[BACK]Return to README CVS log [TXT][DIR] Up to [local] / OpenXM_contrib / gc

Diff for /OpenXM_contrib/gc/Attic/README between version 1.1 and 1.1.1.3

version 1.1, 1999/11/27 10:58:32 version 1.1.1.3, 2000/12/01 14:48:27
Line 1 
Line 1 
 Copyright 1988, 1989 Hans-J. Boehm, Alan J. Demers  Copyright 1988, 1989 Hans-J. Boehm, Alan J. Demers
 Copyright (c) 1991-1996 by Xerox Corporation.  All rights reserved.  Copyright (c) 1991-1996 by Xerox Corporation.  All rights reserved.
 Copyright (c) 1996-1998 by Silicon Graphics.  All rights reserved.  Copyright (c) 1996-1999 by Silicon Graphics.  All rights reserved.
   Copyright (c) 1999 by Hewlett-Packard Company. All rights reserved.
   
 THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED  THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
 OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.  OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.
Line 11  Permission to modify the code and to distribute modifi
Line 12  Permission to modify the code and to distribute modifi
 provided the above notices are retained, and a notice that the code was  provided the above notices are retained, and a notice that the code was
 modified is included with the above copyright notice.  modified is included with the above copyright notice.
   
 This is version 4.14 of a conservative garbage collector for C and C++.  This is version 5.3 of a conservative garbage collector for C and C++.
   
 You might find a more recent version of this at  You might find a more recent version of this at
   
 http://reality.sgi.com/boehm/gc.html  http://www.hpl.hp.com/personal/Hans_Boehm/gc
   
 HISTORY -  HISTORY -
   
   Early versions of this collector were developed as a part of research    Early versions of this collector were developed as a part of research
 projects supported in part by the National Science Foundation  projects supported in part by the National Science Foundation
 and the Defense Advance Research Projects Agency.  and the Defense Advance Research Projects Agency.
 Much of the code was rewritten by Hans-J. Boehm at Xerox PARC  Much of the code was rewritten by Hans-J. Boehm (boehm@acm.org) at Xerox PARC,
 and is now maintained by him at SGI (boehm@sgi.com).  SGI, and HP Labs.
   
 Some other contributors:  Some other contributors:
   
Line 40  Robert Brazile (brazile@diamond.bbn.com) originally su
Line 41  Robert Brazile (brazile@diamond.bbn.com) originally su
 Al Dosser (dosser@src.dec.com) and Regis Cridlig (Regis.Cridlig@cl.cam.ac.uk)  Al Dosser (dosser@src.dec.com) and Regis Cridlig (Regis.Cridlig@cl.cam.ac.uk)
 subsequently provided updates and information on variation between ULTRIX  subsequently provided updates and information on variation between ULTRIX
 systems.  Parag Patel (parag@netcom.com) supplied the A/UX code.  systems.  Parag Patel (parag@netcom.com) supplied the A/UX code.
 Jesper Peterson(jep@mtiame.mtia.oz.au) and  Jesper Peterson(jep@mtiame.mtia.oz.au), Michel Schinz, and
 Michel Schinz supplied the Amiga port.  Martin Tauchmann (martintauchmann@bigfoot.com) supplied the Amiga port.
 Thomas Funke (thf@zelator.in-berlin.de(?)) and  Thomas Funke (thf@zelator.in-berlin.de(?)) and
 Brian D.Carlstrom (bdc@clark.lcs.mit.edu) supplied the NeXT ports.  Brian D.Carlstrom (bdc@clark.lcs.mit.edu) supplied the NeXT ports.
 Douglas Steel (doug@wg.icl.co.uk) provided ICL DRS6000 code.  Douglas Steel (doug@wg.icl.co.uk) provided ICL DRS6000 code.
Line 475  intended to run with malloc/free (e.g. code with extre
Line 476  intended to run with malloc/free (e.g. code with extre
 portability constraints).  To do so define FIND_LEAK in Makefile  portability constraints).  To do so define FIND_LEAK in Makefile
 This will cause the collector to invoke the report_leak  This will cause the collector to invoke the report_leak
 routine defined near the top of reclaim.c whenever an inaccessible  routine defined near the top of reclaim.c whenever an inaccessible
 object is found that has not been explicitly freed.  The collector will  object is found that has not been explicitly freed.  Such objects will
 no longer reclaim inaccessible memory; in this form it is purely a  also be automatically reclaimed.
 debugging tool.  
   Productive use of this facility normally involves redefining report_leak    Productive use of this facility normally involves redefining report_leak
 to do something more intelligent.  This typically requires annotating  to do something more intelligent.  This typically requires annotating
 objects with additional information (e.g. creation time stack trace) that  objects with additional information (e.g. creation time stack trace) that
Line 612  reclaimed.  Exclusive-or'ing forward and backward link
Line 612  reclaimed.  Exclusive-or'ing forward and backward link
 doesn't cut it.  doesn't cut it.
   Some C optimizers may lose the last undisguised pointer to a memory    Some C optimizers may lose the last undisguised pointer to a memory
 object as a consequence of clever optimizations.  This has almost  object as a consequence of clever optimizations.  This has almost
 never been observed in practice.  Send mail to boehm@sgi.com  never been observed in practice.  Send mail to boehm@acm.org
 for suggestions on how to fix your compiler.  for suggestions on how to fix your compiler.
   This is not a real-time collector.  In the standard configuration,    This is not a real-time collector.  In the standard configuration,
 percentage of time required for collection should be constant across  percentage of time required for collection should be constant across
Line 621  heap sizes.  But collection pauses will increase for l
Line 621  heap sizes.  But collection pauses will increase for l
 per MB of accessible memory that needs to be scanned.  Your mileage  per MB of accessible memory that needs to be scanned.  Your mileage
 may vary.)  The incremental/generational collection facility helps,  may vary.)  The incremental/generational collection facility helps,
 but is portable only if "stubborn" allocation is used.  but is portable only if "stubborn" allocation is used.
   Please address bug reports to boehm@sgi.com.  If you are    Please address bug reports to boehm@acm.org.  If you are
 contemplating a major addition, you might also send mail to ask whether  contemplating a major addition, you might also send mail to ask whether
 it's already been done (or whether we tried and discarded it).  it's already been done (or whether we tried and discarded it).
   
Line 1451  Since 4.14alpha1
Line 1451  Since 4.14alpha1
   
 Since 4.14alpha2  Since 4.14alpha2
  - changed STACKBOTTOM for DJGPP (Thanks to Salvador Eduardo Tropea).   - changed STACKBOTTOM for DJGPP (Thanks to Salvador Eduardo Tropea).
   
   Since 4.14
    - Reworked large block allocator.  Now uses multiple doubly linked free
      lists to approximate best fit.
    - Changed heap expansion heuristic.  Entirely free blocks are no longer
      counted towards the heap size.  This seems to have a major impact on
      heap size stability; the old version could expand the heap way too
      much in the presence of large block fragmentation.
    - added -DGC_ASSERTIONS and some simple assertions inside the collector.
      This is mainlyt for collector debugging.
    - added -DUSE_MUNMAP to allow the heap to shrink.  Suupported on only
      a few UNIX-like platforms for now.
    - added GC_dump_regions() for debugging of fragmentation issues.
    - Changed PowerPC pointer alignment under Linux to 4.  (This needs
      checking by someone who has one.  The suggestions came to me via a
      rather circuitous path.)
    - Changed the Linux/Alpha port to walk the data segment backwards until
      it encounters a SIGSEGV.  The old way to find the start of the data
      segment broke with a recent release.
    - cordxtra.c needed to call GC_REGISTER_FINALIZER instead of
      GC_register_finalizer, so that it would continue to work with GC_DEBUG.
    - allochblk sometimes cleared the wrong block for debugging purposes
      when it dropped blacklisted blocks.  This could result in spurious
      error reports with GC_DEBUG.
    - added MACOS X Server support.  (Thanks to Andrew Stone.)
    - Changed the Solaris threads code to ignore stack limits > 8 MB with
      a warning.  Empirically, it is not safe to access arbitrary pages
      in such large stacks.  And the dirty bit implementation does not
      guarantee that none of them will be accessed.
    - Integrated Martin Tauchmann's Amiga changes.
    - Integrated James Dominy's OpenBSD/SPARC port.
   
   Since 5.0alpha1
    - Fixed bugs introduced in alpha1 (OpenBSD & large block initialization).
    - Added -DKEEP_BACK_PTRS and backptr.h interface.  (The implementation
      idea came from Al Demers.)
   
   Since 5.0alpha2
    - Added some highly incomplete code to support a copied young generation.
      Comments on nursery.h are appreciated.
    - Changed -DFIND_LEAK, -DJAVA_FINALIZATION, and -DFINALIZE_ON_DEMAND,
      so the same effect could be obtained with a runtime switch.   This is
      a step towards standardizing on a single dynamic GC library.
    - Significantly changed the way leak detection is handled, as a consequence
      of the above.
   
   Since 5.0 alpha3
    - Added protection fault handling patch for Linux/M68K from Fergus
      Henderson and Roman Hodek.
    - Removed the tests for SGI_SOURCE in new_gc_alloc.h.  This was causing that
      interface to fail on nonSGI platforms.
    - Changed the Linux stack finding code to use /proc, after changing it
      to use HEURISTIC1.  (Thanks to David Mossberger for pointing out the
      /proc hook.)
    - Added HP/UX incremental GC support and HP/UX 11 thread support.
      Thread support is currently still flakey.
    - Added basic Linux/IA64 support.
    - Integrated Anthony Green's PicoJava support.
    - Integrated Scott Ananian's StrongARM/NetBSD support.
    - Fixed some fairly serious performance bugs in the incremental
      collector.  These have probably been there essentially forever.
      (Mark bits were sometimes set before scanning dirty pages.
      The reclaim phase unnecessarily dirtied full small object pages.)
    - Changed the reclaim phase to ignore nearly full pages to avoid
      touching them.
    - Limited GC_black_list_spacing to roughly the heap growth increment.
    - Changed full collection triggering heuristic to decrease full GC
      frequency by default, but to explicitly trigger full GCs during
      heap growth.  This doesn't always improve things, but on average it's
      probably a win.
    - GC_debug_free(0, ...) failed.  Thanks to Fergus Henderson for the
      bug report and fix.
   
   Since 5.0 alpha4
    - GC_malloc_explicitly_typed and friends sometimes failed to
      initialize first word.
    - Added allocation routines and support in the marker for mark descriptors
      in a type structure referenced by the first word of an object.  This was
      introduced to support gcj, but hopefully in a way that makes it
      generically useful.
    - Added GC_requested_heapsize, and inhibited collections in nonincremental
      mode if the actual used heap size is less than what was explicitly
      requested.
    - The Solaris pthreads version of GC_pthread_create didn't handle a NULL
      attribute pointer.  Solaris thread support used the wrong default thread
      stack size.  (Thanks to Melissa O'Neill for the patch.)
    - Changed PUSH_CONTENTS macro to no longer modify first parameter.
      This usually doesn't matter, but it was certainly an accident waiting
      to happen ...
    - Added GC_register_finalizer_no_order and friends to gc.h.  They're
      needed by Java implementations.
    - Integrated a fix for a win32 deadlock resulting from clock() calling
      malloc.  (Thanks to Chris Dodd.)
    - Integrated Hiroshi Kawashima's port to Linux/MIPS.  This was designed
      for a handheld platform, and may or may not be sufficient for other
      machines.
    - Fixed a va_arg problem with the %c specifier in cordprnt.c.  It appears
      that this was always broken, but recent versions of gcc are the first to
      report the (statically detectable) bug.
    - Added an attempt at a more general solution to dlopen races/deadlocks.
      GC_dlopen now temporarily disables collection.  Still not ideal, but ...
    - Added -DUSE_I686_PREFETCH, -DUSE_3DNOW_PREFETCH, and support for IA64
      prefetch instructions.  May improve performance measurably, but I'm not
      sure the code will run correctly on processors that don't support the
      instruction.  Won't build except with very recent gcc.
    - Added caching for header lookups in the marker.  This seems to result
      in a barely measurable performance gain.  Added support for interleaved
      lookups of two pointers, but unconfigured that since the performance
      gain is currently near zero, and it adds to code size.
    - Changed Linux DATA_START definition to check both data_start and
      __data_start, since nothing else seems to be portable.
    - Added -DUSE_LD_WRAP to optionally take advantage of the GNU ld function
      wrapping mechanism.  Probably currently useful only on Linux.
    - Moved some variables for the scratch allocator into GC_arrays, on
      Martin Hirzel's suggestion.
    - Fixed a win32 threads bug that caused the collector to not look for
      interior pointers from one of the thread stacks without
      ALL_INTERIOR_POINTERS.  (Thanks to Jeff Sturm.)
    - Added Mingw32 support.  (Thanks again to Jeff Sturm for the patch.)
    - Changed the alpha port to use the generic register scanning code instead
      of alpha_mach_dep.s.  Alpha_mach_dep.s doesn't look for pointers in fp
      registers, but gcc sometimes spills pointers there.  (Thanks to Manuel
      Serrano for helping me debug this by email.)  Changed the IA64 code to
      do something similar for similar reasons.
   
   Since 5.0alpha6:
    - -DREDIRECT_MALLOC was broken in alpha6. Fixed.
    - Cleaned up gc_ccp.h slightly, thus also causing the HP C++ compiler to
      accept it.
    - Removed accidental reference to dbg_mlc.c, which caused dbg_mlc.o to be
      linked into every executable.
    - Added PREFETCH to bitmap marker.  Changed it to use the header cache.
    - GC_push_marked sometimes pushed one object too many, resulting in a
      segmentation fault in GC_mark_from_mark_stack.  This was probably an old
      bug.  It finally showed up in gctest on win32.
    - Gc_priv.h erroneously #defined GC_incremental to be TRUE instead of FALSE
      when SMALL_CONFIG was defined.  This was no doubt a major performance bug for
      the default win32 configuration.
    - Removed -DSMALL_CONFIG from NT_MAKEFILE.  It seemed like an anchronism now
      that the average PC has 64MB or so.
    - Integrated Bryce McKinley's patches for linux threads and dynamic loading
      from the libgcj tree.  Turned on dynamic loading support for Linux/PPC.
    - Changed the stack finding code to use environ on HP/UX.  (Thanks
      to Gustavo Rodriguez-Rivera for the suggestion.)  This should probably
      be done on other platforms, too.  Since I can't test those, that'll
      wait until after 5.0.
   
   Since 5.0alpha7:
    - Fixed threadlibs.c for linux threads.  -DUSE_LD_WRAP was broken and
      -ldl was omitted.  Fixed Linux stack finding code to handle
      -DUSE_LD_WRAP correctly.
    - Added MSWIN32 exception handler around marker, so that the collector
      can recover from root segments that are unmapped during the collection.
      This caused occasional failures under Windows 98, and may also be
      an issue under Windows NT/2000.
   
   Since 5.0
    - Fixed a gc.h header bug which showed up under Irix.  (Thanks to
      Dan Sullivan.)
    - Fixed a typo in GC_double_descr in typd_mlc.c not getting traced correctly.
      This probably could result in objects described by array descriptors not
      getting traced correctly.  (Thanks to Ben Hutchings for pointing this out.)
    - The block nearly full tests in reclaim.c were not correct for 64 bit
      environments.  This could result in unnecessary heap growth under unlikely
      conditions.
    - Removed use of CLEAR_DOUBLE from generic reclaim code, since odd sizes
      could occur.
   
   Since 5.1
    - dyn_load.c declared GC_scratch_last_end_ptr as an extern even if it
      was defined as a macro.  This prevented the collector from building on
      Irix.
    - We quietly assumed that indirect mark descriptors were never 0.
      Our own typed allocation interface violated that.  This could result
      in segmentation faults in the marker with typed allocation.
    - Fixed a _DUSE_MUNMAP bug in the heap block allocation code.
      (Thanks to Ben Hutchings for the patch.)
    - Taught the collector about VC++ handling array operator new.
      (Thanks again to Ben Hutchings for the patch.)
    - The two copies of gc_hdrs.h had diverged.  Made one a link to the other
      again.
   
   Since 5.2
    - Fixed _end declaration for OSF1.
    - There were lots of spurious leak reports in leak detection mode, caused
      by the fact that some pages were not being swept, and hence unmarked
      objects weren't making it onto free lists.  (This bug dated back to 5.0.)
    - Fixed a typo in the liblinuxgc.so Makefile rule.
    - Added the GetExitCodeThread to Win32 GC_stop_world to (mostly) work
      around a Windows 95 GetOpenFileName problem.  (Thanks to Jacob Navia.)
   
 To do:  To do:
    - Integrate Linux/SPARC fixes.
  - Very large root set sizes (> 16 MB or so) could cause the collector   - Very large root set sizes (> 16 MB or so) could cause the collector
    to abort with an unexpected mark stack overflow.  (Thanks again to     to abort with an unexpected mark stack overflow.  (Thanks again to
    Peter Chubb.)  NOT YET FIXED.  Workaround is to increase the initial     Peter Chubb.)  NOT YET FIXED.  Workaround is to increase the initial
Line 1465  To do:
Line 1656  To do:
    be possible to conditionally intercept mmap and use GC_exclude_static_roots.     be possible to conditionally intercept mmap and use GC_exclude_static_roots.
    The real fix is to walk rld data structures, which looks possible.     The real fix is to walk rld data structures, which looks possible.
  - Integrate MIT and DEC pthreads ports.   - Integrate MIT and DEC pthreads ports.
  - Deal with very uneven black-listing distributions.  If all the black listed   - Incremental collector should handle large objects better.  Currently,
    blocks reside in the newly allocated heap section, the heuristic for     it looks like the whole object is treated as dirty if any part of it
    temporarily ignoring black-listing fails, and the heap grows too much.     is.
    (This was observed in only one case, and could be worked around, but ...)   - Cord/cordprnt.c doesn't build on a few platforms (notably PowerPC), since
  - I've started work on rewriting the large block allocator to use approximate     we make some unwarranted assumptions about how varargs are handled.  This
    best fit.  There are rare cases in which the current allocator results in     currently makes the cord-aware versions of printf unusable on some platforms.
    excessive large block fragmentation, even with the 4.13 fixes.  This should     Fixing this is unfortunately not trivial.
    also reduce large block allocation time, whcih has become occasionally  
    noticable in 4.13.  
   
Legend:
Removed from v.1.1  
changed lines
  Added in v.1.1.1.3
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>