1============== 2System Library 3============== 4 5Abstract 6======== 7 8This document provides some details on LLVM's System Library, located in the 9source at ``lib/System`` and ``include/llvm/System``. The library's purpose is 10to shield LLVM from the differences between operating systems for the few 11services LLVM needs from the operating system. Much of LLVM is written using 12portability features of standard C++. However, in a few areas, system dependent 13facilities are needed and the System Library is the wrapper around those system 14calls. 15 16By centralizing LLVM's use of operating system interfaces, we make it possible 17for the LLVM tool chain and runtime libraries to be more easily ported to new 18platforms since (theoretically) only ``lib/System`` needs to be ported. This 19library also unclutters the rest of LLVM from #ifdef use and special cases for 20specific operating systems. Such uses are replaced with simple calls to the 21interfaces provided in ``include/llvm/System``. 22 23Note that the System Library is not intended to be a complete operating system 24wrapper (such as the Adaptive Communications Environment (ACE) or Apache 25Portable Runtime (APR)), but only provides the functionality necessary to 26support LLVM. 27 28The System Library was written by Reid Spencer who formulated the design based 29on similar work originating from the eXtensible Programming System (XPS). 30Several people helped with the effort; especially, Jeff Cohen and Henrik Bach 31on the Win32 port. 32 33Keeping LLVM Portable 34===================== 35 36In order to keep LLVM portable, LLVM developers should adhere to a set of 37portability rules associated with the System Library. Adherence to these rules 38should help the System Library achieve its goal of shielding LLVM from the 39variations in operating system interfaces and doing so efficiently. The 40following sections define the rules needed to fulfill this objective. 41 42Don't Include System Headers 43---------------------------- 44 45Except in ``lib/System``, no LLVM source code should directly ``#include`` a 46system header. Care has been taken to remove all such ``#includes`` from LLVM 47while ``lib/System`` was being developed. Specifically this means that header 48files like "``unistd.h``", "``windows.h``", "``stdio.h``", and "``string.h``" 49are forbidden to be included by LLVM source code outside the implementation of 50``lib/System``. 51 52To obtain system-dependent functionality, existing interfaces to the system 53found in ``include/llvm/System`` should be used. If an appropriate interface is 54not available, it should be added to ``include/llvm/System`` and implemented in 55``lib/System`` for all supported platforms. 56 57Don't Expose System Headers 58--------------------------- 59 60The System Library must shield LLVM from **all** system headers. To obtain 61system level functionality, LLVM source must ``#include "llvm/System/Thing.h"`` 62and nothing else. This means that ``Thing.h`` cannot expose any system header 63files. This protects LLVM from accidentally using system specific functionality 64and only allows it via the ``lib/System`` interface. 65 66Use Standard C Headers 67---------------------- 68 69The **standard** C headers (the ones beginning with "c") are allowed to be 70exposed through the ``lib/System`` interface. These headers and the things they 71declare are considered to be platform agnostic. LLVM source files may include 72them directly or obtain their inclusion through ``lib/System`` interfaces. 73 74Use Standard C++ Headers 75------------------------ 76 77The **standard** C++ headers from the standard C++ library and standard 78template library may be exposed through the ``lib/System`` interface. These 79headers and the things they declare are considered to be platform agnostic. 80LLVM source files may include them or obtain their inclusion through 81``lib/System`` interfaces. 82 83High Level Interface 84-------------------- 85 86The entry points specified in the interface of ``lib/System`` must be aimed at 87completing some reasonably high level task needed by LLVM. We do not want to 88simply wrap each operating system call. It would be preferable to wrap several 89operating system calls that are always used in conjunction with one another by 90LLVM. 91 92For example, consider what is needed to execute a program, wait for it to 93complete, and return its result code. On Unix, this involves the following 94operating system calls: ``getenv``, ``fork``, ``execve``, and ``wait``. The 95correct thing for ``lib/System`` to provide is a function, say 96``ExecuteProgramAndWait``, that implements the functionality completely. what 97we don't want is wrappers for the operating system calls involved. 98 99There must **not** be a one-to-one relationship between operating system 100calls and the System library's interface. Any such interface function will be 101suspicious. 102 103No Unused Functionality 104----------------------- 105 106There must be no functionality specified in the interface of ``lib/System`` 107that isn't actually used by LLVM. We're not writing a general purpose operating 108system wrapper here, just enough to satisfy LLVM's needs. And, LLVM doesn't 109need much. This design goal aims to keep the ``lib/System`` interface small and 110understandable which should foster its actual use and adoption. 111 112No Duplicate Implementations 113---------------------------- 114 115The implementation of a function for a given platform must be written exactly 116once. This implies that it must be possible to apply a function's 117implementation to multiple operating systems if those operating systems can 118share the same implementation. This rule applies to the set of operating 119systems supported for a given class of operating system (e.g. Unix, Win32). 120 121No Virtual Methods 122------------------ 123 124The System Library interfaces can be called quite frequently by LLVM. In order 125to make those calls as efficient as possible, we discourage the use of virtual 126methods. There is no need to use inheritance for implementation differences, it 127just adds complexity. The ``#include`` mechanism works just fine. 128 129No Exposed Functions 130-------------------- 131 132Any functions defined by system libraries (i.e. not defined by ``lib/System``) 133must not be exposed through the ``lib/System`` interface, even if the header 134file for that function is not exposed. This prevents inadvertent use of system 135specific functionality. 136 137For example, the ``stat`` system call is notorious for having variations in the 138data it provides. ``lib/System`` must not declare ``stat`` nor allow it to be 139declared. Instead it should provide its own interface to discovering 140information about files and directories. Those interfaces may be implemented in 141terms of ``stat`` but that is strictly an implementation detail. The interface 142provided by the System Library must be implemented on all platforms (even those 143without ``stat``). 144 145No Exposed Data 146--------------- 147 148Any data defined by system libraries (i.e. not defined by ``lib/System``) must 149not be exposed through the ``lib/System`` interface, even if the header file 150for that function is not exposed. As with functions, this prevents inadvertent 151use of data that might not exist on all platforms. 152 153Minimize Soft Errors 154-------------------- 155 156Operating system interfaces will generally provide error results for every 157little thing that could go wrong. In almost all cases, you can divide these 158error results into two groups: normal/good/soft and abnormal/bad/hard. That is, 159some of the errors are simply information like "file not found", "insufficient 160privileges", etc. while other errors are much harder like "out of space", "bad 161disk sector", or "system call interrupted". We'll call the first group "*soft*" 162errors and the second group "*hard*" errors. 163 164``lib/System`` must always attempt to minimize soft errors. This is a design 165requirement because the minimization of soft errors can affect the granularity 166and the nature of the interface. In general, if you find that you're wanting to 167throw soft errors, you must review the granularity of the interface because it 168is likely you're trying to implement something that is too low level. The rule 169of thumb is to provide interface functions that **can't** fail, except when 170faced with hard errors. 171 172For a trivial example, suppose we wanted to add an "``OpenFileForWriting``" 173function. For many operating systems, if the file doesn't exist, attempting to 174open the file will produce an error. However, ``lib/System`` should not simply 175throw that error if it occurs because its a soft error. The problem is that the 176interface function, ``OpenFileForWriting`` is too low level. It should be 177``OpenOrCreateFileForWriting``. In the case of the soft "doesn't exist" error, 178this function would just create it and then open it for writing. 179 180This design principle needs to be maintained in ``lib/System`` because it 181avoids the propagation of soft error handling throughout the rest of LLVM. 182Hard errors will generally just cause a termination for an LLVM tool so don't 183be bashful about throwing them. 184 185Rules of thumb: 186 187#. Don't throw soft errors, only hard errors. 188 189#. If you're tempted to throw a soft error, re-think the interface. 190 191#. Handle internally the most common normal/good/soft error conditions 192 so the rest of LLVM doesn't have to. 193 194No throw Specifications 195----------------------- 196 197None of the ``lib/System`` interface functions may be declared with C++ 198``throw()`` specifications on them. This requirement makes sure that the 199compiler does not insert additional exception handling code into the interface 200functions. This is a performance consideration: ``lib/System`` functions are at 201the bottom of many call chains and as such can be frequently called. We need 202them to be as efficient as possible. However, no routines in the system 203library should actually throw exceptions. 204 205Code Organization 206----------------- 207 208Implementations of the System Library interface are separated by their general 209class of operating system. Currently only Unix and Win32 classes are defined 210but more could be added for other operating system classifications. To 211distinguish which implementation to compile, the code in ``lib/System`` uses 212the ``LLVM_ON_UNIX`` and ``LLVM_ON_WIN32`` ``#defines`` provided via configure 213through the ``llvm/Config/config.h`` file. Each source file in ``lib/System``, 214after implementing the generic (operating system independent) functionality 215needs to include the correct implementation using a set of 216``#if defined(LLVM_ON_XYZ)`` directives. For example, if we had 217``lib/System/File.cpp``, we'd expect to see in that file: 218 219.. code-block:: c++ 220 221 #if defined(LLVM_ON_UNIX) 222 #include "Unix/File.cpp" 223 #endif 224 #if defined(LLVM_ON_WIN32) 225 #include "Win32/File.cpp" 226 #endif 227 228The implementation in ``lib/System/Unix/File.cpp`` should handle all Unix 229variants. The implementation in ``lib/System/Win32/File.cpp`` should handle all 230Win32 variants. What this does is quickly differentiate the basic class of 231operating system that will provide the implementation. The specific details for 232a given platform must still be determined through the use of ``#ifdef``. 233 234Consistent Semantics 235-------------------- 236 237The implementation of a ``lib/System`` interface can vary drastically between 238platforms. That's okay as long as the end result of the interface function is 239the same. For example, a function to create a directory is pretty straight 240forward on all operating system. System V IPC on the other hand isn't even 241supported on all platforms. Instead of "supporting" System V IPC, 242``lib/System`` should provide an interface to the basic concept of 243inter-process communications. The implementations might use System V IPC if 244that was available or named pipes, or whatever gets the job done effectively 245for a given operating system. In all cases, the interface and the 246implementation must be semantically consistent. 247 248