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 * 3.0.0 (May 11, 2012)
  10
  11   Although this version adds some major new features, the primary focus is on
  12   internal code cleanup that facilitates maintainability and portability, most
  13   of which is not reflected in the ChangeLog.  This is the first release to
  14   incorporate substantial contributions from numerous other developers, and the
  15   result is a more broadly useful allocator (see the git revision history for
  16   contribution details).  Note that the license has been unified, thanks to
  17   Facebook granting a license under the same terms as the other copyright
  18   holders (see COPYING).
  19
  20   New features:
  21   - Implement Valgrind support, redzones, and quarantine.
  22   - Add support for additional platforms:
  23     + FreeBSD
  24     + Mac OS X Lion
  25     + MinGW
  26     + Windows (no support yet for replacing the system malloc)
  27   - Add support for additional architectures:
  28     + MIPS
  29     + SH4
  30     + Tilera
  31   - Add support for cross compiling.
  32   - Add nallocm(), which rounds a request size up to the nearest size class
  33     without actually allocating.
  34   - Implement aligned_alloc() (blame C11).
  35   - Add the "thread.tcache.enabled" mallctl.
  36   - Add the "opt.prof_final" mallctl.
  37   - Update pprof (from gperftools 2.0).
  38   - Add the --with-mangling option.
  39   - Add the --disable-experimental option.
  40   - Add the --disable-munmap option, and make it the default on Linux.
  41   - Add the --enable-mremap option, which disables use of mremap(2) by default.
  42
  43   Incompatible changes:
  44   - Enable stats by default.
  45   - Enable fill by default.
  46   - Disable lazy locking by default.
  47   - Rename the "tcache.flush" mallctl to "thread.tcache.flush".
  48   - Rename the "arenas.pagesize" mallctl to "arenas.page".
  49   - Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB).
  50   - Change the "opt.prof_accum" default from true to false.
  51
  52   Removed features:
  53   - Remove the swap feature, including the "config.swap", "swap.avail",
  54     "swap.prezeroed", "swap.nfds", and "swap.fds" mallctls.
  55   - Remove highruns statistics, including the
  56     "stats.arenas.<i>.bins.<j>.highruns" and
  57     "stats.arenas.<i>.lruns.<j>.highruns" mallctls.
  58   - As part of small size class refactoring, remove the "opt.lg_[qc]space_max",
  59     "arenas.cacheline", "arenas.subpage", "arenas.[tqcs]space_{min,max}", and
  60     "arenas.[tqcs]bins" mallctls.
  61   - Remove the "arenas.chunksize" mallctl.
  62   - Remove the "opt.lg_prof_tcmax" option.
  63   - Remove the "opt.lg_prof_bt_max" option.
  64   - Remove the "opt.lg_tcache_gc_sweep" option.
  65   - Remove the --disable-tiny option, including the "config.tiny" mallctl.
  66   - Remove the --enable-dynamic-page-shift configure option.
  67   - Remove the --enable-sysv configure option.
  68
  69   Bug fixes:
  70   - Fix a statistics-related bug in the "thread.arena" mallctl that could cause
  71     invalid statistics and crashes.
  72   - Work around TLS deallocation via free() on Linux.  This bug could cause
  73     write-after-free memory corruption.
  74   - Fix a potential deadlock that could occur during interval- and
  75     growth-triggered heap profile dumps.
  76   - Fix large calloc() zeroing bugs due to dropping chunk map unzeroed flags.
  77   - Fix chunk_alloc_dss() to stop claiming memory is zeroed.  This bug could
  78     cause memory corruption and crashes with --enable-dss specified.
  79   - Fix fork-related bugs that could cause deadlock in children between fork
  80     and exec.
  81   - Fix malloc_stats_print() to honor 'b' and 'l' in the opts parameter.
  82   - Fix realloc(p, 0) to act like free(p).
  83   - Do not enforce minimum alignment in memalign().
  84   - Check for NULL pointer in malloc_usable_size().
  85   - Fix an off-by-one heap profile statistics bug that could be observed in
  86     interval- and growth-triggered heap profiles.
  87   - Fix the "epoch" mallctl to update cached stats even if the passed in epoch
  88     is 0.
  89   - Fix bin->runcur management to fix a layout policy bug.  This bug did not
  90     affect correctness.
  91   - Fix a bug in choose_arena_hard() that potentially caused more arenas to be
  92     initialized than necessary.
  93   - Add missing "opt.lg_tcache_max" mallctl implementation.
  94   - Use glibc allocator hooks to make mixed allocator usage less likely.
  95   - Fix build issues for --disable-tcache.
  96   - Don't mangle pthread_create() when --with-private-namespace is specified.
  97
  98 * 2.2.5 (November 14, 2011)
  99
 100   Bug fixes:
 101   - Fix huge_ralloc() race when using mremap(2).  This is a serious bug that
 102     could cause memory corruption and/or crashes.
 103   - Fix huge_ralloc() to maintain chunk statistics.
 104   - Fix malloc_stats_print(..., "a") output.
 105
 106 * 2.2.4 (November 5, 2011)
 107
 108   Bug fixes:
 109   - Initialize arenas_tsd before using it.  This bug existed for 2.2.[0-3], as
 110     well as for --disable-tls builds in earlier releases.
 111   - Do not assume a 4 KiB page size in test/rallocm.c.
 112
 113 * 2.2.3 (August 31, 2011)
 114
 115   This version fixes numerous bugs related to heap profiling.
 116
 117   Bug fixes:
 118   - Fix a prof-related race condition.  This bug could cause memory corruption,
 119     but only occurred in non-default configurations (prof_accum:false).
 120   - Fix off-by-one backtracing issues (make sure that prof_alloc_prep() is
 121     excluded from backtraces).
 122   - Fix a prof-related bug in realloc() (only triggered by OOM errors).
 123   - Fix prof-related bugs in allocm() and rallocm().
 124   - Fix prof_tdata_cleanup() for --disable-tls builds.
 125   - Fix a relative include path, to fix objdir builds.
 126
 127 * 2.2.2 (July 30, 2011)
 128
 129   Bug fixes:
 130   - Fix a build error for --disable-tcache.
 131   - Fix assertions in arena_purge() (for real this time).
 132   - Add the --with-private-namespace option.  This is a workaround for symbol
 133     conflicts that can inadvertently arise when using static libraries.
 134
 135 * 2.2.1 (March 30, 2011)
 136
 137   Bug fixes:
 138   - Implement atomic operations for x86/x64.  This fixes compilation failures
 139     for versions of gcc that are still in wide use.
 140   - Fix an assertion in arena_purge().
 141
 142 * 2.2.0 (March 22, 2011)
 143
 144   This version incorporates several improvements to algorithms and data
 145   structures that tend to reduce fragmentation and increase speed.
 146
 147   New features:
 148   - Add the "stats.cactive" mallctl.
 149   - Update pprof (from google-perftools 1.7).
 150   - Improve backtracing-related configuration logic, and add the
 151     --disable-prof-libgcc option.
 152
 153   Bug fixes:
 154   - Change default symbol visibility from "internal", to "hidden", which
 155     decreases the overhead of library-internal function calls.
 156   - Fix symbol visibility so that it is also set on OS X.
 157   - Fix a build dependency regression caused by the introduction of the .pic.o
 158     suffix for PIC object files.
 159   - Add missing checks for mutex initialization failures.
 160   - Don't use libgcc-based backtracing except on x64, where it is known to work.
 161   - Fix deadlocks on OS X that were due to memory allocation in
 162     pthread_mutex_lock().
 163   - Heap profiling-specific fixes:
 164     + Fix memory corruption due to integer overflow in small region index
 165       computation, when using a small enough sample interval that profiling
 166       context pointers are stored in small run headers.
 167     + Fix a bootstrap ordering bug that only occurred with TLS disabled.
 168     + Fix a rallocm() rsize bug.
 169     + Fix error detection bugs for aligned memory allocation.
 170
 171 * 2.1.3 (March 14, 2011)
 172
 173   Bug fixes:
 174   - Fix a cpp logic regression (due to the "thread.{de,}allocatedp" mallctl fix
 175     for OS X in 2.1.2).
 176   - Fix a "thread.arena" mallctl bug.
 177   - Fix a thread cache stats merging bug.
 178
 179 * 2.1.2 (March 2, 2011)
 180
 181   Bug fixes:
 182   - Fix "thread.{de,}allocatedp" mallctl for OS X.
 183   - Add missing jemalloc.a to build system.
 184
 185 * 2.1.1 (January 31, 2011)
 186
 187   Bug fixes:
 188   - Fix aligned huge reallocation (affected allocm()).
 189   - Fix the ALLOCM_LG_ALIGN macro definition.
 190   - Fix a heap dumping deadlock.
 191   - Fix a "thread.arena" mallctl bug.
 192
 193 * 2.1.0 (December 3, 2010)
 194
 195   This version incorporates some optimizations that can't quite be considered
 196   bug fixes.
 197
 198   New features:
 199   - Use Linux's mremap(2) for huge object reallocation when possible.
 200   - Avoid locking in mallctl*() when possible.
 201   - Add the "thread.[de]allocatedp" mallctl's.
 202   - Convert the manual page source from roff to DocBook, and generate both roff
 203     and HTML manuals.
 204
 205   Bug fixes:
 206   - Fix a crash due to incorrect bootstrap ordering.  This only impacted
 207     --enable-debug --enable-dss configurations.
 208   - Fix a minor statistics bug for mallctl("swap.avail", ...).
 209
 210 * 2.0.1 (October 29, 2010)
 211
 212   Bug fixes:
 213   - Fix a race condition in heap profiling that could cause undefined behavior
 214     if "opt.prof_accum" were disabled.
 215   - Add missing mutex unlocks for some OOM error paths in the heap profiling
 216     code.
 217   - Fix a compilation error for non-C99 builds.
 218
 219 * 2.0.0 (October 24, 2010)
 220
 221   This version focuses on the experimental *allocm() API, and on improved
 222   run-time configuration/introspection.  Nonetheless, numerous performance
 223   improvements are also included.
 224
 225   New features:
 226   - Implement the experimental {,r,s,d}allocm() API, which provides a superset
 227     of the functionality available via malloc(), calloc(), posix_memalign(),
 228     realloc(), malloc_usable_size(), and free().  These functions can be used to
 229     allocate/reallocate aligned zeroed memory, ask for optional extra memory
 230     during reallocation, prevent object movement during reallocation, etc.
 231   - Replace JEMALLOC_OPTIONS/JEMALLOC_PROF_PREFIX with MALLOC_CONF, which is
 232     more human-readable, and more flexible.  For example:
 233       JEMALLOC_OPTIONS=AJP
 234     is now:
 235       MALLOC_CONF=abort:true,fill:true,stats_print:true
 236   - Port to Apple OS X.  Sponsored by Mozilla.
 237   - Make it possible for the application to control thread-->arena mappings via
 238     the "thread.arena" mallctl.
 239   - Add compile-time support for all TLS-related functionality via pthreads TSD.
 240     This is mainly of interest for OS X, which does not support TLS, but has a
 241     TSD implementation with similar performance.
 242   - Override memalign() and valloc() if they are provided by the system.
 243   - Add the "arenas.purge" mallctl, which can be used to synchronously purge all
 244     dirty unused pages.
 245   - Make cumulative heap profiling data optional, so that it is possible to
 246     limit the amount of memory consumed by heap profiling data structures.
 247   - Add per thread allocation counters that can be accessed via the
 248     "thread.allocated" and "thread.deallocated" mallctls.
 249
 250   Incompatible changes:
 251   - Remove JEMALLOC_OPTIONS and malloc_options (see MALLOC_CONF above).
 252   - Increase default backtrace depth from 4 to 128 for heap profiling.
 253   - Disable interval-based profile dumps by default.
 254
 255   Bug fixes:
 256   - Remove bad assertions in fork handler functions.  These assertions could
 257     cause aborts for some combinations of configure settings.
 258   - Fix strerror_r() usage to deal with non-standard semantics in GNU libc.
 259   - Fix leak context reporting.  This bug tended to cause the number of contexts
 260     to be underreported (though the reported number of objects and bytes were
 261     correct).
 262   - Fix a realloc() bug for large in-place growing reallocation.  This bug could
 263     cause memory corruption, but it was hard to trigger.
 264   - Fix an allocation bug for small allocations that could be triggered if
 265     multiple threads raced to create a new run of backing pages.
 266   - Enhance the heap profiler to trigger samples based on usable size, rather
 267     than request size.
 268   - Fix a heap profiling bug due to sometimes losing track of requested object
 269     size for sampled objects.
 270
 271 * 1.0.3 (August 12, 2010)
 272
 273   Bug fixes:
 274   - Fix the libunwind-based implementation of stack backtracing (used for heap
 275     profiling).  This bug could cause zero-length backtraces to be reported.
 276   - Add a missing mutex unlock in library initialization code.  If multiple
 277     threads raced to initialize malloc, some of them could end up permanently
 278     blocked.
 279
 280 * 1.0.2 (May 11, 2010)
 281
 282   Bug fixes:
 283   - Fix junk filling of large objects, which could cause memory corruption.
 284   - Add MAP_NORESERVE support for chunk mapping, because otherwise virtual
 285     memory limits could cause swap file configuration to fail.  Contributed by
 286     Jordan DeLong.
 287
 288 * 1.0.1 (April 14, 2010)
 289
 290   Bug fixes:
 291   - Fix compilation when --enable-fill is specified.
 292   - Fix threads-related profiling bugs that affected accuracy and caused memory
 293     to be leaked during thread exit.
 294   - Fix dirty page purging race conditions that could cause crashes.
 295   - Fix crash in tcache flushing code during thread destruction.
 296
 297 * 1.0.0 (April 11, 2010)
 298
 299   This release focuses on speed and run-time introspection.  Numerous
 300   algorithmic improvements make this release substantially faster than its
 301   predecessors.
 302
 303   New features:
 304   - Implement autoconf-based configuration system.
 305   - Add mallctl*(), for the purposes of introspection and run-time
 306     configuration.
 307   - Make it possible for the application to manually flush a thread's cache, via
 308     the "tcache.flush" mallctl.
 309   - Base maximum dirty page count on proportion of active memory.
 310   - Compute various addtional run-time statistics, including per size class
 311     statistics for large objects.
 312   - Expose malloc_stats_print(), which can be called repeatedly by the
 313     application.
 314   - Simplify the malloc_message() signature to only take one string argument,
 315     and incorporate an opaque data pointer argument for use by the application
 316     in combination with malloc_stats_print().
 317   - Add support for allocation backed by one or more swap files, and allow the
 318     application to disable over-commit if swap files are in use.
 319   - Implement allocation profiling and leak checking.
 320
 321   Removed features:
 322   - Remove the dynamic arena rebalancing code, since thread-specific caching
 323     reduces its utility.
 324
 325   Bug fixes:
 326   - Modify chunk allocation to work when address space layout randomization
 327     (ASLR) is in use.
 328   - Fix thread cleanup bugs related to TLS destruction.
 329   - Handle 0-size allocation requests in posix_memalign().
 330   - Fix a chunk leak.  The leaked chunks were never touched, so this impacted
 331     virtual memory usage, but not physical memory usage.
 332
 333 * linux_2008082[78]a (August 27/28, 2008)
 334
 335   These snapshot releases are the simple result of incorporating Linux-specific
 336   support into the FreeBSD malloc sources.
 337
 338 --------------------------------------------------------------------------------
 339 vim:filetype=text:textwidth=80