1Noteworthy changes in GCC after EGCS 1.1. 2----------------------------------------- 3 4Target specific NEWS 5 6 RS6000/PowerPC: -mcpu=401 was added as an alias for -mcpu=403. -mcpu=e603e 7 was added to do -mcpu=603e and -msoft-float. 8 9Noteworthy changes in GCC for EGCS 1.1. 10--------------------------------------- 11 12The compiler now implements global common subexpression elimination (gcse) as 13well as global constant/copy propagation. (link to gcse page). 14 15More major improvements have been made to the alias analysis code. A new 16option to allow front-ends to provide alias information to the optimizers 17has also been added (-fstrict-aliasing). -fstrict-aliasing is off by default 18now, but will be enabled by default in the future. (link to alias page) 19 20Major changes continue in the exception handling support. This release 21includes some changes to reduce static overhead for exception handling. It 22also includes some major changes to the setjmp/longjmp based EH mechanism to 23make it less pessimistic. And finally, major infrastructure improvements 24to the dwarf2 EH mechanism have been made to make our EH support extensible. 25 26We have fixed the infamous security problems with temporary files. 27 28The "regmove" optimization pass has been nearly completely rewritten. It now 29uses much more information about the target to determine profitability of 30transformations. 31 32The compiler now recomputes register usage information immediately before 33register allocation. Previously such information was only not kept up to 34date after instruction combination which led to poor register allocation 35choices by our priority based register allocator. 36 37The register reloading phase of the compiler has been improved to better 38optimize spill code. This primarily helps targets which generate lots of 39spills (like the x86 ports and many register poor embedded ports). 40 41A few changes in the heuristics used by the register allocator and scheduler 42have been made which can significantly improve performance for certain 43applications. 44 45The compiler's branch shortening algorithms have been significantly improved 46to work better on targets which align jump targets. 47 48The compiler now supports the "ADDRESSOF" optimization which can significantly 49reduce the overhead for certain inline calls (and inline calls in general). 50 51The compiler now supports a code size optimization switch (-Os). When enabled 52the compiler will prefer optimizations which improve code size over those 53which improve code speed. 54 55The compiler has been improved to completely eliminate library calls which 56compute constant values. This is particularly useful on machines which 57do not have integer mul/div or floating point support on-chip. 58 59GCC now supports a "--help" option to print detailed help information. 60 61cpplib has been greatly improved. It is probably useable for some sites now 62(major missing feature is trigraphs). 63 64Memory footprint for the compiler has been significantly reduced for certain 65pathalogical cases. 66 67Build time improvements for targets which support lots of sched parameters 68(alpha and mips primarily). 69 70Compile time for certain programs using large constant initializers has been 71improved (effects glibc significantly). 72 73Plus an incredible number of infrastructure changes, warning fixes, bugfixes 74and local optimizations. 75 76Various improvements have been made to better support cross compilations. They 77are still not easy, but they are improving. 78 79Target specific NEWS 80 81 Sparc: Now includes V8 plus and V9 support, lots of tuning for Ultrasparcs 82 and uses the Haifa scheduler by default. 83 84 Alpha: EV6 tuned, optimized expansion of memcpy/bzero. 85 86 x86: Data in the static store is aligned per Intel recommendations. Jump 87 targets are aligned per Intel recommendations. Improved epilogue 88 sequences for Pentium chips. Backend improvements which should help 89 register allocation on all x86 variants. Support for PPro conditional 90 move instructions has been fixed and enabled. Random changes 91 throughout the port to make generated code more Pentium friendly. 92 Improved support for 64bit integer operations. 93 Unixware 7, a System V Release 5 target is now supported. 94 SCO OpenServer targets can support GAS. See gcc/INSTALL for details. 95 96 RS6000/PowerPC: Includes AIX4.3 support as well as PowerPC64 support. 97 Haifa instruction scheduling is enabled by default now. 98 99 MIPS: Multiply/Multiply-Add support has been largely rewritten to generate 100 more efficient code. Includes mips16 support. 101 102 M68K: Various micro-optimizations and Coldfire fixes. 103 104 M32r: Major improvements to this port. 105 106 Arm: Includes Thumb and super interworking support. 107 108EGCS includes all gcc2 changes up to and including the June 9, 1998 snapshot. 109 110 111Noteworthy changes in GCC version 2.8.1 112--------------------------------------- 113 114Numerous bugs have been fixed and some minor performance 115improvements (compilation speed) have been made. 116 117Noteworthy changes in GCC version 2.8.0 118--------------------------------------- 119 120A major change in this release is the addition of a framework for 121exception handling, currently used by C++. Many internal changes and 122optimization improvements have been made. These increase the 123maintainability and portability of GCC. GCC now uses autoconf to 124compute many host parameters. 125 126The following lists changes that add new features or targets. 127 128See cp/NEWS for new features of C++ in this release. 129 130New tools and features: 131 132 The Dwarf 2 debugging information format is supported on ELF systems, and 133 is the default for -ggdb where available. It can also be used for C++. 134 The Dwarf version 1 debugging format is also permitted for C++, but 135 does not work well. 136 137 gcov.c is provided for test coverage analysis and branch profiling 138 analysis is also supported; see -fprofile-arcs, -ftest-coverage, 139 and -fbranch-probabilities. 140 141 Support for the Checker memory checking tool. 142 143 New switch, -fstack-check, to check for stack overflow on systems that 144 don't have such built into their ABI. 145 146 New switches, -Wundef and -Wno-undef to warn if an undefined identifier 147 is evaluated in an #if directive. 148 149 Options -Wall and -Wimplicit now cause GCC to warn about implicit int 150 in declarations (e.g. `register i;'), since the C Standard committee 151 has decided to disallow this in the next revision of the standard; 152 -Wimplicit-function-declarations and -Wimplicit-int are subsets of 153 this. 154 155 Option -Wsign-compare causes GCC to warn about comparison of signed and 156 unsigned values. 157 158 Add -dI option of cccp for cxref. 159 160New features in configuration, installation and specs file handling: 161 162 New option --enable-c-cpplib to configure script. 163 164 You can use --with-cpu on the configure command to specify the default 165 CPU that GCC should generate code for. 166 167 The -specs=file switch allows you to override default specs used in 168 invoking programs like cc1, as, etc. 169 170 Allow including one specs file from another and renaming a specs 171 variable. 172 173 You can now relocate all GCC files with a single environment variable 174 or a registry entry under Windows 95 and Windows NT. 175 176Changes in Objective-C: 177 178 The Objective-C Runtime Library has been made thread-safe. 179 180 The Objective-C Runtime Library contains an interface for creating 181 mutexes, condition mutexes, and threads; it requires a back-end 182 implementation for the specific platform and/or thread package. 183 Currently supported are DEC/OSF1, IRIX, Mach, OS/2, POSIX, PCThreads, 184 Solaris, and Windows32. The --enable-threads parameter can be used 185 when configuring GCC to enable and select a thread back-end. 186 187 Objective-C is now configured as separate front-end language to GCC, 188 making it more convenient to conditionally build it. 189 190 The internal structures of the Objective-C Runtime Library have 191 changed sufficiently to warrant a new version number; now version 8. 192 Programs compiled with an older version must be recompiled. 193 194 The Objective-C Runtime Library can be built as a DLL on Windows 95 195 and Windows NT systems. 196 197 The Objective-C Runtime Library implements +load. 198 199The following new targets are supported (see also list under each 200individual CPU below): 201 202 Embedded target m32r-elf. 203 Embedded Hitachi Super-H using ELF. 204 RTEMS real-time system on various CPU targets. 205 ARC processor. 206 NEC V850 processor. 207 Matsushita MN10200 processor. 208 Matsushita MN10300 processor. 209 Sparc and PowerPC running on VxWorks. 210 Support both glibc versions 1 and 2 on Linux-based GNU systems. 211 212New features for DEC Alpha systems: 213 214 Allow detailed specification of IEEE fp support: 215 -mieee, -mieee-with-inexact, and -mieee-conformant 216 -mfp-trap-mode=xxx, -mfp-round-mode=xxx, -mtrap-precision=xxx 217 -mcpu=xxx for CPU selection 218 Support scheduling parameters for EV5. 219 Add support for BWX, CIX, and MAX instruction set extensions. 220 Support Linux-based GNU systems. 221 Support VMS. 222 223Additional supported processors and systems for MIPS targets: 224 225 MIPS4 instruction set. 226 R4100, R4300 and R5000 processors. 227 N32 and N64 ABI. 228 IRIX 6.2. 229 SNI SINIX. 230 231New features for Intel x86 family: 232 233 Add scheduling parameters for Pentium and Pentium Pro. 234 Support stabs on Solaris-x86. 235 Intel x86 processors running the SCO OpenServer 5 family. 236 Intel x86 processors running DG/UX. 237 Intel x86 using Cygwin32 or Mingw32 on Windows 95 and Windows NT. 238 239New features for Motorola 68k family: 240 241 Support for 68060 processor. 242 More consistent switches to specify processor. 243 Motorola 68k family running AUX. 244 68040 running pSOS, ELF object files, DBX debugging. 245 Coldfire variant of Motorola m68k family. 246 247New features for the HP PA RISC: 248 249 -mspace and -mno-space 250 -mlong-load-store and -mno-long-load-store 251 -mbig-switch -mno-big-switch 252 253 GCC on the PA requires either gas-2.7 or the HP assembler; for best 254 results using GAS is highly recommended. GAS is required for -g and 255 exception handling support. 256 257New features for SPARC-based systems: 258 259 The ultrasparc cpu. 260 The sparclet cpu, supporting only a.out file format. 261 Sparc running SunOS 4 with the GNU assembler. 262 Sparc running the Linux-based GNU system. 263 Embedded Sparc processors running the ELF object file format. 264 -mcpu=xxx 265 -mtune=xxx 266 -malign-loops=xxx 267 -malign-jumps=xxx 268 -malign-functions=xxx 269 -mimpure-text and -mno-impure-text 270 271 Options -mno-v8 and -mno-sparclite are no longer supported on SPARC 272 targets. Options -mcypress, -mv8, -msupersparc, -msparclite, -mf930, 273 and -mf934 are deprecated and will be deleted in GCC 2.9. Use 274 -mcpu=xxx instead. 275 276New features for rs6000 and PowerPC systems: 277 278 Solaris 2.51 running on PowerPC's. 279 The Linux-based GNU system running on PowerPC's. 280 -mcpu=604e,602,603e,620,801,823,mpc505,821,860,power2 281 -mtune=xxx 282 -mrelocatable-lib, -mno-relocatable-lib 283 -msim, -mmve, -memb 284 -mupdate, -mno-update 285 -mfused-madd, -mno-fused-madd 286 287 -mregnames 288 -meabi 289 -mcall-linux, -mcall-solaris, -mcall-sysv-eabi, -mcall-sysv-noeabi 290 -msdata, -msdata=none, -msdata=default, -msdata=sysv, -msdata=eabi 291 -memb, -msim, -mmvme 292 -myellowknife, -mads 293 wchar_t is now of type long as per the ABI, not unsigned short. 294 -p/-pg support 295 -mcpu=403 now implies -mstrict-align. 296 Implement System V profiling. 297 298 Aix 4.1 GCC targets now default to -mcpu=common so that programs 299 compiled can be moved between rs6000 and powerpc based systems. A 300 consequence of this is that -static won't work, and that some programs 301 may be slightly slower. 302 303 You can select the default value to use for -mcpu=xxx on rs6000 and 304 powerpc targets by using the --with-cpu=xxx option when configuring the 305 compiler. In addition, a new options, -mtune=xxx was added that 306 selects the machine to schedule for but does not select the 307 architecture level. 308 309 Directory names used for storing the multilib libraries on System V 310 and embedded PowerPC systems have been shortened to work with commands 311 like tar that have fixed limits on pathname size. 312 313New features for the Hitachi H8/300(H): 314 315 -malign-300 316 -ms (for the Hitachi H8/S processor) 317 -mint32 318 319New features for the ARM: 320 321 -march=xxx, -mtune=xxx, -mcpu=xxx 322 Support interworking with Thumb code. 323 ARM processor with a.out object format, COFF, or AOF assembler. 324 ARM on "semi-hosted" platform. 325 ARM running NetBSD. 326 ARM running the Linux-based GNU system. 327 328New feature for Solaris systems: 329 330 GCC installation no longer makes a copy of system include files, 331 thus insulating GCC better from updates to the operating system. 332 333 334Noteworthy changes in GCC version 2.7.2 335--------------------------------------- 336 337A few bugs have been fixed (most notably the generation of an 338invalid assembler opcode on some RS/6000 systems). 339 340Noteworthy changes in GCC version 2.7.1 341--------------------------------------- 342 343This release fixes numerous bugs (mostly minor) in GCC 2.7.0, but 344also contains a few new features, mostly related to specific targets. 345 346Major changes have been made in code to support Windows NT. 347 348The following new targets are supported: 349 350 2.9 BSD on PDP-11 351 Linux on m68k 352 HP/UX version 10 on HP PA RISC (treated like version 9) 353 DEC Alpha running Windows NT 354 355When parsing C, GCC now recognizes C++ style `//' comments unless you 356specify `-ansi' or `-traditional'. 357 358The PowerPC System V targets (powerpc-*-sysv, powerpc-*-eabi) now use the 359calling sequence specified in the System V Application Binary Interface 360Processor Supplement (PowerPC Processor ABI Supplement) rather than the calling 361sequence used in GCC version 2.7.0. That calling sequence was based on the AIX 362calling sequence without function descriptors. To compile code for that older 363calling sequence, either configure the compiler for powerpc-*-eabiaix or use 364the -mcall-aix switch when compiling and linking. 365 366Noteworthy changes in GCC version 2.7.0 367--------------------------------------- 368 369GCC now works better on systems that use ".obj" and ".exe" instead of 370".o" and no extension. This involved changes to the driver program, 371gcc.c, to convert ".o" names to ".obj" and to GCC's Makefile to use 372".obj" and ".exe" in filenames that are not targets. In order to 373build GCC on such systems, you may need versions of GNU make and/or 374compatible shells. At this point, this support is preliminary. 375 376Object file extensions of ".obj" and executable file extensions of 377".exe" are allowed when using appropriate version of GNU Make. 378 379Numerous enhancements were made to the __attribute__ facility including 380more attributes and more places that support it. We now support the 381"packed", "nocommon", "noreturn", "volatile", "const", "unused", 382"transparent_union", "constructor", "destructor", "mode", "section", 383"align", "format", "weak", and "alias" attributes. Each of these 384names may also be specified with added underscores, e.g., "__packed__". 385__attribute__ may now be applied to parameter definitions, function 386definitions, and structure, enum, and union definitions. 387 388GCC now supports returning more structures in registers, as specified by 389many calling sequences (ABIs), such as on the HP PA RISC. 390 391A new option '-fpack-struct' was added to automatically pack all structure 392members together without holes. 393 394There is a new library (cpplib) and program (cppmain) that at some 395point will replace cpp (aka cccp). To use cppmain as cpp now, pass 396the option CCCP=cppmain to make. The library is already used by the 397fix-header program, which should speed up the fixproto script. 398 399New options for supported targets: 400 401 GNU on many targets. 402 NetBSD on MIPS, m68k, VAX, and x86. 403 LynxOS on x86, m68k, Sparc, and RS/6000. 404 VxWorks on many targets. 405 406 Windows/NT on x86 architecture. Initial support for Windows/NT on Alpha 407 (not fully working). 408 409 Many embedded targets, specifically UDI on a29k, aout, coff, elf, 410 and vsta "operating systems" on m68k, m88k, mips, sparc, and x86. 411 412Additional support for x86 (i386, i486, and Pentium): 413 414 Work with old and new linkers for Linux-based GNU systems, 415 supporting both a.out and ELF. 416 FreeBSD on x86. 417 Stdcall convention. 418 -malign-double, -mregparm=, -malign-loops= and -malign-jumps= switches. 419 On ISC systems, support -Xp like -posix. 420 421Additions for RS/6000: 422 423 Instruction scheduling information for PowerPC 403. 424 AIX 4.1 on PowerPC. 425 -mstring and -mno-string. 426 -msoft-float and floating-point emulation included. 427 Preliminary support for PowerPC System V.4 with or without the GNU as. 428 Preliminary support for EABI. 429 Preliminary support for 64-bit systems. 430 Both big and little endian systems. 431 432New features for MIPS-based systems: 433 434 r4650. 435 mips4 and R8000. 436 Irix 6.0. 437 64-bit ABI. 438 Allow dollar signs in labels on SGI/Irix 5.x. 439 440New support for HP PA RISC: 441 442 Generation of PIC (requires binutils-2.5.2.u6 or later). 443 HP-UX version 9 on HP PA RISC (dynamically links even with -g). 444 Processor variants for HP PA RISC: 700, 7100, and 7100LC. 445 Automatic generation of long calls when needed. 446 -mfast-indirect-calls for kernels and static binaries. 447 448 The called routine now copies arguments passed by invisible reference, 449 as required by the calling standard. 450 451Other new miscellaneous target-specific support: 452 453 -mno-multm on a29k. 454 -mold-align for i960. 455 Configuration for "semi-hosted" ARM. 456 -momit-leaf-frame-pointer for M88k. 457 SH3 variant of Hitachi Super-H and support both big and little endian. 458 459Changes to Objective-C: 460 461 Bare-bones implementation of NXConstantString has been added, 462 which is invoked by the @"string" directive. 463 464 Class * has been changed to Class to conform to the NextSTEP and 465 OpenStep runtime. 466 467 Enhancements to make dynamic loading easier. 468 469 The module version number has been updated to Version 7, thus existing 470 code will need to be recompiled to use the current run-time library. 471 472GCC now supports the ISO Normative Addendum 1 to the C Standard. 473As a result: 474 475 The header <iso646.h> defines macros for C programs written 476 in national variants of ISO 646. 477 478 The following digraph tokens are supported: 479 <: :> <% %> %: %:%: 480 These behave like the following, respectively: 481 [ ] { } # ## 482 483 Digraph tokens are supported unless you specify the `-traditional' 484 option; you do not need to specify `-ansi' or `-trigraphs'. Except 485 for contrived and unlikely examples involving preprocessor 486 stringizing, digraph interpretation doesn't change the meaning of 487 programs; this is unlike trigraph interpretation, which changes the 488 meanings of relatively common strings. 489 490 The macro __STDC_VERSION__ has the value 199409L. 491 492 As usual, for full conformance to the standard, you also need a 493 C library that conforms. 494 495The following lists changes that have been made to g++. If some 496features mentioned below sound unfamiliar, you will probably want to 497look at the recently-released public review copy of the C++ Working 498Paper. For PostScript and PDF (Adobe Acrobat) versions, see the 499archive at ftp://research.att.com/dist/stdc++/WP. For HTML and ASCII 500versions, see ftp://ftp.cygnus.com/pub/g++. On the web, see 501http://www.cygnus.com/~mrs/wp-draft. 502 503The scope of variables declared in the for-init-statement has been changed 504to conform to http://www.cygnus.com/~mrs/wp-draft/stmt.html#stmt.for; as a 505result, packages such as groff 1.09 will not compile unless you specify the 506-fno-for-scope flag. PLEASE DO NOT REPORT THIS AS A BUG; this is a change 507mandated by the C++ standardization committee. 508 509Binary incompatibilities: 510 511 The builtin 'bool' type is now the size of a machine word on RISC targets, 512 for code efficiency; it remains one byte long on CISC targets. 513 514 Code that does not use #pragma interface/implementation will most 515 likely shrink dramatically, as g++ now only emits the vtable for a 516 class in the translation unit where its first non-inline, non-abstract 517 virtual function is defined. 518 519 Classes that do not define the copy constructor will sometimes be 520 passed and returned in registers. This may illuminate latent bugs in 521 your code. 522 523Support for automatic template instantiation has *NOT* been added, due 524to a disagreement over design philosophies. 525 526Support for exception handling has been improved; more targets are now 527supported, and throws will use the RTTI mechanism to match against the 528catch parameter type. Optimization is NOT SUPPORTED with 529-fhandle-exceptions; no need to report this as a bug. 530 531Support for Run-Time Type Identification has been added with -frtti. 532This support is still in alpha; one major restriction is that any file 533compiled with -frtti must include <typeinfo.h>. 534 535Preliminary support for namespaces has been added. This support is far 536from complete, and probably not useful. 537 538Synthesis of compiler-generated constructors, destructors and 539assignment operators is now deferred until the functions are used. 540 541The parsing of expressions such as `a ? b : c = 1' has changed from 542`(a ? b : c) = 1' to `a : b ? (c = 1)'. 543 544The code generated for testing conditions, especially those using || 545and &&, is now more efficient. 546 547The operator keywords and, and_eq, bitand, bitor, compl, not, not_eq, 548or, or_eq, xor and xor_eq are now supported. Use -ansi or 549-foperator-names to enable them. 550 551The 'explicit' keyword is now supported. 'explicit' is used to mark 552constructors and type conversion operators that should not be used 553implicitly. 554 555g++ now accepts the typename keyword, though it currently has no 556semantics; it can be a no-op in the current template implementation. 557You may want to start using it in your code, however, since the 558pending rewrite of the template implementation to compile STL properly 559(perhaps for 2.8.0, perhaps not) will require you to use it as 560indicated by the current draft. 561 562Handling of user-defined type conversion has been overhauled so that 563type conversion operators are now found and used properly in 564expressions and function calls. 565 566-fno-strict-prototype now only applies to function declarations with 567"C" linkage. 568 569g++ now warns about 'if (x=0)' with -Wparentheses or -Wall. 570 571#pragma weak and #pragma pack are supported on System V R4 targets, as 572are various other target-specific #pragmas supported by gcc. 573 574new and delete of const types is now allowed (with no additional 575semantics). 576 577Explicit instantiation of template methods is now supported. Also, 578'inline template class foo<int>;' can be used to emit only the vtable 579for a template class. 580 581With -fcheck-new, g++ will check the return value of all calls to 582operator new, and not attempt to modify a returned null pointer. 583 584The template instantiation code now handles more conversions when 585passing to a parameter that does not depend on template arguments. 586This means that code like 'string s; cout << s;' now works. 587 588Invalid jumps in a switch statement past declarations that require 589initializations are now caught. 590 591Functions declared 'extern inline' now have the same linkage semantics 592as inline member functions. On supported targets, where previously 593these functions (and vtables, and template instantiations) would have 594been defined statically, they will now be defined as weak symbols so 595that only one out-of-line definition is used. 596 597collect2 now demangles linker output, and c++filt has become part of 598the gcc distribution. 599 600Noteworthy changes in GCC version 2.6.3: 601 602A few more bugs have been fixed. 603 604Noteworthy changes in GCC version 2.6.2: 605 606A few bugs have been fixed. 607 608Names of attributes can now be preceded and followed by double underscores. 609 610Noteworthy changes in GCC version 2.6.1: 611 612Numerous (mostly minor) bugs have been fixed. 613 614The following new configurations are supported: 615 616 GNU on x86 (instead of treating it like MACH) 617 NetBSD on Sparc and Motorola 68k 618 AIX 4.1 on RS/6000 and PowerPC systems 619 Sequent DYNIX/ptx 1.x and 2.x. 620 Both COFF and ELF configurations on AViiON without using /bin/gcc 621 Windows/NT on x86 architecture; preliminary 622 AT&T DSP1610 digital signal processor chips 623 i960 systems on bare boards using COFF 624 PDP11; target only and not extensively tested 625 626The -pg option is now supported for Alpha under OSF/1 V3.0 or later. 627 628Files with an extension of ".c++" are treated as C++ code. 629 630The -Xlinker and -Wl arguments are now passed to the linker in the 631position they were specified on the command line. This makes it 632possible, for example, to pass flags to the linker about specific 633object files. 634 635The use of positional arguments to the configure script is no longer 636recommended. Use --target= to specify the target; see the GCC manual. 637 638The 386 now supports two new switches: -mreg-alloc=<string> changes 639the default register allocation order used by the compiler, and 640-mno-wide-multiply disables the use of the mul/imul instructions that 641produce 64 bit results in EAX:EDX from 32 bit operands to do long long 642multiplies and 32-bit division by constants. 643 644Noteworthy changes in GCC version 2.6.0: 645 646Numerous bugs have been fixed, in the C and C++ front-ends, as 647well as in the common compiler code. 648 649This release includes the C, Objective-C, and C++ compilers. However, 650we have moved the files for the C++ compiler (G++) files to a 651subdirectory, cp. Subsequent releases of GCC will split these files 652to a separate TAR file. 653 654The G++ team has been tracking the development of the ANSI standard for C++. 655Here are some new features added from the latest working paper: 656 657 * built-in boolean type 'bool', with constants 'true' and 'false'. 658 * array new and delete (operator new [] and delete []). 659 * WP-conforming lifetime of temporaries. 660 * explicit instantiation of templates (template class A<int>;), 661 along with an option (-fno-implicit-templates) to disable emission 662 of implicitly instantiated templates, obsoletes -fexternal-templates. 663 * static member constants (static const int foo = 4; within the 664 class declaration). 665 666Many error messages have been improved to tell the user more about the 667problem. Conformance checking with -pedantic-errors has been 668improved. G++ now compiles Fresco. 669 670There is now an experimental implementation of virtual functions using 671thunks instead of Cfront-style vtables, enabled with -fvtable-thunks. 672This option also enables a heuristic which causes the compiler to only 673emit the vtable in the translation unit where its first non-inline 674virtual function is defined; using this option and 675-fno-implicit-templates, users should be able to avoid #pragma 676interface/implementation altogether. 677 678Signatures have been added as a GNU C++ extension. Using the option 679-fhandle-signatures, users are able to turn on recognition of 680signatures. A short introduction on signatures is in the section 681`Extension to the C++ Language' in the manual. 682 683The `g++' program is now a C program, rather than a shell script. 684 685Lots and lots and lots of bugs fixes, in nested types, access control, 686pointers to member functions, the parser, templates, overload 687resolution, etc, etc. 688 689There have been two major enhancements to the Objective-C compiler: 690 6911) Added portability. It now runs on Alpha, and some problems with 692 message forwarding have been addressed on other platforms. 693 6942) Selectors have been redefined to be pointers to structs like: 695 { void *sel_id, char *sel_types }, where the sel_id is the unique 696 identifier, the selector itself is no longer unique. 697 698 Programmers should use the new function sel_eq to test selector 699 equivalence. 700 701The following major changes have been made to the base compiler and 702machine-specific files. 703 704- The MIL-STD-1750A is a new port, but still preliminary. 705 706- The h8/300h is now supported; both the h8/300 and h8/300h ports come 707 with 32 bit IEEE 754 software floating point support. 708 709- The 64-bit Sparc (v9) and 64-bit MIPS chips are supported. 710 711- NetBSD is supported on m68k, Intel x86, and pc523 systems and FreeBSD 712 on x86. 713 714- COFF is supported on x86, m68k, and Sparc systems running LynxOS. 715 716- 68K systems from Bull and Concurrent are supported and System V 717 Release 4 is supported on the Atari. 718 719- GCC supports GAS on the Motorola 3300 (sysV68) and debugging 720 (assuming GAS) on the Plexus 68K system. (However, GAS does not yet 721 work on those systems). 722 723- System V Release 4 is supported on MIPS (Tandem). 724 725- For DG/UX, an ELF configuration is now supported, and both the ELF 726 and BCS configurations support ELF and COFF object file formats. 727 728- OSF/1 V2.0 is supported on Alpha. 729 730- Function profiling is also supported on Alpha. 731 732- GAS and GDB is supported for Irix 5 (MIPS). 733 734- "common mode" (code that will run on both POWER and PowerPC 735 architectures) is now supported for the RS/6000 family; the 736 compiler knows about more PPC chips. 737 738- Both NeXTStep 2.1 and 3 are supported on 68k-based architectures. 739 740- On the AMD 29k, the -msoft-float is now supported, as well as 741 -mno-sum-in-toc for RS/6000, -mapp-regs and -mflat for Sparc, and 742 -membedded-pic for MIPS. 743 744- GCC can now convert division by integer constants into the equivalent 745 multiplication and shift operations when that is faster than the 746 division. 747 748- Two new warning options, -Wbad-function-cast and 749 -Wmissing-declarations have been added. 750 751- Configurations may now add machine-specific __attribute__ options on 752 type; many machines support the `section' attribute. 753 754- The -ffast-math flag permits some optimization that violate strict 755 IEEE rules, such as converting X * 0.0 to 0.0. 756 757Noteworthy changes in GCC version 2.5.8: 758 759This release only fixes a few serious bugs. These include fixes for a 760bug that prevented most programs from working on the RS/6000, a bug 761that caused invalid assembler code for programs with a `switch' 762statement on the NS32K, a G++ problem that caused undefined names in 763some configurations, and several less serious problems, some of which 764can affect most configuration. 765 766Noteworthy change in GCC version 2.5.7: 767 768This release only fixes a few bugs, one of which was causing bootstrap 769compare errors on some systems. 770 771Noteworthy change in GCC version 2.5.6: 772 773A few backend bugs have been fixed, some of which only occur on one 774machine. 775 776The C++ compiler in 2.5.6 includes: 777 778 * fixes for some common crashes 779 * correct handling of nested types that are referenced as `foo::bar' 780 * spurious warnings about friends being declared static and never 781 defined should no longer appear 782 * enums that are local to a method in a class, or a class that's 783 local to a function, are now handled correctly. For example: 784 class foo { void bar () { enum { x, y } E; x; } }; 785 void bar () { class foo { enum { x, y } E; E baz; }; } 786 787Noteworthy change in GCC version 2.5.5: 788 789A large number of C++ bugs have been fixed. 790 791The fixproto script adds prototypes conditionally on __cplusplus. 792 793Noteworthy change in GCC version 2.5.4: 794 795A bug fix in passing of structure arguments for the HP-PA architecture 796makes code compiled with GCC 2.5.4 incompatible with code compiled 797with earlier versions (if it passes struct arguments of 33 to 64 bits, 798interspersed with other types of arguments). 799 800Noteworthy change in gcc version 2.5.3: 801 802The method of "mangling" C++ function names has been changed. So you 803must recompile all C++ programs completely when you start using GCC 8042.5. Also, GCC 2.5 requires libg++ version 2.5. Earlier libg++ 805versions won't work with GCC 2.5. (This is generally true--GCC 806version M.N requires libg++ version M.N.) 807 808Noteworthy GCC changes in version 2.5: 809 810* There is now support for the IBM 370 architecture as a target. 811Currently the only operating system supported is MVS; GCC does not run 812on MVS, so you must produce .s files using GCC as a cross compiler, 813then transfer them to MVS to assemble them. This port is not reliable 814yet. 815 816* The Power PC is now supported. 817 818* The i860-based Paragon machine is now supported. 819 820* The Hitachi 3050 (an HP-PA machine) is now supported. 821 822* The variable __GNUC_MINOR__ holds the minor version number of GCC, as 823an integer. For version 2.5.X, the value is 5. 824 825* In C, initializers for static and global variables are now processed 826an element at a time, so that they don't need a lot of storage. 827 828* The C syntax for specifying which structure field comes next in an 829initializer is now `.FIELDNAME='. The corresponding syntax for 830array initializers is now `[INDEX]='. For example, 831 832 char whitespace[256] 833 = { [' '] = 1, ['\t'] = 1, ['\n'] = 1 }; 834 835This was changed to accord with the syntax proposed by the Numerical 836C Extensions Group (NCEG). 837 838* Complex numbers are now supported in C. Use the keyword __complex__ 839to declare complex data types. See the manual for details. 840 841* GCC now supports `long double' meaningfully on the Sparc (128-bit 842floating point) and on the 386 (96-bit floating point). The Sparc 843support is enabled on Solaris 2.x because earlier system versions 844(SunOS 4) have bugs in the emulation. 845 846* All targets now have assertions for cpu, machine and system. So you 847can now use assertions to distinguish among all supported targets. 848 849* Nested functions in C may now be inline. Just declare them inline 850in the usual way. 851 852* Packed structure members are now supported fully; it should be possible 853to access them on any supported target, no matter how little alignment 854they have. 855 856* To declare that a function does not return, you must now write 857something like this (works only in 2.5): 858 859 void fatal () __attribute__ ((noreturn)); 860 861or like this (works in older versions too): 862 863 typedef void voidfn (); 864 865 volatile voidfn fatal; 866 867It used to be possible to do so by writing this: 868 869 volatile void fatal (); 870 871but it turns out that ANSI C requires that to mean something 872else (which is useless). 873 874Likewise, to declare that a function is side-effect-free 875so that calls may be deleted or combined, write 876something like this (works only in 2.5): 877 878 int computation () __attribute__ ((const)); 879 880or like this (works in older versions too): 881 882 typedef int intfn (); 883 884 const intfn computation; 885 886* The new option -iwithprefixbefore specifies a directory to add to 887the search path for include files in the same position where -I would 888put it, but uses the specified prefix just like -iwithprefix. 889 890* Basic block profiling has been enhanced to record the function the 891basic block comes from, and if the module was compiled for debugging, 892the line number and filename. A default version of the basic block 893support module has been added to libgcc2 that appends the basic block 894information to a text file 'bb.out'. Machine descriptions can now 895override the basic block support module in the target macro file. 896 897New features in g++: 898 899* The new flag `-fansi-overloading' for C++. Use a newly implemented 900scheme of argument matching for C++. It makes g++ more accurately 901obey the rules set down in Chapter 13 of the Annotated C++ Reference 902Manual (the ARM). This option will be turned on by default in a 903future release. 904 905* The -finline-debug flag is now gone (it was never really used by the 906 compiler). 907 908* Recognizing the syntax for pointers to members, e.g., "foo::*bar", has been 909 dramatically improved. You should not get any syntax errors or incorrect 910 runtime results while using pointers to members correctly; if you do, it's 911 a definite bug. 912 913* Forward declaration of an enum is now flagged as an error. 914 915* Class-local typedefs are now working properly. 916 917* Nested class support has been significantly improved. The compiler 918 will now (in theory) support up to 240 nested classes before hitting 919 other system limits (like memory size). 920 921* There is a new C version of the `g++' driver, to replace the old 922 shell script. This should significantly improve the performance of 923 executing g++ on a system where a user's PATH environment variable 924 references many NFS-mounted filesystems. This driver also works 925 under MS-DOS and OS/2. 926 927* The ANSI committee working on the C++ standard has adopted a new 928 keyword `mutable'. This will allow you to make a specific member be 929 modifiable in an otherwise const class. 930 931Noteworthy GCC changes in version 2.4.4: 932 933 A crash building g++ on various hosts (including m68k) has been 934 fixed. Also the g++ compiler no longer reports incorrect 935 ambiguities in some situations where they do not exist, and 936 const template member functions are now being found properly. 937 938Noteworthy GCC changes in version 2.4: 939 940* On each target, the default is now to return short structures 941compatibly with the "usual" compiler on that target. 942 943For most targets, this means the default is to return all structures 944in memory, like long structures, in whatever way is used on that 945target. Use -freg-struct-return to enable returning short structures 946(and unions) in registers. 947 948This change means that newly compiled binaries are incompatible with 949binaries compiled with previous versions of GCC. 950 951On some targets, GCC is itself the usual compiler. On these targets, 952the default way to return short structures is still in registers. 953Use -fpcc-struct-return to tell GCC to return them in memory. 954 955* There is now a floating point emulator which can imitate the way all 956supported target machines do floating point arithmetic. 957 958This makes it possible to have cross compilation to and from the VAX, 959and between machines of different endianness. However, this works 960only when the target machine description is updated to use the new 961facilities, and not all have been updated. 962 963This also makes possible support for longer floating point types. 964GCC 2.4 supports extended format on the 68K if you use `long double', 965for targets that have a 68881. (When we have run time library 966routines for extended floating point, then `long double' will use 967extended format on all 68K targets.) 968 969We expect to support extended floating point on the i386 and Sparc in 970future versions. 971 972* Building GCC now automatically fixes the system's header files. 973This should require no attention. 974 975* GCC now installs an unsigned data type as size_t when it fixes the 976header files (on all but a handful of old target machines). 977Therefore, the bug that size_t failed to be unsigned is fixed. 978 979* Building and installation are now completely separate. 980All new files are constructed during the build process; 981installation just copies them. 982 983* New targets supported: Clipper, Hitachi SH, Hitachi 8300, and Sparc 984Lite. 985 986* A totally new and much better Objective C run time system is included. 987 988* Objective C supports many new features. Alas, I can't describe them 989since I don't use that language; however, they are the same ones 990supported in recent versions of the NeXT operating system. 991 992* The builtin functions __builtin_apply_args, __builtin_apply and 993__builtin_return let you record the arguments and returned 994value of a function without knowing their number or type. 995 996* The builtin string variables __FUNCTION__ and __PRETTY_FUNCTION__ 997give the name of the function in the source, and a pretty-printed 998version of the name. The two are the same in C, but differ in C++. 999 1000* Casts to union types do not yield lvalues. 1001 1002* ## before an empty rest argument discards the preceding sequence 1003of non-whitespace characters from the macro definition. 1004(This feature is subject to change.) 1005 1006 1007New features specific to C++: 1008 1009* The manual contains a new section ``Common Misunderstandings with 1010GNU C++'' that C++ users should read. 1011 1012* #pragma interface and #pragma implementation let you use the same 1013C++ source file for both interface and implementation. 1014However, this mechanism is still in transition. 1015 1016* Named returned values let you avoid an extra constructor call 1017when a function result has a class type. 1018 1019* The C++ operators <? and >? yield min and max, respectively. 1020 1021* C++ gotos can exit a block safely even if the block has 1022aggregates that require destructors. 1023 1024* gcc defines the macro __GNUG__ when compiling C++ programs. 1025 1026* GNU C++ now correctly distinguishes between the prefix and postfix 1027forms of overloaded operator ++ and --. To avoid breaking old 1028code, if a class defines only the prefix form, the compiler 1029accepts either ++obj or obj++, unless -pedantic is used. 1030 1031* If you are using version 2.3 of libg++, you need to rebuild it with 1032`make CC=gcc' to avoid mismatches in the definition of `size_t'. 1033 1034Newly documented compiler options: 1035 1036-fnostartfiles 1037 Omit the standard system startup files when linking. 1038 1039-fvolatile-global 1040 Consider memory references to extern and global data items to 1041 be volatile. 1042 1043-idirafter DIR 1044 Add DIR to the second include path. 1045 1046-iprefix PREFIX 1047 Specify PREFIX for later -iwithprefix options. 1048 1049-iwithprefix DIR 1050 Add PREFIX/DIR to the second include path. 1051 1052-mv8 1053 Emit Sparc v8 code (with integer multiply and divide). 1054-msparclite 1055 Emit Sparclite code (roughly v7.5). 1056 1057-print-libgcc-file-name 1058 Search for the libgcc.a file, print its absolute file name, and exit. 1059 1060-Woverloaded-virtual 1061 Warn when a derived class function declaration may be an error 1062 in defining a C++ virtual function. 1063 1064-Wtemplate-debugging 1065 When using templates in a C++ program, warn if debugging is 1066 not yet fully available. 1067 1068+eN 1069 Control how C++ virtual function definitions are used 1070 (like cfront 1.x). 1071 1072