gcc/NEWS

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