attr_set(pth_attr_t attr, int field, ...);

This sets the attribute field field in attr to a value specified as an additional argument on the variable argument list. The following attribute 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 *

int pth_attr_get(pth_attr_t attr, int field, ...);

This retrieves the attribute field field in attr and stores its value in the variable specified through a pointer in an additional argument on the variable argument list. The following 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 *

int pth_attr_destroy(pth_attr_t attr);

This destroys a attribute object attr. After this attr is no longer a valid attribute object. The following functions control the threading itself and make up the main \s-1API\s0 of the Pth library.

pth_t pth_spawn(pth_attr_t attr, void *(*entry)(void *), void *arg);

This spawns a new thread with the attributes given in attr (or \*(C`PTH_ATTR_DEFAULT\*(C' for default attributes - which means that thread priority, joinability and cancel state are inherited from the current thread) with the starting point at routine entry; the dispatch count is not inherited from the current thread if attr is not specified - rather, it is initialized to zero. This entry routine is called as `pth_exit(entry(arg))' inside the new thread unit, i.e., 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 \s-1POSIX\s0 function exit\|(3) still terminates the complete process and not just the current thread. There is no Pth-internal limit on the number of threads one can spawn, except the limit implied by the available virtual memory. Pth internally keeps track of thread in dynamic data structures. The function returns \*(C`NULL\*(C' on error.

int pth_once(pth_once_t *ctrlvar, void (*func)(void *), void *arg);

This is a convenience function which uses a control variable of type \*(C`pth_once_t\*(C' to make sure a constructor function func is called only once as `func(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 ctrlvar should be declared as `\*(C`pth_once_t\*(C' variable-name = \*(C`PTH_ONCE_INIT\*(C';' before calling this function.

pth_t 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\*(C'.

int pth_suspend(pth_t tid);

This suspends a thread tid until it is manually resumed again via pth_resume\|(3). For this, the thread is moved to the \s-1SUSPENDED\s0 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\*(C' on success and \*(C`FALSE\*(C' on errors.

int pth_resume(pth_t tid);

This function resumes a previously suspended thread tid, i.e. tid has to stay on the \s-1SUSPENDED\s0 queue. The thread is moved to the \s-1NEW\s0, \s-1READY\s0 or \s-1WAITING\s0 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\*(C' on success and \*(C`FALSE\*(C' on errors.

int pth_raise(pth_t tid, int sig)

This function raises a signal for delivery to thread 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 action is still configured process-wide. When sig is 0 plain thread checking is performed, i.e., `\*(C`pth_raise(tid, 0)\*(C'' returns \*(C`TRUE\*(C' when thread tid still exists in the \s-1PTH\s0 system but doesn't send any signal to it.

int pth_yield(pth_t 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 \s-1CPU\s0 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 Pth. A thread 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 \s-1CPU\s0 bursts into smaller units with this call. Usually one specifies tid as \*(C`NULL\*(C' 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 coroutines where a thread/routine switches to a particular cooperating thread. If tid is not \*(C`NULL\*(C' and points to a new or ready thread, it is guaranteed that this thread receives execution control on the next dispatching step. If tid is in a different state (that is, not in \*(C`PTH_STATE_NEW\*(C' or \*(C`PTH_STATE_READY\*(C') an error is reported. The function usually returns \*(C`TRUE\*(C' for success and only \*(C`FALSE\*(C' (with \*(C`errno\*(C' set to \*(C`EINVAL\*(C') if tid specified an invalid or still not new or ready thread.

int pth_nap(pth_time_t naptime);

This functions suspends the execution of the current thread until naptime is elapsed. naptime is of type \*(C`pth_time_t\*(C' 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 naptime has elapsed. It's only guarantees that the thread will sleep at least naptime. But because of the non-preemptive nature of Pth it can last longer (when another thread kept the \s-1CPU\s0 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.

int pth_wait(pth_event_t 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 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 ev argument is a 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.

int pth_cancel(pth_t tid);

This cancels a thread tid. How the cancellation is done depends on the cancellation state of tid which the thread can configure itself. When its state is \*(C`PTH_CANCEL_DISABLE\*(C' a cancellation request is just made pending. When it is \*(C`PTH_CANCEL_ENABLE\*(C' it depends on the cancellation type what is performed. When its \*(C`PTH_CANCEL_DEFERRED\*(C' again the cancellation request is just made pending. But when its \*(C`PTH_CANCEL_ASYNCHRONOUS\*(C' 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)\*(C'' at one of his cancellation points. In Pth thread enter a cancellation point either explicitly via pth_cancel_point\|(3) or implicitly by waiting for an event.

int pth_abort(pth_t tid);

This is the cruel way to cancel a thread tid. When it's already dead and waits to be joined it just joins it (via `\*(C`pth_join(\*(C'tid\*(C`, NULL)\*(C'') 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(\*(C'tid\*(C`)\*(C''.

int pth_join(pth_t tid, void **value);

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

void pth_exit(void *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\*(C' set to \*(C`FALSE\*(C', it's immediately removed and value is ignored. Else the thread is inserted into the dead queue and value remembered for a subsequent pth_join\|(3) call by another thread. Utility functions.

int pth_fdmode(int fd, int mode);

This switches the non-blocking mode flag on file descriptor fd. The argument mode can be \*(C`PTH_FDMODE_BLOCK\*(C' for switching fd into blocking I/O mode, \*(C`PTH_FDMODE_NONBLOCK\*(C' for switching fd into non-blocking I/O mode or \*(C`PTH_FDMODE_POLL\*(C' for just polling the current mode. The current mode is returned (either \*(C`PTH_FDMODE_BLOCK\*(C' or \*(C`PTH_FDMODE_NONBLOCK\*(C') or \*(C`PTH_FDMODE_ERROR\*(C' on error. Keep in mind that since 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 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.

pth_time_t pth_time(long sec, long usec);

This is a constructor for a \*(C`pth_time_t\*(C' structure which is a convenient function to avoid temporary structure values. It returns a pth_time_t structure which holds the absolute time value specified by sec and usec.

pth_time_t pth_timeout(long sec, long usec);

This is a constructor for a \*(C`pth_time_t\*(C' structure which is a convenient function to avoid temporary structure values. It returns a pth_time_t structure which holds the absolute time value calculated by adding sec and usec to the current time.

Sfdisc_t *pth_sfiodisc(void);

This functions is always available, but only reasonably usable when Pth was built with Sfio support (\*(C`--with-sfio\*(C' option) and \*(C`PTH_EXT_SFIO\*(C' is then defined by \*(C`pth.h\*(C'. It is useful for applications which want to use the comprehensive Sfio I/O library with the Pth threading library. Then this function can be used to get an Sfio discipline structure (\*(C`Sfdisc_t\*(C') which can be pushed onto Sfio streams (\*(C`Sfio_t\*(C') 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 Sfio stream does only block the current thread instead of the whole process. The application has to free\|(3) the \*(C`Sfdisc_t\*(C' structure when it is no longer needed. The Sfio package can be found at http://www.research.att.com/sw/tools/sfio/. Pth supports \s-1POSIX\s0 style thread cancellation via pth_cancel\|(3) and the following two related functions:

void pth_cancel_state(int newstate, int *oldstate);

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

void pth_cancel_point(void);

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

pth_event_t pth_event(unsigned long spec, ...);

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

This is a file descriptor event. One or more of \*(C`PTH_UNTIL_FD_READABLE\*(C', \*(C`PTH_UNTIL_FD_WRITEABLE\*(C' or \*(C`PTH_UNTIL_FD_EXCEPTION\*(C' have to be OR-ed into 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)\*(C''. 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 \s-1BSD\s0 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\*(C' events). Example: `\*(C`pth_event(PTH_EVENT_SELECT, &rc, nfd, rfds, wfds, efds)\*(C'' where \*(C`rc\*(C' has to be of type `\*(C`int *\*(C'', \*(C`nfd\*(C' has to be of type `\*(C`int\*(C'' and \*(C`rfds\*(C', \*(C`wfds\*(C' and \*(C`efds\*(C' have to be of type `\*(C`fd_set *\*(C'' (see select\|(2)). The number of occurred file descriptors are stored in \*(C`rc\*(C'. This is a signal set event. The two additional arguments have to be a pointer to a signal set (type `\*(C`sigset_t *\*(C'') and a pointer to a signal number variable (type `\*(C`int *\*(C''). 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 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);\*(C''. This is a time point event. The additional argument has to be of type \*(C`pth_time_t\*(C' (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))\*(C''. This is a message port event. The additional argument has to be of type \*(C`pth_msgport_t\*(C'. This events waits until one or more messages were received on the specified message port. Example: `\*(C`pth_event(PTH_EVENT_MSG, mp)\*(C''. This is a thread event. The additional argument has to be of type \*(C`pth_t\*(C'. One of \*(C`PTH_UNTIL_TID_NEW\*(C', \*(C`PTH_UNTIL_TID_READY\*(C', \*(C`PTH_UNTIL_TID_WAITING\*(C' or \*(C`PTH_UNTIL_TID_DEAD\*(C' has to be OR-ed into 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)\*(C''. This is a custom callback function event. Three additional arguments have to be given with the following types: `\*(C`int (*)(void *)\*(C'', `\*(C`void *\*(C'' and `\*(C`pth_time_t\*(C''. 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\*(C'. Once it returned \*(C`TRUE\*(C' 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))\*(C''.

unsigned long pth_event_typeof(pth_event_t ev);

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

int pth_event_extract(pth_event_t 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 ev via `\*(C`ev = pth_event(PTH_EVENT_FD, fd);\*(C'' you have to extract it via `\*(C`pth_event_extract(ev, &fd)\*(C'', 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 pointers to variables and these variables have to be of the same type as the argument of pth_event\|(3) required.

pth_event_t pth_event_concat(pth_event_t ev, ...);

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

pth_event_t pth_event_isolate(pth_event_t ev);

This isolates the event ev from possibly appended events in the event ring. When in ev only one event exists, this returns \*(C`NULL\*(C'. When remaining events exists, they form a new event ring which is returned.

pth_event_t pth_event_walk(pth_event_t ev, int direction);

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

pth_status_t pth_event_status(pth_event_t ev);

This returns the status of event ev. This is a fast operation because only a tag on 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\*(C' (event is still pending), \*(C`PTH_STATUS_OCCURRED\*(C' (event successfully occurred), \*(C`PTH_STATUS_FAILED\*(C' (event failed).

int pth_event_free(pth_event_t ev, int mode);

This deallocates the event ev (when mode is \*(C`PTH_FREE_THIS\*(C') or all events appended to the event ring under ev (when mode is \*(C`PTH_FREE_ALL\*(C'). The following functions provide thread-local storage through unique keys similar to the \s-1POSIX\s0 Pthread \s-1API\s0. Use this for thread specific global data.

int pth_key_create(pth_key_t *key, void (*func)(void *));

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

int pth_key_delete(pth_key_t key);

This explicitly destroys a key key.

int pth_key_setdata(pth_key_t key, const void *value);

This stores value under key.

void *pth_key_getdata(pth_key_t key);

This retrieves the value under key. The following functions provide message ports which can be used for efficient and flexible inter-thread communication.

pth_msgport_t pth_msgport_create(const char *name);

This returns a pointer to a new message port. If name name is not \*(C`NULL\*(C', the 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.

void pth_msgport_destroy(pth_msgport_t mp);

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

pth_msgport_t pth_msgport_find(const char *name);

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

int pth_msgport_pending(pth_msgport_t mp);

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

int pth_msgport_put(pth_msgport_t mp, pth_message_t *m);

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

pth_message_t *pth_msgport_get(pth_msgport_t mp);

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

int pth_msgport_reply(pth_message_t *m);

This replies a message m to the message port of the sender. Per-thread cleanup functions.

int pth_cleanup_push(void (*handler)(void *), void *arg);

This pushes the routine handler onto the stack of cleanup routines for the current thread. These routines are called in \s-1LIFO\s0 order when the thread terminates.

int pth_cleanup_pop(int execute);

This pops the top-most routine from the stack of cleanup routines for the current thread. When execute is \*(C`TRUE\*(C' the routine is additionally called. The following functions provide some special support for process forking situations inside the threading environment.

int pth_atfork_push(void (*prepare)(void *), void (*)(void *parent), void (*)(void *child), void *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 prepare handler is called before fork\|(2) processing commences. The parent handler is called after fork\|(2) processing completes in the parent process. The 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\*(C'. Each handler is called with arg as the argument. The order of calls to pth_atfork_push\|(3) is significant. The parent and child handlers are called in the order in which they were established by calls to pth_atfork_push\|(3), i.e., \s-1FIFO\s0. The prepare fork handlers are called in the opposite order, i.e., \s-1LIFO\s0.

int 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\*(C' when no more handlers couldn't be removed from the stack.

pid_t 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). The following functions provide synchronization support via mutual exclusion locks (mutex), read-write locks (rwlock), condition variables (cond) and barriers (barrier). Keep in mind that in a non-preemptive threading system like 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.

int pth_mutex_init(pth_mutex_t *mutex);

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

int pth_mutex_acquire(pth_mutex_t *mutex, int try, pth_event_t ev);

This acquires a mutex 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 ev occurred (when ev is not \*(C`NULL\*(C'). 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 try is \*(C`TRUE\*(C' this function never suspends execution. Instead it returns \*(C`FALSE\*(C' with \*(C`errno\*(C' set to \*(C`EBUSY\*(C'.

int pth_mutex_release(pth_mutex_t *mutex);

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

int pth_rwlock_init(pth_rwlock_t *rwlock);

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

int pth_rwlock_acquire(pth_rwlock_t *rwlock, int op, int try, pth_event_t ev);

This acquires a read-only (when op is \*(C`PTH_RWLOCK_RD\*(C') or a read-write (when op is \*(C`PTH_RWLOCK_RW\*(C') lock 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 ev events can be given to let the locking timeout, etc. When try is \*(C`TRUE\*(C' this function never suspends execution. Instead it returns \*(C`FALSE\*(C' with \*(C`errno\*(C' set to \*(C`EBUSY\*(C'.

int pth_rwlock_release(pth_rwlock_t *rwlock);

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

int pth_cond_init(pth_cond_t *cond);

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

int pth_cond_await(pth_cond_t *cond, pth_mutex_t *mutex, pth_event_t ev);

This awaits a condition situation. The caller has to follow the semantics of the \s-1POSIX\s0 condition variables: mutex has to be acquired before this function is called. The execution of the current thread is then suspended either until the events in ev occurred (when ev is not \*(C`NULL\*(C') or cond was notified by another thread via pth_cond_notify\|(3). While the thread is waiting, mutex is released. Before it returns mutex is reacquired.

int pth_cond_notify(pth_cond_t *cond, int broadcast);

This notified one or all threads which are waiting on cond. When broadcast is \*(C`TRUE\*(C' all thread are notified, else only a single (unspecified) one.

int pth_barrier_init(pth_barrier_t *barrier, int threshold);

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

int pth_barrier_reach(pth_barrier_t *barrier);

This function reaches a barrier barrier. If this is the last thread (as specified by threshold on init of 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\*(C' on error) the value \*(C`TRUE\*(C' for any thread which neither reached the barrier as the first nor the last thread; \*(C`PTH_BARRIER_HEADLIGHT\*(C' for the thread which reached the barrier as the first thread and \*(C`PTH_BARRIER_TAILLIGHT\*(C' for the thread which reached the barrier as the last thread. 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 \s-1GNU\s0 Pth are based on. Hence these functions you can use for implementing your own simple user-space threads. The \*(C`pth_uctx_t\*(C' context is somewhat modeled after \s-1POSIX\s0 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 \s-1CPU\s0 with 800Mhz running under FreeBSD 4 one usually achieves about 260,000 user-space context switches (via pth_uctx_switch\|(3)) per second.

int pth_uctx_create(pth_uctx_t *uctx);

This function creates a user-space context and stores it into 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\*(C', else \*(C`FALSE\*(C'.

int pth_uctx_make(pth_uctx_t uctx, char *sk_addr, size_t sk_size, const sigset_t *sigmask, void (*start_func)(void *), void *start_arg, pth_uctx_t uctx_after);

This function makes a new user-space context in uctx which will operate on the run-time stack sk_addr (which is of maximum size sk_size), with the signals in sigmask blocked (if sigmask is not \*(C`NULL\*(C') and starting to execute with the call start_func(start_arg). If sk_addr is \*(C`NULL\*(C', a stack is dynamically allocated. The stack size sk_size has to be at least 16384 (16KB). If the start function start_func returns and uctx_after is not \*(C`NULL\*(C', an implicit user-space context switch to this context is performed. Else (if uctx_after is \*(C`NULL\*(C') the process is terminated with exit\|(3). This function is somewhat modeled after \s-1POSIX\s0 makecontext\|(3). On success, this function returns \*(C`TRUE\*(C', else \*(C`FALSE\*(C'.

int pth_uctx_switch(pth_uctx_t uctx_from, pth_uctx_t uctx_to);

This function saves the current user-space context in uctx_from for later restoring by another call to pth_uctx_switch\|(3) and restores the new user-space context from 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 \s-1POSIX\s0 swapcontext\|(3). If uctx_from or uctx_to are \*(C`NULL\*(C' or if uctx_to contains no valid user-space context, \*(C`FALSE\*(C' is returned instead of \*(C`TRUE\*(C'. These are the only errors possible.

int pth_uctx_destroy(pth_uctx_t uctx);

This function destroys the user-space context in uctx. The run-time stack associated with the user-space context is deallocated only if it was not given by the application (see sk_addr of pth_uctx_create\|(3)). If uctx is \*(C`NULL\*(C', \*(C`FALSE\*(C' is returned instead of \*(C`TRUE\*(C'. This is the only error possible. The following functions are generalized replacements functions for the \s-1POSIX\s0 \s-1API\s0, i.e., they are similar to the functions under `Standard \s-1POSIX\s0 Replacement \s-1API\s0' but all have an additional event argument which can be used for timeouts, etc.

int pth_sigwait_ev(const sigset_t *set, int *sig, pth_event_t ev);

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

int pth_connect_ev(int s, const struct sockaddr *addr, socklen_t addrlen, pth_event_t ev);

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

int pth_accept_ev(int s, struct sockaddr *addr, socklen_t *addrlen, pth_event_t ev);

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

int pth_select_ev(int nfd, fd_set *rfds, fd_set *wfds, fd_set *efds, struct timeval *timeout, pth_event_t ev);

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

int pth_poll_ev(struct pollfd *fds, unsigned int nfd, int timeout, pth_event_t ev);

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

ssize_t pth_read_ev(int fd, void *buf, size_t nbytes, pth_event_t ev);

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

ssize_t pth_readv_ev(int fd, const struct iovec *iovec, int iovcnt, pth_event_t ev);

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

ssize_t pth_write_ev(int fd, const void *buf, size_t nbytes, pth_event_t ev);

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

ssize_t pth_writev_ev(int fd, const struct iovec *iovec, int iovcnt, pth_event_t ev);

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

ssize_t pth_recv_ev(int fd, void *buf, size_t nbytes, int flags, pth_event_t ev);

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

ssize_t pth_recvfrom_ev(int fd, void *buf, size_t nbytes, int flags, struct sockaddr *from, socklen_t *fromlen, pth_event_t ev);

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

ssize_t pth_send_ev(int fd, const void *buf, size_t nbytes, int flags, pth_event_t ev);

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

ssize_t pth_sendto_ev(int fd, const void *buf, size_t nbytes, int flags, const struct sockaddr *to, socklen_t tolen, pth_event_t ev);

This is equal to pth_sendto\|(3) (see below), but has an additional event argument ev. When pth_sendto\|(3) suspends the current threads execution it usually only uses the I/O event on fd to awake. With this function any number of extra events can be used to awake the current thread (remember that ev actually is an event ring). The following functions are standard replacements functions for the \s-1POSIX\s0 \s-1API\s0. The difference is mainly that they suspend the current thread only instead of the whole process in case the file descriptors will block.

int pth_nanosleep(const struct timespec *rqtp, struct timespec *rmtp);

This is a variant of the \s-1POSIX\s0 nanosleep\|(3) function. It suspends the current threads execution until the amount of time in rqtp elapsed. The thread is guaranteed to not wake up before this time, but because of the non-preemptive scheduling nature of Pth, it can be awakened later, of course. If rmtp is not \*(C`NULL\*(C', the \*(C`timespec\*(C' 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.

int pth_usleep(unsigned int usec);

This is a variant of the 4.3BSD usleep\|(3) function. It suspends the current threads execution until usec microseconds (= 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 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.

unsigned int pth_sleep(unsigned int sec);

This is a variant of the \s-1POSIX\s0 sleep\|(3) function. It suspends the current threads execution until sec seconds elapsed. The thread is guaranteed to not wake up before this time, but because of the non-preemptive scheduling nature of 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.

pid_t pth_waitpid(pid_t pid, int *status, int options);

This is a variant of the \s-1POSIX\s0 waitpid\|(2) function. It suspends the current threads execution until status information is available for a terminated child process 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).

int pth_system(const char *cmd);

This is a variant of the \s-1POSIX\s0 system\|(3) function. It executes the shell command cmd with Bourne Shell (\*(C`sh\*(C') 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).

int pth_sigmask(int how, const sigset_t *set, sigset_t *oset)

This is the Pth thread-related equivalent of \s-1POSIX\s0 sigprocmask\|(2) respectively pthread_sigmask\|(3). The arguments how, set and oset directly relate to sigprocmask\|(2), because 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).

int pth_sigwait(const sigset_t *set, int *sig);

This is a variant of the \s-1POSIX\s0.1c sigwait\|(3) function. It suspends the current threads execution until a signal in set occurred and stores the signal number in 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 asynchronous safe functions you should recognize that this is a great benefit.

int pth_connect(int s, const struct sockaddr *addr, socklen_t addrlen);

This is a variant of the 4.2BSD connect\|(2) function. It establishes a connection on a socket s to target specified in addr and 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).

int pth_accept(int s, struct sockaddr *addr, socklen_t *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 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).

int pth_select(int nfd, fd_set *rfds, fd_set *wfds, fd_set *efds, struct timeval *timeout);

This is a variant of the 4.2BSD select\|(2) function. It examines the I/O descriptor sets whose addresses are passed in rfds, wfds, and 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).

int pth_pselect(int nfd, fd_set *rfds, fd_set *wfds, fd_set *efds, const struct timespec *timeout, const sigset_t *sigmask);

This is a variant of the \s-1POSIX\s0 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\*(C' is passed instead of the lower-resolution \*(C`struct timeval\*(C' 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).

int pth_poll(struct pollfd *fds, unsigned int nfd, int timeout);

This is a variant of the SysV poll\|(2) function. It examines the I/O descriptors which are passed in the array 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).

ssize_t pth_read(int fd, void *buf, size_t nbytes);

This is a variant of the \s-1POSIX\s0 read\|(2) function. It reads up to nbytes bytes into buf from file descriptor 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).

ssize_t pth_readv(int fd, const struct iovec *iovec, int iovcnt);

This is a variant of the \s-1POSIX\s0 readv\|(2) function. It reads data from file descriptor fd into the first iovcnt rows of the 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).

ssize_t pth_write(int fd, const void *buf, size_t nbytes);

This is a variant of the \s-1POSIX\s0 write\|(2) function. It writes nbytes bytes from buf to file descriptor 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).

ssize_t pth_writev(int fd, const struct iovec *iovec, int iovcnt);

This is a variant of the \s-1POSIX\s0 writev\|(2) function. It writes data to file descriptor fd from the first iovcnt rows of the 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).

ssize_t pth_pread(int fd, void *buf, size_t nbytes, off_t offset);

This is a variant of the \s-1POSIX\s0 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 offset for the desired position inside the file.

ssize_t pth_pwrite(int fd, const void *buf, size_t nbytes, off_t offset);

This is a variant of the \s-1POSIX\s0 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 offset for the desired position inside the file.

ssize_t pth_recv(int fd, void *buf, size_t nbytes, int flags);

This is a variant of the SUSv2 recv\|(2) function and equal to ``pth_recvfrom(fd, buf, nbytes, flags, \s-1NULL\s0, 0)''.

ssize_t pth_recvfrom(int fd, void *buf, size_t nbytes, int flags, struct sockaddr *from, socklen_t *fromlen);

This is a variant of the SUSv2 recvfrom\|(2) function. It reads up to nbytes bytes into buf from file descriptor fd while using flags and from/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).

ssize_t pth_send(int fd, const void *buf, size_t nbytes, int flags);

This is a variant of the SUSv2 send\|(2) function and equal to ``pth_sendto(fd, buf, nbytes, flags, \s-1NULL\s0, 0)''.

ssize_t pth_sendto(int fd, const void *buf, size_t nbytes, int flags, const struct sockaddr *to, socklen_t tolen);

This is a variant of the SUSv2 sendto\|(2) function. It writes nbytes bytes from buf to file descriptor fd while using flags and to/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).

EXAMPLE

The following example is a useless server which does nothing more than listening on \s-1TCP\s0 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\*(C' every 5 seconds. The example contains no error checking and is only intended to show you the look and feel of 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); } }

BUILD ENVIRONMENTS

In this section we will discuss the canonical ways to establish the build environment for a Pth based program. The possibilities supported by Pth range from very simple environments to rather complex ones. As a first example, assume we have the above test program staying in the source file \*(C`foo.c\*(C'. Then we can create a very simple build environment by just adding the following \*(C`Makefile\*(C':

$ 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 Pth installation via its \*(C`pth-config\*(C' program. This approach is straight-forward and works fine for small projects. 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 Pth, that is, against a Pth source tree which was just configured and built, but not installed. Third, it would be also useful to allow checking of the Pth version to make sure it is at least a minimum required version. And finally, it would be also great to make sure Pth works correctly by first performing some sanity compile and run-time checks. All this can be done if we use \s-1GNU\s0 autoconf and the \*(C`AC_CHECK_PTH\*(C' macro provided by Pth. For this, we establish the following three files:

First we again need the \*(C`Makefile\*(C', but this time it contains autoconf placeholders and additional cleanup targets. And we create it under the name \*(C`Makefile.in\*(C', because it is now an input file for 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 autoconf generates additional files, we added a canonical \*(C`distclean\*(C' target which cleans this up. Secondly, we wrote \*(C`configure.ac\*(C', a (minimal) autoconf script specification:

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

Then we let autoconf's \*(C`aclocal\*(C' program generate for us an \*(C`aclocal.m4\*(C' file containing Pth's \*(C`AC_CHECK_PTH\*(C' macro. Then we generate the final \*(C`configure\*(C' script out of this \*(C`aclocal.m4\*(C' file and the \*(C`configure.ac\*(C' 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\*(C' we get a correct \*(C`Makefile\*(C' which immediately can be used to build \*(C`foo\*(C' (assuming that Pth is already installed somewhere, so that \*(C`pth-config\*(C' is in $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 Pth is installed in non-standard locations or \*(C`pth-config\*(C' is not in $PATH, one just has to drop the \*(C`configure\*(C' script a note about the location by running \*(C`configure\*(C' with the option \*(C`--with-pth=\*(C'dir (where dir is the argument which was used with the \*(C`--prefix\*(C' option when Pth was installed). Finally let us assume the \*(C`foo\*(C' program stays under either a \s-1GPL\s0 or \s-1LGPL\s0 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 Pth just to allow our \*(C`foo\*(C' package to compile. For this, it is a convenient practice to include the required libraries (here Pth) into the source tree of the package (here \*(C`foo\*(C'). Pth ships with all necessary support to allow us to easily achieve this approach. Say, we want Pth in a subdirectory named \*(C`pth/\*(C' and this directory should be seamlessly integrated into the configuration and build process of \*(C`foo\*(C'.

First we again start with the \*(C`Makefile.in\*(C', 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 autoconf script \*(C`configure.ac\*(C':

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

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

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

Now we have to create the \*(C`pth/\*(C' subdirectory itself. For this, we extract the Pth distribution to the \*(C`foo\*(C' source tree and just rename it to \*(C`pth/\*(C':

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

Optionally to reduce the size of the \*(C`pth/\*(C' subdirectory, we can strip down the Pth sources to a minimum with the striptease feature:

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

After this the source tree of \*(C`foo\*(C' 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\*(C' 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, autoconf now automatically configures the local (stripped down) copy of Pth in the subdirectory \*(C`pth/\*(C' and the \*(C`Makefile\*(C' automatically builds the subdirectory, too.

SYSTEM CALL WRAPPER FACILITY

Pth per default uses an explicit \s-1API\s0, 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 Pth cannot provide per default because it's not really portable. Nevertheless Pth provides a two step approach to solve this problem: This variant is available on all platforms and can always be enabled by building Pth with \*(C`--enable-syscall-soft\*(C'. This then triggers some \*(C`#define\*(C''s in the \*(C`pth.h\*(C' 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\*(C', of course. And this also means that existing libraries, including the vendor's stdio, usually will still block the whole process if one of its I/O functions block. This variant is available only on those platforms where the syscall\|(2) function exists and there it can be enabled by building Pth with \*(C`--enable-syscall-hard\*(C'. This then builds wrapper functions (for instances read\|(3)) into the Pth library which internally call the real 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\*(C' and that existing libraries, including the vendor's stdio, magically become thread-aware (and then block only the current thread).

IMPLEMENTATION NOTES

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\*(C' for details]. Any other Pth code is \s-1POSIX\s0 and \s-1ANSI\s0 C based only.

The context switching is done via either SUSv2 makecontext\|(2) or \s-1POSIX\s0 make[sig]setjmp\|(3) and [sig]longjmp\|(3). Here all \s-1CPU\s0 registers, the program counter and the stack pointer are switched. Additionally the Pth dispatcher switches also the global Unix \*(C`errno\*(C' variable [see \*(C`pth_mctx.c\*(C' for details] and the signal mask (either implicitly via sigsetjmp\|(3) or in an emulated way via explicit setprocmask\|(2) calls).

The 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\*(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\*(C' and \*(C`pth_pqueue.c\*(C' for details].

Most time critical code sections (especially the dispatcher and event manager) are speeded up by inline functions (implemented as \s-1ANSI\s0 C pre-processor macros). Additionally any debugging code is completely removed from the source when not built with \*(C`-DPTH_DEBUG\*(C' (see Autoconf \*(C`--enable-debug\*(C' option), i.e., not only stub functions remain [see \*(C`pth_debug.c\*(C' for details].

RESTRICTIONS

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\*(C' of the platform vendor). For an asynchronous \s-1DNS\s0 resolver library use the \s-1GNU\s0 adns package from Ian Jackson ( see http://www.gnu.org/software/adns/adns.html ).

HISTORY

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

Pth was then implemented in order to combine the non-preemptive approach of multithreading (which provides better portability and performance) with an \s-1API\s0 similar to the popular one found in Pthread libraries (which provides easy programming).

So the essential idea of the non-preemptive approach was taken over from Peter Simons scheduler. The priority based scheduling algorithm was suggested by Martin Kraemer. Some code inspiration also came from an experimental threading library (rsthreads) written by Robert S. Thau for an ancient internal test version of the Apache webserver. The concept and \s-1API\s0 of message ports was borrowed from AmigaOS' Exec subsystem. The concept and idea for the flexible event mechanism came from Paul Vixie's eventlib (which can be found as a part of \s-1BIND\s0 v8).

BUG REPORTS AND SUPPORT

If you think you have found a bug in Pth, you should send a report as complete as possible to [email protected]. If you can, please try to fix the problem and include a patch, made with '\*(C`diff -u3\*(C'', 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 [email protected] mailing list by sending an Email to [email protected] with `\*(C`subscribe pth-users\*(C'' (or `\*(C`subscribe pth-users\*(C' address' if you want to subscribe from a particular Email address) in the body. Then you can discuss your issues with other Pth users by sending messages to [email protected]. Currently (as of August 2000) you can reach about 110 Pth users on this mailing list. Old postings you can find at http://www.mail-archive.com/[email protected]/.

RELATED TO pth…

`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

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

`The Single \s-1UNIX\s0 Specification, Version 2 - Threads', The Open Group 1997; http://www.opengroup.org/onlinepubs /007908799/xsh/threads.html

\s-1SMI\s0 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 B. Nichols, D. Buttlar, J.P. Farrel: `Pthreads Programming - A \s-1POSIX\s0 Standard for Better Multiprocessing', O'Reilly 1996; \s-1ISBN\s0 1-56592-115-1

B. Lewis, D. J. Berg: `Multithreaded Programming with Pthreads', Sun Microsystems Press, Prentice Hall 1998; \s-1ISBN\s0 0-13-680729-1

B. Lewis, D. J. Berg: `Threads Primer - A Guide To Multithreaded Programming', Prentice Hall 1996; \s-1ISBN\s0 0-13-443698-9

S. J. Norton, M. D. Dipasquale: `Thread Time - The Multithreaded Programming Guide', Prentice Hall 1997; \s-1ISBN\s0 0-13-190067-6

D. R. Butenhof: `Programming with \s-1POSIX\s0 Threads', Addison Wesley 1997; \s-1ISBN\s0 0-201-63392-2 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).

AUTHOR

Ralf S. Engelschall [email protected] www.engelschall.com