NetScape Portable Runtime (NSPR) provides platform independence for non-GUI operating system facilities. These facilities include threads, thread synchronization, normal file and network I/O, interval timing and calendar time, basic memory management (malloc and free) and shared library linking.
A good portion of the library's purpose, and perhaps the primary purpose in the Gromit environment, was to provide the underpinnings of the Java VM, more or less mapping the sys layer that Sun defined for the porting of the Java VM to various platforms. NSPR went beyond that requirement in some areas and since it was also the platform independent layer for most of the servers produced by Netscape. It was expected and preferred that existing code be restructured and perhaps even rewritten in order to use the NSPR API. It is not a goal to provide a platform for the porting into Netscape of externally developed code.
At the time of writing the current generation of NSPR was known as NSPR20. The first generation of NSPR was originally conceived just to satisfy the requirements of porting Java to various host environments. NSPR20, an effort started in 1996, built on that original idea, though very little is left of the original code. (The "20" in "NSPR20" does not mean "version 2.0" but rather "second generation".) Many of the concepts have been reformed, expanded, and matured. Today NSPR may still be appropriate as the platform dependent layer under Java, but its primary application is supporting clients written entirely in C or C++.
NSPR's goal is to provide uniform service over a wide range of operating system environments. It strives to not export the lowest common denominator, but to exploit the best features of each operating system on which it runs, and still provide a uniform service across a wide range of host offerings.
Threads are the major feature of NSPR. The industry's offering of threads is quite sundry. NSPR, while far from perfect, does provide a single API to which clients may program and expect reasonably consistent behavior. The operating systems provide everything from no concept of threading at all up to and including sophisticated, scalable and efficient implementations. NSPR makes as much use of what the systems offer as it can. It is a goal of NSPR that NSPR impose as little overhead as possible in accessing those appropriate system features.
Thread synchronization is loosely based on Monitors as described by C.A.R. Hoare in Monitors: An operating system structuring concept , Communications of the ACM, 17(10), October 1974 and then formalized by Xerox' Mesa programming language ("Mesa Language Manual", J.G. Mitchell et al, Xerox PARC, CSL-79-3 (Apr 1979)). This mechanism provides the basic mutual exclusion (mutex) and thread notification facilities (condition variables) implemented by NSPR. Additionally, NSPR provides synchronization methods more suited for use by Java. The Java-like facilities include monitor reentrancy, implicit and tightly bound notification capabilities with the ability to associate the synchronization objects dynamically.
NSPR's I/O is a slightly augmented BSD sockets model that allows arbitrary layering. It was originally intended to export synchronous I/O methods only, relying on threads to provide the concurrency needed for complex applications. That method of operation is preferred though it is possible to configure the network I/O channels as non-blocking in the traditional sense.
Part of NSPR deals with manipulation of network addresses. NSPR defines a network address object that is Internet Protocol (IP) centric. While the object is not declared as opaque, the API provides methods that allow and encourage clients to treat the addresses as polymorphic items. The goal in this area is to provide a migration path between IPv4 and IPv6. To that end it is possible to perform translations of ASCII strings (DNS names) into NSPR's network address structures, with no regard to whether the addressing technology is IPv4 or IPv6.
Timing facilities are available in two forms: interval timing and calendar functions.
Interval timers are based on a free running, 32-bit, platform dependent resolution timer. Such timers are normally used to specify timeouts on I/O, waiting on condition variables and other rudimentary thread scheduling. Since these timers have finite namespace and are free running, they can wrap at any time. NSPR does not provide an epoch , but expects clients to deal with that issue. The granularity of the timers is guaranteed to be between 10 microseconds and 1 millisecond. This allows a minimal timer period in of approximately 12 hours. But in order to deal with the wrap-around issue, only half that namespace may be utilized. Therefore, the minimal usable interval available from the timers is slightly less than six hours.
Calendar times are 64-bit signed numbers with units of microseconds. The epoch for calendar times is midnight, January 1, 1970, Greenwich Mean Time. Negative times extend to times before 1970, and positive numbers forward. Use of 64 bits allows a representation of times approximately in the range of -30000 to the year 30000. There is a structural representation (i.e., exploded view), routines to acquire the current time from the host system, and convert them to and from the 64-bit and structural representation. Additionally there are routines to convert to and from most well-known forms of ASCII into the 64-bit NSPR representation.
NSPR provides API to perform the basic malloc, calloc, realloc and free functions. Depending on the platform, the functions may be implemented almost entirely in the NSPR runtime or simply shims that call immediately into the host operating system's offerings.
Support for linking (shared library loading and unloading) is part of NSPR's feature set. In most cases this is simply a smoothing over of the facilities offered by the various platform providers.
NSPR is applicable as a platform on which to write threaded applications that need to be ported to multiple platforms. The current implementation supports Macintosh (PPC), WIN-32 (WinNT, Win9x) and 20 versions of UNIX and is still expanding. The basic API is stable and expected to remain that way.
NSPR is functionally complete and has entered a mode of sustaining engineering. As operating system vendors issue new releases of their operating systems, NSPR will be moved forward to these new releases by interested players.