deps/jemalloc/ChangeLog

   1 Following are change highlights associated with official releases.  Important
   2 bug fixes are all mentioned, but internal enhancements are omitted here for
   3 brevity (even though they are more fun to write about).  Much more detail can be
   4 found in the git revision history:
   5
   6     http://www.canonware.com/cgi-bin/gitweb.cgi?p=jemalloc.git
   7     git://canonware.com/jemalloc.git
   8
   9 * 2.2.1 (March 30, 2011)
  10
  11   Bug fixes:
  12   - Implement atomic operations for x86/x64.  This fixes compilation failures
  13     for versions of gcc that are still in wide use.
  14   - Fix an assertion in arena_purge().
  15
  16 * 2.2.0 (March 22, 2011)
  17
  18   This version incorporates several improvements to algorithms and data
  19   structures that tend to reduce fragmentation and increase speed.
  20
  21   New features:
  22   - Add the "stats.cactive" mallctl.
  23   - Update pprof (from google-perftools 1.7).
  24   - Improve backtracing-related configuration logic, and add the
  25     --disable-prof-libgcc option.
  26
  27   Bug fixes:
  28   - Change default symbol visibility from "internal", to "hidden", which
  29     decreases the overhead of library-internal function calls.
  30   - Fix symbol visibility so that it is also set on OS X.
  31   - Fix a build dependency regression caused by the introduction of the .pic.o
  32     suffix for PIC object files.
  33   - Add missing checks for mutex initialization failures.
  34   - Don't use libgcc-based backtracing except on x64, where it is known to work.
  35   - Fix deadlocks on OS X that were due to memory allocation in
  36     pthread_mutex_lock().
  37   - Heap profiling-specific fixes:
  38     + Fix memory corruption due to integer overflow in small region index
  39       computation, when using a small enough sample interval that profiling
  40       context pointers are stored in small run headers.
  41     + Fix a bootstrap ordering bug that only occurred with TLS disabled.
  42     + Fix a rallocm() rsize bug.
  43     + Fix error detection bugs for aligned memory allocation.
  44
  45 * 2.1.3 (March 14, 2011)
  46
  47   Bug fixes:
  48   - Fix a cpp logic regression (due to the "thread.{de,}allocatedp" mallctl fix
  49     for OS X in 2.1.2).
  50   - Fix a "thread.arena" mallctl bug.
  51   - Fix a thread cache stats merging bug.
  52
  53 * 2.1.2 (March 2, 2011)
  54
  55   Bug fixes:
  56   - Fix "thread.{de,}allocatedp" mallctl for OS X.
  57   - Add missing jemalloc.a to build system.
  58
  59 * 2.1.1 (January 31, 2011)
  60
  61   Bug fixes:
  62   - Fix aligned huge reallocation (affected allocm()).
  63   - Fix the ALLOCM_LG_ALIGN macro definition.
  64   - Fix a heap dumping deadlock.
  65   - Fix a "thread.arena" mallctl bug.
  66
  67 * 2.1.0 (December 3, 2010)
  68
  69   This version incorporates some optimizations that can't quite be considered
  70   bug fixes.
  71
  72   New features:
  73   - Use Linux's mremap(2) for huge object reallocation when possible.
  74   - Avoid locking in mallctl*() when possible.
  75   - Add the "thread.[de]allocatedp" mallctl's.
  76   - Convert the manual page source from roff to DocBook, and generate both roff
  77     and HTML manuals.
  78
  79   Bug fixes:
  80   - Fix a crash due to incorrect bootstrap ordering.  This only impacted
  81     --enable-debug --enable-dss configurations.
  82   - Fix a minor statistics bug for mallctl("swap.avail", ...).
  83
  84 * 2.0.1 (October 29, 2010)
  85
  86   Bug fixes:
  87   - Fix a race condition in heap profiling that could cause undefined behavior
  88     if "opt.prof_accum" were disabled.
  89   - Add missing mutex unlocks for some OOM error paths in the heap profiling
  90     code.
  91   - Fix a compilation error for non-C99 builds.
  92
  93 * 2.0.0 (October 24, 2010)
  94
  95   This version focuses on the experimental *allocm() API, and on improved
  96   run-time configuration/introspection.  Nonetheless, numerous performance
  97   improvements are also included.
  98
  99   New features:
 100   - Implement the experimental {,r,s,d}allocm() API, which provides a superset
 101     of the functionality available via malloc(), calloc(), posix_memalign(),
 102     realloc(), malloc_usable_size(), and free().  These functions can be used to
 103     allocate/reallocate aligned zeroed memory, ask for optional extra memory
 104     during reallocation, prevent object movement during reallocation, etc.
 105   - Replace JEMALLOC_OPTIONS/JEMALLOC_PROF_PREFIX with MALLOC_CONF, which is
 106     more human-readable, and more flexible.  For example:
 107       JEMALLOC_OPTIONS=AJP
 108     is now:
 109       MALLOC_CONF=abort:true,fill:true,stats_print:true
 110   - Port to Apple OS X.  Sponsored by Mozilla.
 111   - Make it possible for the application to control thread-->arena mappings via
 112     the "thread.arena" mallctl.
 113   - Add compile-time support for all TLS-related functionality via pthreads TSD.
 114     This is mainly of interest for OS X, which does not support TLS, but has a
 115     TSD implementation with similar performance.
 116   - Override memalign() and valloc() if they are provided by the system.
 117   - Add the "arenas.purge" mallctl, which can be used to synchronously purge all
 118     dirty unused pages.
 119   - Make cumulative heap profiling data optional, so that it is possible to
 120     limit the amount of memory consumed by heap profiling data structures.
 121   - Add per thread allocation counters that can be accessed via the
 122     "thread.allocated" and "thread.deallocated" mallctls.
 123
 124   Incompatible changes:
 125   - Remove JEMALLOC_OPTIONS and malloc_options (see MALLOC_CONF above).
 126   - Increase default backtrace depth from 4 to 128 for heap profiling.
 127   - Disable interval-based profile dumps by default.
 128
 129   Bug fixes:
 130   - Remove bad assertions in fork handler functions.  These assertions could
 131     cause aborts for some combinations of configure settings.
 132   - Fix strerror_r() usage to deal with non-standard semantics in GNU libc.
 133   - Fix leak context reporting.  This bug tended to cause the number of contexts
 134     to be underreported (though the reported number of objects and bytes were
 135     correct).
 136   - Fix a realloc() bug for large in-place growing reallocation.  This bug could
 137     cause memory corruption, but it was hard to trigger.
 138   - Fix an allocation bug for small allocations that could be triggered if
 139     multiple threads raced to create a new run of backing pages.
 140   - Enhance the heap profiler to trigger samples based on usable size, rather
 141     than request size.
 142   - Fix a heap profiling bug due to sometimes losing track of requested object
 143     size for sampled objects.
 144
 145 * 1.0.3 (August 12, 2010)
 146
 147   Bug fixes:
 148   - Fix the libunwind-based implementation of stack backtracing (used for heap
 149     profiling).  This bug could cause zero-length backtraces to be reported.
 150   - Add a missing mutex unlock in library initialization code.  If multiple
 151     threads raced to initialize malloc, some of them could end up permanently
 152     blocked.
 153
 154 * 1.0.2 (May 11, 2010)
 155
 156   Bug fixes:
 157   - Fix junk filling of large objects, which could cause memory corruption.
 158   - Add MAP_NORESERVE support for chunk mapping, because otherwise virtual
 159     memory limits could cause swap file configuration to fail.  Contributed by
 160     Jordan DeLong.
 161
 162 * 1.0.1 (April 14, 2010)
 163
 164   Bug fixes:
 165   - Fix compilation when --enable-fill is specified.
 166   - Fix threads-related profiling bugs that affected accuracy and caused memory
 167     to be leaked during thread exit.
 168   - Fix dirty page purging race conditions that could cause crashes.
 169   - Fix crash in tcache flushing code during thread destruction.
 170
 171 * 1.0.0 (April 11, 2010)
 172
 173   This release focuses on speed and run-time introspection.  Numerous
 174   algorithmic improvements make this release substantially faster than its
 175   predecessors.
 176
 177   New features:
 178   - Implement autoconf-based configuration system.
 179   - Add mallctl*(), for the purposes of introspection and run-time
 180     configuration.
 181   - Make it possible for the application to manually flush a thread's cache, via
 182     the "tcache.flush" mallctl.
 183   - Base maximum dirty page count on proportion of active memory.
 184   - Compute various addtional run-time statistics, including per size class
 185     statistics for large objects.
 186   - Expose malloc_stats_print(), which can be called repeatedly by the
 187     application.
 188   - Simplify the malloc_message() signature to only take one string argument,
 189     and incorporate an opaque data pointer argument for use by the application
 190     in combination with malloc_stats_print().
 191   - Add support for allocation backed by one or more swap files, and allow the
 192     application to disable over-commit if swap files are in use.
 193   - Implement allocation profiling and leak checking.
 194
 195   Removed features:
 196   - Remove the dynamic arena rebalancing code, since thread-specific caching
 197     reduces its utility.
 198
 199   Bug fixes:
 200   - Modify chunk allocation to work when address space layout randomization
 201     (ASLR) is in use.
 202   - Fix thread cleanup bugs related to TLS destruction.
 203   - Handle 0-size allocation requests in posix_memalign().
 204   - Fix a chunk leak.  The leaked chunks were never touched, so this impacted
 205     virtual memory usage, but not physical memory usage.
 206
 207 * linux_2008082[78]a (August 27/28, 2008)
 208
 209   These snapshot releases are the simple result of incorporating Linux-specific
 210   support into the FreeBSD malloc sources.
 211
 212 --------------------------------------------------------------------------------
 213 vim:filetype=text:textwidth=80