1This directory contains the libffi package, which is not part of GCC but 2shipped with GCC as convenience. 3 4Copied without changes from CPython 2.7 head (e04e1f253ed8). 5 6Status 7====== 8 9libffi-2.00 has not been released yet! This is a development snapshot! 10 11libffi-1.20 was released on October 5, 1998. Check the libffi web 12page for updates: <URL:http://sources.redhat.com/libffi/>. 13 14 15What is libffi? 16=============== 17 18Compilers for high level languages generate code that follow certain 19conventions. These conventions are necessary, in part, for separate 20compilation to work. One such convention is the "calling 21convention". The "calling convention" is essentially a set of 22assumptions made by the compiler about where function arguments will 23be found on entry to a function. A "calling convention" also specifies 24where the return value for a function is found. 25 26Some programs may not know at the time of compilation what arguments 27are to be passed to a function. For instance, an interpreter may be 28told at run-time about the number and types of arguments used to call 29a given function. Libffi can be used in such programs to provide a 30bridge from the interpreter program to compiled code. 31 32The libffi library provides a portable, high level programming 33interface to various calling conventions. This allows a programmer to 34call any function specified by a call interface description at run 35time. 36 37Ffi stands for Foreign Function Interface. A foreign function 38interface is the popular name for the interface that allows code 39written in one language to call code written in another language. The 40libffi library really only provides the lowest, machine dependent 41layer of a fully featured foreign function interface. A layer must 42exist above libffi that handles type conversions for values passed 43between the two languages. 44 45 46Supported Platforms and Prerequisites 47===================================== 48 49Libffi has been ported to: 50 51 SunOS 4.1.3 & Solaris 2.x (SPARC-V8, SPARC-V9) 52 53 Irix 5.3 & 6.2 (System V/o32 & n32) 54 55 Intel x86 - Linux (System V ABI) 56 57 Alpha - Linux and OSF/1 58 59 m68k - Linux (System V ABI) 60 61 PowerPC - Linux (System V ABI, Darwin, AIX) 62 63 ARM - Linux (System V ABI) 64 65Libffi has been tested with the egcs 1.0.2 gcc compiler. Chances are 66that other versions will work. Libffi has also been built and tested 67with the SGI compiler tools. 68 69On PowerPC, the tests failed (see the note below). 70 71You must use GNU make to build libffi. SGI's make will not work. 72Sun's probably won't either. 73 74If you port libffi to another platform, please let me know! I assume 75that some will be easy (x86 NetBSD), and others will be more difficult 76(HP). 77 78 79Installing libffi 80================= 81 82[Note: before actually performing any of these installation steps, 83 you may wish to read the "Platform Specific Notes" below.] 84 85First you must configure the distribution for your particular 86system. Go to the directory you wish to build libffi in and run the 87"configure" program found in the root directory of the libffi source 88distribution. 89 90You may want to tell configure where to install the libffi library and 91header files. To do that, use the --prefix configure switch. Libffi 92will install under /usr/local by default. 93 94If you want to enable extra run-time debugging checks use the the 95--enable-debug configure switch. This is useful when your program dies 96mysteriously while using libffi. 97 98Another useful configure switch is --enable-purify-safety. Using this 99will add some extra code which will suppress certain warnings when you 100are using Purify with libffi. Only use this switch when using 101Purify, as it will slow down the library. 102 103Configure has many other options. Use "configure --help" to see them all. 104 105Once configure has finished, type "make". Note that you must be using 106GNU make. SGI's make will not work. Sun's probably won't either. 107You can ftp GNU make from prep.ai.mit.edu:/pub/gnu. 108 109To ensure that libffi is working as advertised, type "make test". 110 111To install the library and header files, type "make install". 112 113 114Using libffi 115============ 116 117 The Basics 118 ---------- 119 120Libffi assumes that you have a pointer to the function you wish to 121call and that you know the number and types of arguments to pass it, 122as well as the return type of the function. 123 124The first thing you must do is create an ffi_cif object that matches 125the signature of the function you wish to call. The cif in ffi_cif 126stands for Call InterFace. To prepare a call interface object, use the 127following function: 128 129ffi_status ffi_prep_cif(ffi_cif *cif, ffi_abi abi, 130 unsigned int nargs, 131 ffi_type *rtype, ffi_type **atypes); 132 133 CIF is a pointer to the call interface object you wish 134 to initialize. 135 136 ABI is an enum that specifies the calling convention 137 to use for the call. FFI_DEFAULT_ABI defaults 138 to the system's native calling convention. Other 139 ABI's may be used with care. They are system 140 specific. 141 142 NARGS is the number of arguments this function accepts. 143 libffi does not yet support vararg functions. 144 145 RTYPE is a pointer to an ffi_type structure that represents 146 the return type of the function. Ffi_type objects 147 describe the types of values. libffi provides 148 ffi_type objects for many of the native C types: 149 signed int, unsigned int, signed char, unsigned char, 150 etc. There is also a pointer ffi_type object and 151 a void ffi_type. Use &ffi_type_void for functions that 152 don't return values. 153 154 ATYPES is a vector of ffi_type pointers. ARGS must be NARGS long. 155 If NARGS is 0, this is ignored. 156 157 158ffi_prep_cif will return a status code that you are responsible 159for checking. It will be one of the following: 160 161 FFI_OK - All is good. 162 163 FFI_BAD_TYPEDEF - One of the ffi_type objects that ffi_prep_cif 164 came across is bad. 165 166 167Before making the call, the VALUES vector should be initialized 168with pointers to the appropriate argument values. 169 170To call the the function using the initialized ffi_cif, use the 171ffi_call function: 172 173void ffi_call(ffi_cif *cif, void *fn, void *rvalue, void **avalues); 174 175 CIF is a pointer to the ffi_cif initialized specifically 176 for this function. 177 178 FN is a pointer to the function you want to call. 179 180 RVALUE is a pointer to a chunk of memory that is to hold the 181 result of the function call. Currently, it must be 182 at least one word in size (except for the n32 version 183 under Irix 6.x, which must be a pointer to an 8 byte 184 aligned value (a long long). It must also be at least 185 word aligned (depending on the return type, and the 186 system's alignment requirements). If RTYPE is 187 &ffi_type_void, this is ignored. If RVALUE is NULL, 188 the return value is discarded. 189 190 AVALUES is a vector of void* that point to the memory locations 191 holding the argument values for a call. 192 If NARGS is 0, this is ignored. 193 194 195If you are expecting a return value from FN it will have been stored 196at RVALUE. 197 198 199 200 An Example 201 ---------- 202 203Here is a trivial example that calls puts() a few times. 204 205 #include <stdio.h> 206 #include <ffi.h> 207 208 int main() 209 { 210 ffi_cif cif; 211 ffi_type *args[1]; 212 void *values[1]; 213 char *s; 214 int rc; 215 216 /* Initialize the argument info vectors */ 217 args[0] = &ffi_type_uint; 218 values[0] = &s; 219 220 /* Initialize the cif */ 221 if (ffi_prep_cif(&cif, FFI_DEFAULT_ABI, 1, 222 &ffi_type_uint, args) == FFI_OK) 223 { 224 s = "Hello World!"; 225 ffi_call(&cif, puts, &rc, values); 226 /* rc now holds the result of the call to puts */ 227 228 /* values holds a pointer to the function's arg, so to 229 call puts() again all we need to do is change the 230 value of s */ 231 s = "This is cool!"; 232 ffi_call(&cif, puts, &rc, values); 233 } 234 235 return 0; 236 } 237 238 239 240 Aggregate Types 241 --------------- 242 243Although libffi has no special support for unions or bit-fields, it is 244perfectly happy passing structures back and forth. You must first 245describe the structure to libffi by creating a new ffi_type object 246for it. Here is the definition of ffi_type: 247 248 typedef struct _ffi_type 249 { 250 unsigned size; 251 short alignment; 252 short type; 253 struct _ffi_type **elements; 254 } ffi_type; 255 256All structures must have type set to FFI_TYPE_STRUCT. You may set 257size and alignment to 0. These will be calculated and reset to the 258appropriate values by ffi_prep_cif(). 259 260elements is a NULL terminated array of pointers to ffi_type objects 261that describe the type of the structure elements. These may, in turn, 262be structure elements. 263 264The following example initializes a ffi_type object representing the 265tm struct from Linux's time.h: 266 267 struct tm { 268 int tm_sec; 269 int tm_min; 270 int tm_hour; 271 int tm_mday; 272 int tm_mon; 273 int tm_year; 274 int tm_wday; 275 int tm_yday; 276 int tm_isdst; 277 /* Those are for future use. */ 278 long int __tm_gmtoff__; 279 __const char *__tm_zone__; 280 }; 281 282 { 283 ffi_type tm_type; 284 ffi_type *tm_type_elements[12]; 285 int i; 286 287 tm_type.size = tm_type.alignment = 0; 288 tm_type.elements = &tm_type_elements; 289 290 for (i = 0; i < 9; i++) 291 tm_type_elements[i] = &ffi_type_sint; 292 293 tm_type_elements[9] = &ffi_type_slong; 294 tm_type_elements[10] = &ffi_type_pointer; 295 tm_type_elements[11] = NULL; 296 297 /* tm_type can now be used to represent tm argument types and 298 return types for ffi_prep_cif() */ 299 } 300 301 302 303Platform Specific Notes 304======================= 305 306 Intel x86 307 --------- 308 309There are no known problems with the x86 port. 310 311 Sun SPARC - SunOS 4.1.3 & Solaris 2.x 312 ------------------------------------- 313 314You must use GNU Make to build libffi on Sun platforms. 315 316 MIPS - Irix 5.3 & 6.x 317 --------------------- 318 319Irix 6.2 and better supports three different calling conventions: o32, 320n32 and n64. Currently, libffi only supports both o32 and n32 under 321Irix 6.x, but only o32 under Irix 5.3. Libffi will automatically be 322configured for whichever calling convention it was built for. 323 324By default, the configure script will try to build libffi with the GNU 325development tools. To build libffi with the SGI development tools, set 326the environment variable CC to either "cc -32" or "cc -n32" before 327running configure under Irix 6.x (depending on whether you want an o32 328or n32 library), or just "cc" for Irix 5.3. 329 330With the n32 calling convention, when returning structures smaller 331than 16 bytes, be sure to provide an RVALUE that is 8 byte aligned. 332Here's one way of forcing this: 333 334 double struct_storage[2]; 335 my_small_struct *s = (my_small_struct *) struct_storage; 336 /* Use s for RVALUE */ 337 338If you don't do this you are liable to get spurious bus errors. 339 340"long long" values are not supported yet. 341 342You must use GNU Make to build libffi on SGI platforms. 343 344 ARM - System V ABI 345 ------------------ 346 347The ARM port was performed on a NetWinder running ARM Linux ELF 348(2.0.31) and gcc 2.8.1. 349 350 351 352 PowerPC System V ABI 353 -------------------- 354 355There are two `System V ABI's which libffi implements for PowerPC. 356They differ only in how small structures are returned from functions. 357 358In the FFI_SYSV version, structures that are 8 bytes or smaller are 359returned in registers. This is what GCC does when it is configured 360for solaris, and is what the System V ABI I have (dated September 3611995) says. 362 363In the FFI_GCC_SYSV version, all structures are returned the same way: 364by passing a pointer as the first argument to the function. This is 365what GCC does when it is configured for linux or a generic sysv 366target. 367 368EGCS 1.0.1 (and probably other versions of EGCS/GCC) also has a 369inconsistency with the SysV ABI: When a procedure is called with many 370floating-point arguments, some of them get put on the stack. They are 371all supposed to be stored in double-precision format, even if they are 372only single-precision, but EGCS stores single-precision arguments as 373single-precision anyway. This causes one test to fail (the `many 374arguments' test). 375 376 377What's With The Crazy Comments? 378=============================== 379 380You might notice a number of cryptic comments in the code, delimited 381by /*@ and @*/. These are annotations read by the program LCLint, a 382tool for statically checking C programs. You can read all about it at 383<http://larch-www.lcs.mit.edu:8001/larch/lclint/index.html>. 384 385 386History 387======= 388 3891.20 Oct-5-98 390 Raffaele Sena produces ARM port. 391 3921.19 Oct-5-98 393 Fixed x86 long double and long long return support. 394 m68k bug fixes from Andreas Schwab. 395 Patch for DU assembler compatibility for the Alpha from Richard 396 Henderson. 397 3981.18 Apr-17-98 399 Bug fixes and MIPS configuration changes. 400 4011.17 Feb-24-98 402 Bug fixes and m68k port from Andreas Schwab. PowerPC port from 403 Geoffrey Keating. Various bug x86, Sparc and MIPS bug fixes. 404 4051.16 Feb-11-98 406 Richard Henderson produces Alpha port. 407 4081.15 Dec-4-97 409 Fixed an n32 ABI bug. New libtool, auto* support. 410 4111.14 May-13-97 412 libtool is now used to generate shared and static libraries. 413 Fixed a minor portability problem reported by Russ McManus 414 <mcmanr@eq.gs.com>. 415 4161.13 Dec-2-96 417 Added --enable-purify-safety to keep Purify from complaining 418 about certain low level code. 419 Sparc fix for calling functions with < 6 args. 420 Linux x86 a.out fix. 421 4221.12 Nov-22-96 423 Added missing ffi_type_void, needed for supporting void return 424 types. Fixed test case for non MIPS machines. Cygnus Support 425 is now Cygnus Solutions. 426 4271.11 Oct-30-96 428 Added notes about GNU make. 429 4301.10 Oct-29-96 431 Added configuration fix for non GNU compilers. 432 4331.09 Oct-29-96 434 Added --enable-debug configure switch. Clean-ups based on LCLint 435 feedback. ffi_mips.h is always installed. Many configuration 436 fixes. Fixed ffitest.c for sparc builds. 437 4381.08 Oct-15-96 439 Fixed n32 problem. Many clean-ups. 440 4411.07 Oct-14-96 442 Gordon Irlam rewrites v8.S again. Bug fixes. 443 4441.06 Oct-14-96 445 Gordon Irlam improved the sparc port. 446 4471.05 Oct-14-96 448 Interface changes based on feedback. 449 4501.04 Oct-11-96 451 Sparc port complete (modulo struct passing bug). 452 4531.03 Oct-10-96 454 Passing struct args, and returning struct values works for 455 all architectures/calling conventions. Expanded tests. 456 4571.02 Oct-9-96 458 Added SGI n32 support. Fixed bugs in both o32 and Linux support. 459 Added "make test". 460 4611.01 Oct-8-96 462 Fixed float passing bug in mips version. Restructured some 463 of the code. Builds cleanly with SGI tools. 464 4651.00 Oct-7-96 466 First release. No public announcement. 467 468 469Authors & Credits 470================= 471 472libffi was written by Anthony Green <green@cygnus.com>. 473 474Portions of libffi were derived from Gianni Mariani's free gencall 475library for Silicon Graphics machines. 476 477The closure mechanism was designed and implemented by Kresten Krab 478Thorup. 479 480The Sparc port was derived from code contributed by the fine folks at 481Visible Decisions Inc <http://www.vdi.com>. Further enhancements were 482made by Gordon Irlam at Cygnus Solutions <http://www.cygnus.com>. 483 484The Alpha port was written by Richard Henderson at Cygnus Solutions. 485 486Andreas Schwab ported libffi to m68k Linux and provided a number of 487bug fixes. 488 489Geoffrey Keating ported libffi to the PowerPC. 490 491Raffaele Sena ported libffi to the ARM. 492 493Jesper Skov and Andrew Haley both did more than their fair share of 494stepping through the code and tracking down bugs. 495 496Thanks also to Tom Tromey for bug fixes and configuration help. 497 498Thanks to Jim Blandy, who provided some useful feedback on the libffi 499interface. 500 501If you have a problem, or have found a bug, please send a note to 502green@cygnus.com. 503