On Thu, May 17, 2001 at 03:49:36PM +0400, Sergey Krivulja wrote:
> DVL> Это ошибка в g++, исправлена она в gcc-c++ >= 2.96-ipl9mdk
> DVL> Текущая версия в Сизифе: 2.96-ipl11mdk
> А что добавлось?

По сравнению с gcc-2.96-ipl8mdk, в cpp/gcc/g++/.. добавилось множество
патчей (в основном от jj):
- fix typo in frame_state patch
- fix __frame_state_for incompatibility between egcs 1.* and gcc-2.96-RH (#37933)
- fix a tradcpp ICE with -Dfoo= (#38551)
- fix a sibcall and eh related ICE (#38466)
- fix a template argument mangling bug (#31168)
- fix i386 if conversion optimization (#37054)
- only warn about multi-line comments if -Wcomment or -Wall is given
- fix a SUBREG_BYTE ICE with ia64 builtins
- properly honour mode in cse lookups (Bernd Schmidt, #35122)
- fix an ICE in dwarf-2 support (Jason Merrill)
- fix an dwarf-2 output bug (Jason Merrill, #32801)
- fix 2 bugs in ia64 exception handling
- include ia64intrin.h header on ia64
- fix yet another alpha unaligned access (Richard Kenner, #37698)
- fix gcc driver with -traditional -traditional-cpp (Neil Booth, Chris Demetriou, #35963)
- fix tree checking bug on bad arguments to inlined functions (#36986)
- support -G option on ia64 (#33354)
- fix -MD -o name_without_dot (#37697)
- fix ia64 ICE seen in xfig (nested CONST rtl)
- add a patch for m68k requested by Jes Sorensen (Jeff Law)
- sync with ia64-000717-branch new patches
- fix functions using all local regs needing a frame pointer (Jim Wilson)
- fix constant folding of pointer comparisons where one pointer is (void *)(-1UL >> 1) (#33690)
- handle STRICT_LOW_PART in ifcvt (#33474)
- throttle C++ inlining (Mark Mitchell, #29556)
- define -D__EXCEPTIONS in g++ unless -fno-exceptions is passed
- fix GCSE for function parameters whose addresses are taken (#34050)
- fix null pointer check optimization (#23891)
- fix dwarf-2 in presence of inline function redeclared locally as extern (Jason Merrill, #25253)
- fix a SUBREG_BYTE related ICE (#32739)
- update regrename.c to fix some -frename-registers bugs (but -frename-registers is not considered supported)
- fix Fortran lexer, so that it does not eat one whole line after #line directive (#32230)
- fix implicit operator= handling for classes with anonymous aggregates (#32162)
- fix libstdc++'s bastring.cc so that it does not warn with -Wshadow
- update documentation about -Wshadow (#32160)
- optimize C++ frontend by killing -fnew-abi support (it is not ABI compatible with gcc 3.0 anyway)
- store static variables from inline functions into linkonce sections (to save space e.g. in mozilla)
- fix reg-stack return value handling (#20952, #21002, #31756)
- fix glibc getdents64.c build on ia32 with --enable-kernel 2.4.1 (Bernd Schmidt, Jan Hubicka)
- fix aspell miscompilation (#30297)
- issue a clear error message about invalid ia32 floating point asm constraints (#27137)
- remove libc5 kludge from ia32 crtendS.o
- fix Alpha combiner patch
- fix bogus cpp warning (Neil Booth)
- macro expand I in Objective C @I (Zack Weinberg)
- fix last cpp patch (so that -MD with -o works)
- fix constant folding bug (#28864)
- fix ADDRESSOF recognition (#29686)
- avoid bogus -Wunused warnings in Fortran (Richard Henderson, #29559)
- fix debugging of static local variables in stabs (Jason Merrill, #29548)
- don't crash if fixup_var_refs_1 changed MEM into REG (Richard Kenner,
- add explicit barriers to flushrs and mov x = ar.bsp on ia64 (Jim Wilson,
- fix ia64 syscall_linkage security problems (David Mosberger)
- Add Neil Booth fixes on cpp :
  * cpplex.c (_cpp_lex_token): Don't warn about directives in
  macro arguments when looking for the '('.
  * cppmacro.c (funlike_invocation_p): Set parsing_args to
  2 when really parsing arguments; 1 when looking for '('.
  Always restore the lexer position.
- don't build java subpackage on ia64 - libgcj does not work there
- fix C++ lookup if aggregate contains field with the same name as itself (#21420)
- clarify error message about taking address of an unqualified non-static member (#24329)
- better trim C++ inlined function graph cycles (#28385)
- fix C++ static aggregate initialization (#20267)
- fix Objective C garbage collection bug (#18827)
- fix Objective C testsuite so that it does not use system libobjc
- put only basename into -M goal (Nathan Sidwell)
- fix Alpha reload bug (Richard Henderson, #26749)
- fix Alpha combiner bug (Richard Henderson, #28509)
- fix i386 reload bug with long long (Bernd Schmidt, #28209)
- integrate patch.057 from gnupro-1117 tree
- ia64 support
- fix ia64 B-step errata handling code
- fix a crash caused by lexing #line directives in some cases (#25851)
- workaround ICE in dwarf2out (#24741)
- avoid unaligned accesses on alpha and ia64 (Richard Henderson)
- handle -foperator-names in the new cpp (#26457)
- add nice diagnostic message about #if defined(xor) with -foperator-names (Neil Booth)
- fix paste avoidance after builtin (Neil Booth)
- fix g++.jason/2371.C test so that new cpp does not complain about it (Alexandre Oliva)
- fix ICE on nested function with variable size array argument (#24506)
- change gcc -v output to include minor release
- fix last cpp patch so that no whitespace is inserted at start of line if last macro expansion resulted in no tokens (Neil Booth)
- fix ICE during printing warning about overloading decisions (#23584)
- honor no implicit extern "C" on linux in cpp
- fix layout of __attribute((packed)) enums in bitfields (showing up in Linux DAC960 driver)
- fix ICE when both prototype and function definition declares variable size array argument (#24507)
- change new cpp so that it does not insert whitespace where no macro expansion happens in C mode (Neil Booth, #25200)
- don't ICE when inlining is replacing with constant a variable which is only modified by assigning to itself (#24641)
- mask constant argument to testl or andl if the operation should be done in 8bit mode (#24921)
- fix C++ anonymous union handling (#23682)
- fix extract_bit_field so that it works even with XFmode (DJ Delorie)
- prevent inlining C++ functions containing static variables (#24253)
- include headers sourced from -include or -imacros in -M output (#24317)
- mangle overloaded operators declared in "C" linkage (#24685)
- optimize out comparisons of two constants if at least one comes from inline function arguments (visible e.g. on undefined __bad_udelay symbols from some linux 2.4.0 kernel modules)
- put in updated preprocessor, it seems stable enough and has tons of bugs against both gcc 2.95.2 cccp and cpp used in 2.96-RH until now
- fix DECL_NEEDED_P (Mark Mitchell, #23401)
- fix i386 assembly output for calls to symbol+offset (#23429)
- fix C++ parser to cure GC crashes (#23489, #24060)
- fix miscompilation due to not copied MEM/f flag (Alexander Oliva, #23502)
- fix CC flags handling after sar on i386 (Richard Henderson, #24062)
- only check inmode after checking type of argument in fcmov_comparison_operator on i386 (#24115)
- fix typos in i386 cmpqi_minus_1 (#24178)
- fix thinko in alpha alpha_expand_block_move (#23816, #24092, #24167, #24315)
- fix fixup_var_refs (#24031)
- fix my last fix for cpplib dependency handling (Graham Stott)


Regards,
	Dmitry

+-------------------------------------------------------------------------+
Dmitry V. Levin     mailto://ldv@alt-linux.org
ALT Linux Team      http://www.altlinux.ru/
Fandra Project      http://www.fandra.org/
+-------------------------------------------------------------------------+
UNIX is user friendly. It's just very selective about who its friends are.