1Following are change highlights associated with official releases. Important 2bug fixes are all mentioned, but some internal enhancements are omitted here for 3brevity. Much more detail can be found in the git revision history: 4 5 https://github.com/jemalloc/jemalloc 6 7* 4.5.0 (February 28, 2017) 8 9 This is the first release to benefit from much broader continuous integration 10 testing, thanks to @davidtgoldblatt. Had we had this testing infrastructure 11 in place for prior releases, it would have caught all of the most serious 12 regressions fixed by this release. 13 14 New features: 15 - Add --disable-thp and the opt.thp to provide opt-out mechanisms for 16 transparent huge page integration. (@jasone) 17 - Update zone allocator integration to work with macOS 10.12. (@glandium) 18 - Restructure *CFLAGS configuration, so that CFLAGS behaves typically, and 19 EXTRA_CFLAGS provides a way to specify e.g. -Werror during building, but not 20 during configuration. (@jasone, @ronawho) 21 22 Bug fixes: 23 - Fix DSS (sbrk(2)-based) allocation. This regression was first released in 24 4.3.0. (@jasone) 25 - Handle race in per size class utilization computation. This functionality 26 was first released in 4.0.0. (@interwq) 27 - Fix lock order reversal during gdump. (@jasone) 28 - Fix-refactor tcache synchronization. This regression was first released in 29 4.0.0. (@jasone) 30 - Fix various JSON-formatted malloc_stats_print() bugs. This functionality 31 was first released in 4.3.0. (@jasone) 32 - Fix huge-aligned allocation. This regression was first released in 4.4.0. 33 (@jasone) 34 - When transparent huge page integration is enabled, detect what state pages 35 start in according to the kernel's current operating mode, and only convert 36 arena chunks to non-huge during purging if that is not their initial state. 37 This functionality was first released in 4.4.0. (@jasone) 38 - Fix lg_chunk clamping for the --enable-cache-oblivious --disable-fill case. 39 This regression was first released in 4.0.0. (@jasone, @428desmo) 40 - Properly detect sparc64 when building for Linux. (@glaubitz) 41 42* 4.4.0 (December 3, 2016) 43 44 New features: 45 - Add configure support for *-*-linux-android. (@cferris1000, @jasone) 46 - Add the --disable-syscall configure option, for use on systems that place 47 security-motivated limitations on syscall(2). (@jasone) 48 - Add support for Debian GNU/kFreeBSD. (@thesam) 49 50 Optimizations: 51 - Add extent serial numbers and use them where appropriate as a sort key that 52 is higher priority than address, so that the allocation policy prefers older 53 extents. This tends to improve locality (decrease fragmentation) when 54 memory grows downward. (@jasone) 55 - Refactor madvise(2) configuration so that MADV_FREE is detected and utilized 56 on Linux 4.5 and newer. (@jasone) 57 - Mark partially purged arena chunks as non-huge-page. This improves 58 interaction with Linux's transparent huge page functionality. (@jasone) 59 60 Bug fixes: 61 - Fix size class computations for edge conditions involving extremely large 62 allocations. This regression was first released in 4.0.0. (@jasone, 63 @ingvarha) 64 - Remove overly restrictive assertions related to the cactive statistic. This 65 regression was first released in 4.1.0. (@jasone) 66 - Implement a more reliable detection scheme for os_unfair_lock on macOS. 67 (@jszakmeister) 68 69* 4.3.1 (November 7, 2016) 70 71 Bug fixes: 72 - Fix a severe virtual memory leak. This regression was first released in 73 4.3.0. (@interwq, @jasone) 74 - Refactor atomic and prng APIs to restore support for 32-bit platforms that 75 use pre-C11 toolchains, e.g. FreeBSD's mips. (@jasone) 76 77* 4.3.0 (November 4, 2016) 78 79 This is the first release that passes the test suite for multiple Windows 80 configurations, thanks in large part to @glandium setting up continuous 81 integration via AppVeyor (and Travis CI for Linux and OS X). 82 83 New features: 84 - Add "J" (JSON) support to malloc_stats_print(). (@jasone) 85 - Add Cray compiler support. (@ronawho) 86 87 Optimizations: 88 - Add/use adaptive spinning for bootstrapping and radix tree node 89 initialization. (@jasone) 90 91 Bug fixes: 92 - Fix large allocation to search starting in the optimal size class heap, 93 which can substantially reduce virtual memory churn and fragmentation. This 94 regression was first released in 4.0.0. (@mjp41, @jasone) 95 - Fix stats.arenas.<i>.nthreads accounting. (@interwq) 96 - Fix and simplify decay-based purging. (@jasone) 97 - Make DSS (sbrk(2)-related) operations lockless, which resolves potential 98 deadlocks during thread exit. (@jasone) 99 - Fix over-sized allocation of radix tree leaf nodes. (@mjp41, @ogaun, 100 @jasone) 101 - Fix over-sized allocation of arena_t (plus associated stats) data 102 structures. (@jasone, @interwq) 103 - Fix EXTRA_CFLAGS to not affect configuration. (@jasone) 104 - Fix a Valgrind integration bug. (@ronawho) 105 - Disallow 0x5a junk filling when running in Valgrind. (@jasone) 106 - Fix a file descriptor leak on Linux. This regression was first released in 107 4.2.0. (@vsarunas, @jasone) 108 - Fix static linking of jemalloc with glibc. (@djwatson) 109 - Use syscall(2) rather than {open,read,close}(2) during boot on Linux. This 110 works around other libraries' system call wrappers performing reentrant 111 allocation. (@kspinka, @Whissi, @jasone) 112 - Fix OS X default zone replacement to work with OS X 10.12. (@glandium, 113 @jasone) 114 - Fix cached memory management to avoid needless commit/decommit operations 115 during purging, which resolves permanent virtual memory map fragmentation 116 issues on Windows. (@mjp41, @jasone) 117 - Fix TSD fetches to avoid (recursive) allocation. This is relevant to 118 non-TLS and Windows configurations. (@jasone) 119 - Fix malloc_conf overriding to work on Windows. (@jasone) 120 - Forcibly disable lazy-lock on Windows (was forcibly *enabled*). (@jasone) 121 122* 4.2.1 (June 8, 2016) 123 124 Bug fixes: 125 - Fix bootstrapping issues for configurations that require allocation during 126 tsd initialization (e.g. --disable-tls). (@cferris1000, @jasone) 127 - Fix gettimeofday() version of nstime_update(). (@ronawho) 128 - Fix Valgrind regressions in calloc() and chunk_alloc_wrapper(). (@ronawho) 129 - Fix potential VM map fragmentation regression. (@jasone) 130 - Fix opt_zero-triggered in-place huge reallocation zeroing. (@jasone) 131 - Fix heap profiling context leaks in reallocation edge cases. (@jasone) 132 133* 4.2.0 (May 12, 2016) 134 135 New features: 136 - Add the arena.<i>.reset mallctl, which makes it possible to discard all of 137 an arena's allocations in a single operation. (@jasone) 138 - Add the stats.retained and stats.arenas.<i>.retained statistics. (@jasone) 139 - Add the --with-version configure option. (@jasone) 140 - Support --with-lg-page values larger than actual page size. (@jasone) 141 142 Optimizations: 143 - Use pairing heaps rather than red-black trees for various hot data 144 structures. (@djwatson, @jasone) 145 - Streamline fast paths of rtree operations. (@jasone) 146 - Optimize the fast paths of calloc() and [m,d,sd]allocx(). (@jasone) 147 - Decommit unused virtual memory if the OS does not overcommit. (@jasone) 148 - Specify MAP_NORESERVE on Linux if [heuristic] overcommit is active, in order 149 to avoid unfortunate interactions during fork(2). (@jasone) 150 151 Bug fixes: 152 - Fix chunk accounting related to triggering gdump profiles. (@jasone) 153 - Link against librt for clock_gettime(2) if glibc < 2.17. (@jasone) 154 - Scale leak report summary according to sampling probability. (@jasone) 155 156* 4.1.1 (May 3, 2016) 157 158 This bugfix release resolves a variety of mostly minor issues, though the 159 bitmap fix is critical for 64-bit Windows. 160 161 Bug fixes: 162 - Fix the linear scan version of bitmap_sfu() to shift by the proper amount 163 even when sizeof(long) is not the same as sizeof(void *), as on 64-bit 164 Windows. (@jasone) 165 - Fix hashing functions to avoid unaligned memory accesses (and resulting 166 crashes). This is relevant at least to some ARM-based platforms. 167 (@rkmisra) 168 - Fix fork()-related lock rank ordering reversals. These reversals were 169 unlikely to cause deadlocks in practice except when heap profiling was 170 enabled and active. (@jasone) 171 - Fix various chunk leaks in OOM code paths. (@jasone) 172 - Fix malloc_stats_print() to print opt.narenas correctly. (@jasone) 173 - Fix MSVC-specific build/test issues. (@rustyx, @yuslepukhin) 174 - Fix a variety of test failures that were due to test fragility rather than 175 core bugs. (@jasone) 176 177* 4.1.0 (February 28, 2016) 178 179 This release is primarily about optimizations, but it also incorporates a lot 180 of portability-motivated refactoring and enhancements. Many people worked on 181 this release, to an extent that even with the omission here of minor changes 182 (see git revision history), and of the people who reported and diagnosed 183 issues, so much of the work was contributed that starting with this release, 184 changes are annotated with author credits to help reflect the collaborative 185 effort involved. 186 187 New features: 188 - Implement decay-based unused dirty page purging, a major optimization with 189 mallctl API impact. This is an alternative to the existing ratio-based 190 unused dirty page purging, and is intended to eventually become the sole 191 purging mechanism. New mallctls: 192 + opt.purge 193 + opt.decay_time 194 + arena.<i>.decay 195 + arena.<i>.decay_time 196 + arenas.decay_time 197 + stats.arenas.<i>.decay_time 198 (@jasone, @cevans87) 199 - Add --with-malloc-conf, which makes it possible to embed a default 200 options string during configuration. This was motivated by the desire to 201 specify --with-malloc-conf=purge:decay , since the default must remain 202 purge:ratio until the 5.0.0 release. (@jasone) 203 - Add MS Visual Studio 2015 support. (@rustyx, @yuslepukhin) 204 - Make *allocx() size class overflow behavior defined. The maximum 205 size class is now less than PTRDIFF_MAX to protect applications against 206 numerical overflow, and all allocation functions are guaranteed to indicate 207 errors rather than potentially crashing if the request size exceeds the 208 maximum size class. (@jasone) 209 - jeprof: 210 + Add raw heap profile support. (@jasone) 211 + Add --retain and --exclude for backtrace symbol filtering. (@jasone) 212 213 Optimizations: 214 - Optimize the fast path to combine various bootstrapping and configuration 215 checks and execute more streamlined code in the common case. (@interwq) 216 - Use linear scan for small bitmaps (used for small object tracking). In 217 addition to speeding up bitmap operations on 64-bit systems, this reduces 218 allocator metadata overhead by approximately 0.2%. (@djwatson) 219 - Separate arena_avail trees, which substantially speeds up run tree 220 operations. (@djwatson) 221 - Use memoization (boot-time-computed table) for run quantization. Separate 222 arena_avail trees reduced the importance of this optimization. (@jasone) 223 - Attempt mmap-based in-place huge reallocation. This can dramatically speed 224 up incremental huge reallocation. (@jasone) 225 226 Incompatible changes: 227 - Make opt.narenas unsigned rather than size_t. (@jasone) 228 229 Bug fixes: 230 - Fix stats.cactive accounting regression. (@rustyx, @jasone) 231 - Handle unaligned keys in hash(). This caused problems for some ARM systems. 232 (@jasone, @cferris1000) 233 - Refactor arenas array. In addition to fixing a fork-related deadlock, this 234 makes arena lookups faster and simpler. (@jasone) 235 - Move retained memory allocation out of the default chunk allocation 236 function, to a location that gets executed even if the application installs 237 a custom chunk allocation function. This resolves a virtual memory leak. 238 (@buchgr) 239 - Fix a potential tsd cleanup leak. (@cferris1000, @jasone) 240 - Fix run quantization. In practice this bug had no impact unless 241 applications requested memory with alignment exceeding one page. 242 (@jasone, @djwatson) 243 - Fix LinuxThreads-specific bootstrapping deadlock. (Cosmin Paraschiv) 244 - jeprof: 245 + Don't discard curl options if timeout is not defined. (@djwatson) 246 + Detect failed profile fetches. (@djwatson) 247 - Fix stats.arenas.<i>.{dss,lg_dirty_mult,decay_time,pactive,pdirty} for 248 --disable-stats case. (@jasone) 249 250* 4.0.4 (October 24, 2015) 251 252 This bugfix release fixes another xallocx() regression. No other regressions 253 have come to light in over a month, so this is likely a good starting point 254 for people who prefer to wait for "dot one" releases with all the major issues 255 shaken out. 256 257 Bug fixes: 258 - Fix xallocx(..., MALLOCX_ZERO to zero the last full trailing page of large 259 allocations that have been randomly assigned an offset of 0 when 260 --enable-cache-oblivious configure option is enabled. 261 262* 4.0.3 (September 24, 2015) 263 264 This bugfix release continues the trend of xallocx() and heap profiling fixes. 265 266 Bug fixes: 267 - Fix xallocx(..., MALLOCX_ZERO) to zero all trailing bytes of large 268 allocations when --enable-cache-oblivious configure option is enabled. 269 - Fix xallocx(..., MALLOCX_ZERO) to zero trailing bytes of huge allocations 270 when resizing from/to a size class that is not a multiple of the chunk size. 271 - Fix prof_tctx_dump_iter() to filter out nodes that were created after heap 272 profile dumping started. 273 - Work around a potentially bad thread-specific data initialization 274 interaction with NPTL (glibc's pthreads implementation). 275 276* 4.0.2 (September 21, 2015) 277 278 This bugfix release addresses a few bugs specific to heap profiling. 279 280 Bug fixes: 281 - Fix ixallocx_prof_sample() to never modify nor create sampled small 282 allocations. xallocx() is in general incapable of moving small allocations, 283 so this fix removes buggy code without loss of generality. 284 - Fix irallocx_prof_sample() to always allocate large regions, even when 285 alignment is non-zero. 286 - Fix prof_alloc_rollback() to read tdata from thread-specific data rather 287 than dereferencing a potentially invalid tctx. 288 289* 4.0.1 (September 15, 2015) 290 291 This is a bugfix release that is somewhat high risk due to the amount of 292 refactoring required to address deep xallocx() problems. As a side effect of 293 these fixes, xallocx() now tries harder to partially fulfill requests for 294 optional extra space. Note that a couple of minor heap profiling 295 optimizations are included, but these are better thought of as performance 296 fixes that were integral to disovering most of the other bugs. 297 298 Optimizations: 299 - Avoid a chunk metadata read in arena_prof_tctx_set(), since it is in the 300 fast path when heap profiling is enabled. Additionally, split a special 301 case out into arena_prof_tctx_reset(), which also avoids chunk metadata 302 reads. 303 - Optimize irallocx_prof() to optimistically update the sampler state. The 304 prior implementation appears to have been a holdover from when 305 rallocx()/xallocx() functionality was combined as rallocm(). 306 307 Bug fixes: 308 - Fix TLS configuration such that it is enabled by default for platforms on 309 which it works correctly. 310 - Fix arenas_cache_cleanup() and arena_get_hard() to handle 311 allocation/deallocation within the application's thread-specific data 312 cleanup functions even after arenas_cache is torn down. 313 - Fix xallocx() bugs related to size+extra exceeding HUGE_MAXCLASS. 314 - Fix chunk purge hook calls for in-place huge shrinking reallocation to 315 specify the old chunk size rather than the new chunk size. This bug caused 316 no correctness issues for the default chunk purge function, but was 317 visible to custom functions set via the "arena.<i>.chunk_hooks" mallctl. 318 - Fix heap profiling bugs: 319 + Fix heap profiling to distinguish among otherwise identical sample sites 320 with interposed resets (triggered via the "prof.reset" mallctl). This bug 321 could cause data structure corruption that would most likely result in a 322 segfault. 323 + Fix irealloc_prof() to prof_alloc_rollback() on OOM. 324 + Make one call to prof_active_get_unlocked() per allocation event, and use 325 the result throughout the relevant functions that handle an allocation 326 event. Also add a missing check in prof_realloc(). These fixes protect 327 allocation events against concurrent prof_active changes. 328 + Fix ixallocx_prof() to pass usize_max and zero to ixallocx_prof_sample() 329 in the correct order. 330 + Fix prof_realloc() to call prof_free_sampled_object() after calling 331 prof_malloc_sample_object(). Prior to this fix, if tctx and old_tctx were 332 the same, the tctx could have been prematurely destroyed. 333 - Fix portability bugs: 334 + Don't bitshift by negative amounts when encoding/decoding run sizes in 335 chunk header maps. This affected systems with page sizes greater than 8 336 KiB. 337 + Rename index_t to szind_t to avoid an existing type on Solaris. 338 + Add JEMALLOC_CXX_THROW to the memalign() function prototype, in order to 339 match glibc and avoid compilation errors when including both 340 jemalloc/jemalloc.h and malloc.h in C++ code. 341 + Don't assume that /bin/sh is appropriate when running size_classes.sh 342 during configuration. 343 + Consider __sparcv9 a synonym for __sparc64__ when defining LG_QUANTUM. 344 + Link tests to librt if it contains clock_gettime(2). 345 346* 4.0.0 (August 17, 2015) 347 348 This version contains many speed and space optimizations, both minor and 349 major. The major themes are generalization, unification, and simplification. 350 Although many of these optimizations cause no visible behavior change, their 351 cumulative effect is substantial. 352 353 New features: 354 - Normalize size class spacing to be consistent across the complete size 355 range. By default there are four size classes per size doubling, but this 356 is now configurable via the --with-lg-size-class-group option. Also add the 357 --with-lg-page, --with-lg-page-sizes, --with-lg-quantum, and 358 --with-lg-tiny-min options, which can be used to tweak page and size class 359 settings. Impacts: 360 + Worst case performance for incrementally growing/shrinking reallocation 361 is improved because there are far fewer size classes, and therefore 362 copying happens less often. 363 + Internal fragmentation is limited to 20% for all but the smallest size 364 classes (those less than four times the quantum). (1B + 4 KiB) 365 and (1B + 4 MiB) previously suffered nearly 50% internal fragmentation. 366 + Chunk fragmentation tends to be lower because there are fewer distinct run 367 sizes to pack. 368 - Add support for explicit tcaches. The "tcache.create", "tcache.flush", and 369 "tcache.destroy" mallctls control tcache lifetime and flushing, and the 370 MALLOCX_TCACHE(tc) and MALLOCX_TCACHE_NONE flags to the *allocx() API 371 control which tcache is used for each operation. 372 - Implement per thread heap profiling, as well as the ability to 373 enable/disable heap profiling on a per thread basis. Add the "prof.reset", 374 "prof.lg_sample", "thread.prof.name", "thread.prof.active", 375 "opt.prof_thread_active_init", "prof.thread_active_init", and 376 "thread.prof.active" mallctls. 377 - Add support for per arena application-specified chunk allocators, configured 378 via the "arena.<i>.chunk_hooks" mallctl. 379 - Refactor huge allocation to be managed by arenas, so that arenas now 380 function as general purpose independent allocators. This is important in 381 the context of user-specified chunk allocators, aside from the scalability 382 benefits. Related new statistics: 383 + The "stats.arenas.<i>.huge.allocated", "stats.arenas.<i>.huge.nmalloc", 384 "stats.arenas.<i>.huge.ndalloc", and "stats.arenas.<i>.huge.nrequests" 385 mallctls provide high level per arena huge allocation statistics. 386 + The "arenas.nhchunks", "arenas.hchunk.<i>.size", 387 "stats.arenas.<i>.hchunks.<j>.nmalloc", 388 "stats.arenas.<i>.hchunks.<j>.ndalloc", 389 "stats.arenas.<i>.hchunks.<j>.nrequests", and 390 "stats.arenas.<i>.hchunks.<j>.curhchunks" mallctls provide per size class 391 statistics. 392 - Add the 'util' column to malloc_stats_print() output, which reports the 393 proportion of available regions that are currently in use for each small 394 size class. 395 - Add "alloc" and "free" modes for for junk filling (see the "opt.junk" 396 mallctl), so that it is possible to separately enable junk filling for 397 allocation versus deallocation. 398 - Add the jemalloc-config script, which provides information about how 399 jemalloc was configured, and how to integrate it into application builds. 400 - Add metadata statistics, which are accessible via the "stats.metadata", 401 "stats.arenas.<i>.metadata.mapped", and 402 "stats.arenas.<i>.metadata.allocated" mallctls. 403 - Add the "stats.resident" mallctl, which reports the upper limit of 404 physically resident memory mapped by the allocator. 405 - Add per arena control over unused dirty page purging, via the 406 "arenas.lg_dirty_mult", "arena.<i>.lg_dirty_mult", and 407 "stats.arenas.<i>.lg_dirty_mult" mallctls. 408 - Add the "prof.gdump" mallctl, which makes it possible to toggle the gdump 409 feature on/off during program execution. 410 - Add sdallocx(), which implements sized deallocation. The primary 411 optimization over dallocx() is the removal of a metadata read, which often 412 suffers an L1 cache miss. 413 - Add missing header includes in jemalloc/jemalloc.h, so that applications 414 only have to #include <jemalloc/jemalloc.h>. 415 - Add support for additional platforms: 416 + Bitrig 417 + Cygwin 418 + DragonFlyBSD 419 + iOS 420 + OpenBSD 421 + OpenRISC/or1k 422 423 Optimizations: 424 - Maintain dirty runs in per arena LRUs rather than in per arena trees of 425 dirty-run-containing chunks. In practice this change significantly reduces 426 dirty page purging volume. 427 - Integrate whole chunks into the unused dirty page purging machinery. This 428 reduces the cost of repeated huge allocation/deallocation, because it 429 effectively introduces a cache of chunks. 430 - Split the arena chunk map into two separate arrays, in order to increase 431 cache locality for the frequently accessed bits. 432 - Move small run metadata out of runs, into arena chunk headers. This reduces 433 run fragmentation, smaller runs reduce external fragmentation for small size 434 classes, and packed (less uniformly aligned) metadata layout improves CPU 435 cache set distribution. 436 - Randomly distribute large allocation base pointer alignment relative to page 437 boundaries in order to more uniformly utilize CPU cache sets. This can be 438 disabled via the --disable-cache-oblivious configure option, and queried via 439 the "config.cache_oblivious" mallctl. 440 - Micro-optimize the fast paths for the public API functions. 441 - Refactor thread-specific data to reside in a single structure. This assures 442 that only a single TLS read is necessary per call into the public API. 443 - Implement in-place huge allocation growing and shrinking. 444 - Refactor rtree (radix tree for chunk lookups) to be lock-free, and make 445 additional optimizations that reduce maximum lookup depth to one or two 446 levels. This resolves what was a concurrency bottleneck for per arena huge 447 allocation, because a global data structure is critical for determining 448 which arenas own which huge allocations. 449 450 Incompatible changes: 451 - Replace --enable-cc-silence with --disable-cc-silence to suppress spurious 452 warnings by default. 453 - Assure that the constness of malloc_usable_size()'s return type matches that 454 of the system implementation. 455 - Change the heap profile dump format to support per thread heap profiling, 456 rename pprof to jeprof, and enhance it with the --thread=<n> option. As a 457 result, the bundled jeprof must now be used rather than the upstream 458 (gperftools) pprof. 459 - Disable "opt.prof_final" by default, in order to avoid atexit(3), which can 460 internally deadlock on some platforms. 461 - Change the "arenas.nlruns" mallctl type from size_t to unsigned. 462 - Replace the "stats.arenas.<i>.bins.<j>.allocated" mallctl with 463 "stats.arenas.<i>.bins.<j>.curregs". 464 - Ignore MALLOC_CONF in set{uid,gid,cap} binaries. 465 - Ignore MALLOCX_ARENA(a) in dallocx(), in favor of using the 466 MALLOCX_TCACHE(tc) and MALLOCX_TCACHE_NONE flags to control tcache usage. 467 468 Removed features: 469 - Remove the *allocm() API, which is superseded by the *allocx() API. 470 - Remove the --enable-dss options, and make dss non-optional on all platforms 471 which support sbrk(2). 472 - Remove the "arenas.purge" mallctl, which was obsoleted by the 473 "arena.<i>.purge" mallctl in 3.1.0. 474 - Remove the unnecessary "opt.valgrind" mallctl; jemalloc automatically 475 detects whether it is running inside Valgrind. 476 - Remove the "stats.huge.allocated", "stats.huge.nmalloc", and 477 "stats.huge.ndalloc" mallctls. 478 - Remove the --enable-mremap option. 479 - Remove the "stats.chunks.current", "stats.chunks.total", and 480 "stats.chunks.high" mallctls. 481 482 Bug fixes: 483 - Fix the cactive statistic to decrease (rather than increase) when active 484 memory decreases. This regression was first released in 3.5.0. 485 - Fix OOM handling in memalign() and valloc(). A variant of this bug existed 486 in all releases since 2.0.0, which introduced these functions. 487 - Fix an OOM-related regression in arena_tcache_fill_small(), which could 488 cause cache corruption on OOM. This regression was present in all releases 489 from 2.2.0 through 3.6.0. 490 - Fix size class overflow handling for malloc(), posix_memalign(), memalign(), 491 calloc(), and realloc() when profiling is enabled. 492 - Fix the "arena.<i>.dss" mallctl to return an error if "primary" or 493 "secondary" precedence is specified, but sbrk(2) is not supported. 494 - Fix fallback lg_floor() implementations to handle extremely large inputs. 495 - Ensure the default purgeable zone is after the default zone on OS X. 496 - Fix latent bugs in atomic_*(). 497 - Fix the "arena.<i>.dss" mallctl to handle read-only calls. 498 - Fix tls_model configuration to enable the initial-exec model when possible. 499 - Mark malloc_conf as a weak symbol so that the application can override it. 500 - Correctly detect glibc's adaptive pthread mutexes. 501 - Fix the --without-export configure option. 502 503* 3.6.0 (March 31, 2014) 504 505 This version contains a critical bug fix for a regression present in 3.5.0 and 506 3.5.1. 507 508 Bug fixes: 509 - Fix a regression in arena_chunk_alloc() that caused crashes during 510 small/large allocation if chunk allocation failed. In the absence of this 511 bug, chunk allocation failure would result in allocation failure, e.g. NULL 512 return from malloc(). This regression was introduced in 3.5.0. 513 - Fix backtracing for gcc intrinsics-based backtracing by specifying 514 -fno-omit-frame-pointer to gcc. Note that the application (and all the 515 libraries it links to) must also be compiled with this option for 516 backtracing to be reliable. 517 - Use dss allocation precedence for huge allocations as well as small/large 518 allocations. 519 - Fix test assertion failure message formatting. This bug did not manifest on 520 x86_64 systems because of implementation subtleties in va_list. 521 - Fix inconsequential test failures for hash and SFMT code. 522 523 New features: 524 - Support heap profiling on FreeBSD. This feature depends on the proc 525 filesystem being mounted during heap profile dumping. 526 527* 3.5.1 (February 25, 2014) 528 529 This version primarily addresses minor bugs in test code. 530 531 Bug fixes: 532 - Configure Solaris/Illumos to use MADV_FREE. 533 - Fix junk filling for mremap(2)-based huge reallocation. This is only 534 relevant if configuring with the --enable-mremap option specified. 535 - Avoid compilation failure if 'restrict' C99 keyword is not supported by the 536 compiler. 537 - Add a configure test for SSE2 rather than assuming it is usable on i686 538 systems. This fixes test compilation errors, especially on 32-bit Linux 539 systems. 540 - Fix mallctl argument size mismatches (size_t vs. uint64_t) in the stats unit 541 test. 542 - Fix/remove flawed alignment-related overflow tests. 543 - Prevent compiler optimizations that could change backtraces in the 544 prof_accum unit test. 545 546* 3.5.0 (January 22, 2014) 547 548 This version focuses on refactoring and automated testing, though it also 549 includes some non-trivial heap profiling optimizations not mentioned below. 550 551 New features: 552 - Add the *allocx() API, which is a successor to the experimental *allocm() 553 API. The *allocx() functions are slightly simpler to use because they have 554 fewer parameters, they directly return the results of primary interest, and 555 mallocx()/rallocx() avoid the strict aliasing pitfall that 556 allocm()/rallocm() share with posix_memalign(). Note that *allocm() is 557 slated for removal in the next non-bugfix release. 558 - Add support for LinuxThreads. 559 560 Bug fixes: 561 - Unless heap profiling is enabled, disable floating point code and don't link 562 with libm. This, in combination with e.g. EXTRA_CFLAGS=-mno-sse on x64 563 systems, makes it possible to completely disable floating point register 564 use. Some versions of glibc neglect to save/restore caller-saved floating 565 point registers during dynamic lazy symbol loading, and the symbol loading 566 code uses whatever malloc the application happens to have linked/loaded 567 with, the result being potential floating point register corruption. 568 - Report ENOMEM rather than EINVAL if an OOM occurs during heap profiling 569 backtrace creation in imemalign(). This bug impacted posix_memalign() and 570 aligned_alloc(). 571 - Fix a file descriptor leak in a prof_dump_maps() error path. 572 - Fix prof_dump() to close the dump file descriptor for all relevant error 573 paths. 574 - Fix rallocm() to use the arena specified by the ALLOCM_ARENA(s) flag for 575 allocation, not just deallocation. 576 - Fix a data race for large allocation stats counters. 577 - Fix a potential infinite loop during thread exit. This bug occurred on 578 Solaris, and could affect other platforms with similar pthreads TSD 579 implementations. 580 - Don't junk-fill reallocations unless usable size changes. This fixes a 581 violation of the *allocx()/*allocm() semantics. 582 - Fix growing large reallocation to junk fill new space. 583 - Fix huge deallocation to junk fill when munmap is disabled. 584 - Change the default private namespace prefix from empty to je_, and change 585 --with-private-namespace-prefix so that it prepends an additional prefix 586 rather than replacing je_. This reduces the likelihood of applications 587 which statically link jemalloc experiencing symbol name collisions. 588 - Add missing private namespace mangling (relevant when 589 --with-private-namespace is specified). 590 - Add and use JEMALLOC_INLINE_C so that static inline functions are marked as 591 static even for debug builds. 592 - Add a missing mutex unlock in a malloc_init_hard() error path. In practice 593 this error path is never executed. 594 - Fix numerous bugs in malloc_strotumax() error handling/reporting. These 595 bugs had no impact except for malformed inputs. 596 - Fix numerous bugs in malloc_snprintf(). These bugs were not exercised by 597 existing calls, so they had no impact. 598 599* 3.4.1 (October 20, 2013) 600 601 Bug fixes: 602 - Fix a race in the "arenas.extend" mallctl that could cause memory corruption 603 of internal data structures and subsequent crashes. 604 - Fix Valgrind integration flaws that caused Valgrind warnings about reads of 605 uninitialized memory in: 606 + arena chunk headers 607 + internal zero-initialized data structures (relevant to tcache and prof 608 code) 609 - Preserve errno during the first allocation. A readlink(2) call during 610 initialization fails unless /etc/malloc.conf exists, so errno was typically 611 set during the first allocation prior to this fix. 612 - Fix compilation warnings reported by gcc 4.8.1. 613 614* 3.4.0 (June 2, 2013) 615 616 This version is essentially a small bugfix release, but the addition of 617 aarch64 support requires that the minor version be incremented. 618 619 Bug fixes: 620 - Fix race-triggered deadlocks in chunk_record(). These deadlocks were 621 typically triggered by multiple threads concurrently deallocating huge 622 objects. 623 624 New features: 625 - Add support for the aarch64 architecture. 626 627* 3.3.1 (March 6, 2013) 628 629 This version fixes bugs that are typically encountered only when utilizing 630 custom run-time options. 631 632 Bug fixes: 633 - Fix a locking order bug that could cause deadlock during fork if heap 634 profiling were enabled. 635 - Fix a chunk recycling bug that could cause the allocator to lose track of 636 whether a chunk was zeroed. On FreeBSD, NetBSD, and OS X, it could cause 637 corruption if allocating via sbrk(2) (unlikely unless running with the 638 "dss:primary" option specified). This was completely harmless on Linux 639 unless using mlockall(2) (and unlikely even then, unless the 640 --disable-munmap configure option or the "dss:primary" option was 641 specified). This regression was introduced in 3.1.0 by the 642 mlockall(2)/madvise(2) interaction fix. 643 - Fix TLS-related memory corruption that could occur during thread exit if the 644 thread never allocated memory. Only the quarantine and prof facilities were 645 susceptible. 646 - Fix two quarantine bugs: 647 + Internal reallocation of the quarantined object array leaked the old 648 array. 649 + Reallocation failure for internal reallocation of the quarantined object 650 array (very unlikely) resulted in memory corruption. 651 - Fix Valgrind integration to annotate all internally allocated memory in a 652 way that keeps Valgrind happy about internal data structure access. 653 - Fix building for s390 systems. 654 655* 3.3.0 (January 23, 2013) 656 657 This version includes a few minor performance improvements in addition to the 658 listed new features and bug fixes. 659 660 New features: 661 - Add clipping support to lg_chunk option processing. 662 - Add the --enable-ivsalloc option. 663 - Add the --without-export option. 664 - Add the --disable-zone-allocator option. 665 666 Bug fixes: 667 - Fix "arenas.extend" mallctl to output the number of arenas. 668 - Fix chunk_recycle() to unconditionally inform Valgrind that returned memory 669 is undefined. 670 - Fix build break on FreeBSD related to alloca.h. 671 672* 3.2.0 (November 9, 2012) 673 674 In addition to a couple of bug fixes, this version modifies page run 675 allocation and dirty page purging algorithms in order to better control 676 page-level virtual memory fragmentation. 677 678 Incompatible changes: 679 - Change the "opt.lg_dirty_mult" default from 5 to 3 (32:1 to 8:1). 680 681 Bug fixes: 682 - Fix dss/mmap allocation precedence code to use recyclable mmap memory only 683 after primary dss allocation fails. 684 - Fix deadlock in the "arenas.purge" mallctl. This regression was introduced 685 in 3.1.0 by the addition of the "arena.<i>.purge" mallctl. 686 687* 3.1.0 (October 16, 2012) 688 689 New features: 690 - Auto-detect whether running inside Valgrind, thus removing the need to 691 manually specify MALLOC_CONF=valgrind:true. 692 - Add the "arenas.extend" mallctl, which allows applications to create 693 manually managed arenas. 694 - Add the ALLOCM_ARENA() flag for {,r,d}allocm(). 695 - Add the "opt.dss", "arena.<i>.dss", and "stats.arenas.<i>.dss" mallctls, 696 which provide control over dss/mmap precedence. 697 - Add the "arena.<i>.purge" mallctl, which obsoletes "arenas.purge". 698 - Define LG_QUANTUM for hppa. 699 700 Incompatible changes: 701 - Disable tcache by default if running inside Valgrind, in order to avoid 702 making unallocated objects appear reachable to Valgrind. 703 - Drop const from malloc_usable_size() argument on Linux. 704 705 Bug fixes: 706 - Fix heap profiling crash if sampled object is freed via realloc(p, 0). 707 - Remove const from __*_hook variable declarations, so that glibc can modify 708 them during process forking. 709 - Fix mlockall(2)/madvise(2) interaction. 710 - Fix fork(2)-related deadlocks. 711 - Fix error return value for "thread.tcache.enabled" mallctl. 712 713* 3.0.0 (May 11, 2012) 714 715 Although this version adds some major new features, the primary focus is on 716 internal code cleanup that facilitates maintainability and portability, most 717 of which is not reflected in the ChangeLog. This is the first release to 718 incorporate substantial contributions from numerous other developers, and the 719 result is a more broadly useful allocator (see the git revision history for 720 contribution details). Note that the license has been unified, thanks to 721 Facebook granting a license under the same terms as the other copyright 722 holders (see COPYING). 723 724 New features: 725 - Implement Valgrind support, redzones, and quarantine. 726 - Add support for additional platforms: 727 + FreeBSD 728 + Mac OS X Lion 729 + MinGW 730 + Windows (no support yet for replacing the system malloc) 731 - Add support for additional architectures: 732 + MIPS 733 + SH4 734 + Tilera 735 - Add support for cross compiling. 736 - Add nallocm(), which rounds a request size up to the nearest size class 737 without actually allocating. 738 - Implement aligned_alloc() (blame C11). 739 - Add the "thread.tcache.enabled" mallctl. 740 - Add the "opt.prof_final" mallctl. 741 - Update pprof (from gperftools 2.0). 742 - Add the --with-mangling option. 743 - Add the --disable-experimental option. 744 - Add the --disable-munmap option, and make it the default on Linux. 745 - Add the --enable-mremap option, which disables use of mremap(2) by default. 746 747 Incompatible changes: 748 - Enable stats by default. 749 - Enable fill by default. 750 - Disable lazy locking by default. 751 - Rename the "tcache.flush" mallctl to "thread.tcache.flush". 752 - Rename the "arenas.pagesize" mallctl to "arenas.page". 753 - Change the "opt.lg_prof_sample" default from 0 to 19 (1 B to 512 KiB). 754 - Change the "opt.prof_accum" default from true to false. 755 756 Removed features: 757 - Remove the swap feature, including the "config.swap", "swap.avail", 758 "swap.prezeroed", "swap.nfds", and "swap.fds" mallctls. 759 - Remove highruns statistics, including the 760 "stats.arenas.<i>.bins.<j>.highruns" and 761 "stats.arenas.<i>.lruns.<j>.highruns" mallctls. 762 - As part of small size class refactoring, remove the "opt.lg_[qc]space_max", 763 "arenas.cacheline", "arenas.subpage", "arenas.[tqcs]space_{min,max}", and 764 "arenas.[tqcs]bins" mallctls. 765 - Remove the "arenas.chunksize" mallctl. 766 - Remove the "opt.lg_prof_tcmax" option. 767 - Remove the "opt.lg_prof_bt_max" option. 768 - Remove the "opt.lg_tcache_gc_sweep" option. 769 - Remove the --disable-tiny option, including the "config.tiny" mallctl. 770 - Remove the --enable-dynamic-page-shift configure option. 771 - Remove the --enable-sysv configure option. 772 773 Bug fixes: 774 - Fix a statistics-related bug in the "thread.arena" mallctl that could cause 775 invalid statistics and crashes. 776 - Work around TLS deallocation via free() on Linux. This bug could cause 777 write-after-free memory corruption. 778 - Fix a potential deadlock that could occur during interval- and 779 growth-triggered heap profile dumps. 780 - Fix large calloc() zeroing bugs due to dropping chunk map unzeroed flags. 781 - Fix chunk_alloc_dss() to stop claiming memory is zeroed. This bug could 782 cause memory corruption and crashes with --enable-dss specified. 783 - Fix fork-related bugs that could cause deadlock in children between fork 784 and exec. 785 - Fix malloc_stats_print() to honor 'b' and 'l' in the opts parameter. 786 - Fix realloc(p, 0) to act like free(p). 787 - Do not enforce minimum alignment in memalign(). 788 - Check for NULL pointer in malloc_usable_size(). 789 - Fix an off-by-one heap profile statistics bug that could be observed in 790 interval- and growth-triggered heap profiles. 791 - Fix the "epoch" mallctl to update cached stats even if the passed in epoch 792 is 0. 793 - Fix bin->runcur management to fix a layout policy bug. This bug did not 794 affect correctness. 795 - Fix a bug in choose_arena_hard() that potentially caused more arenas to be 796 initialized than necessary. 797 - Add missing "opt.lg_tcache_max" mallctl implementation. 798 - Use glibc allocator hooks to make mixed allocator usage less likely. 799 - Fix build issues for --disable-tcache. 800 - Don't mangle pthread_create() when --with-private-namespace is specified. 801 802* 2.2.5 (November 14, 2011) 803 804 Bug fixes: 805 - Fix huge_ralloc() race when using mremap(2). This is a serious bug that 806 could cause memory corruption and/or crashes. 807 - Fix huge_ralloc() to maintain chunk statistics. 808 - Fix malloc_stats_print(..., "a") output. 809 810* 2.2.4 (November 5, 2011) 811 812 Bug fixes: 813 - Initialize arenas_tsd before using it. This bug existed for 2.2.[0-3], as 814 well as for --disable-tls builds in earlier releases. 815 - Do not assume a 4 KiB page size in test/rallocm.c. 816 817* 2.2.3 (August 31, 2011) 818 819 This version fixes numerous bugs related to heap profiling. 820 821 Bug fixes: 822 - Fix a prof-related race condition. This bug could cause memory corruption, 823 but only occurred in non-default configurations (prof_accum:false). 824 - Fix off-by-one backtracing issues (make sure that prof_alloc_prep() is 825 excluded from backtraces). 826 - Fix a prof-related bug in realloc() (only triggered by OOM errors). 827 - Fix prof-related bugs in allocm() and rallocm(). 828 - Fix prof_tdata_cleanup() for --disable-tls builds. 829 - Fix a relative include path, to fix objdir builds. 830 831* 2.2.2 (July 30, 2011) 832 833 Bug fixes: 834 - Fix a build error for --disable-tcache. 835 - Fix assertions in arena_purge() (for real this time). 836 - Add the --with-private-namespace option. This is a workaround for symbol 837 conflicts that can inadvertently arise when using static libraries. 838 839* 2.2.1 (March 30, 2011) 840 841 Bug fixes: 842 - Implement atomic operations for x86/x64. This fixes compilation failures 843 for versions of gcc that are still in wide use. 844 - Fix an assertion in arena_purge(). 845 846* 2.2.0 (March 22, 2011) 847 848 This version incorporates several improvements to algorithms and data 849 structures that tend to reduce fragmentation and increase speed. 850 851 New features: 852 - Add the "stats.cactive" mallctl. 853 - Update pprof (from google-perftools 1.7). 854 - Improve backtracing-related configuration logic, and add the 855 --disable-prof-libgcc option. 856 857 Bug fixes: 858 - Change default symbol visibility from "internal", to "hidden", which 859 decreases the overhead of library-internal function calls. 860 - Fix symbol visibility so that it is also set on OS X. 861 - Fix a build dependency regression caused by the introduction of the .pic.o 862 suffix for PIC object files. 863 - Add missing checks for mutex initialization failures. 864 - Don't use libgcc-based backtracing except on x64, where it is known to work. 865 - Fix deadlocks on OS X that were due to memory allocation in 866 pthread_mutex_lock(). 867 - Heap profiling-specific fixes: 868 + Fix memory corruption due to integer overflow in small region index 869 computation, when using a small enough sample interval that profiling 870 context pointers are stored in small run headers. 871 + Fix a bootstrap ordering bug that only occurred with TLS disabled. 872 + Fix a rallocm() rsize bug. 873 + Fix error detection bugs for aligned memory allocation. 874 875* 2.1.3 (March 14, 2011) 876 877 Bug fixes: 878 - Fix a cpp logic regression (due to the "thread.{de,}allocatedp" mallctl fix 879 for OS X in 2.1.2). 880 - Fix a "thread.arena" mallctl bug. 881 - Fix a thread cache stats merging bug. 882 883* 2.1.2 (March 2, 2011) 884 885 Bug fixes: 886 - Fix "thread.{de,}allocatedp" mallctl for OS X. 887 - Add missing jemalloc.a to build system. 888 889* 2.1.1 (January 31, 2011) 890 891 Bug fixes: 892 - Fix aligned huge reallocation (affected allocm()). 893 - Fix the ALLOCM_LG_ALIGN macro definition. 894 - Fix a heap dumping deadlock. 895 - Fix a "thread.arena" mallctl bug. 896 897* 2.1.0 (December 3, 2010) 898 899 This version incorporates some optimizations that can't quite be considered 900 bug fixes. 901 902 New features: 903 - Use Linux's mremap(2) for huge object reallocation when possible. 904 - Avoid locking in mallctl*() when possible. 905 - Add the "thread.[de]allocatedp" mallctl's. 906 - Convert the manual page source from roff to DocBook, and generate both roff 907 and HTML manuals. 908 909 Bug fixes: 910 - Fix a crash due to incorrect bootstrap ordering. This only impacted 911 --enable-debug --enable-dss configurations. 912 - Fix a minor statistics bug for mallctl("swap.avail", ...). 913 914* 2.0.1 (October 29, 2010) 915 916 Bug fixes: 917 - Fix a race condition in heap profiling that could cause undefined behavior 918 if "opt.prof_accum" were disabled. 919 - Add missing mutex unlocks for some OOM error paths in the heap profiling 920 code. 921 - Fix a compilation error for non-C99 builds. 922 923* 2.0.0 (October 24, 2010) 924 925 This version focuses on the experimental *allocm() API, and on improved 926 run-time configuration/introspection. Nonetheless, numerous performance 927 improvements are also included. 928 929 New features: 930 - Implement the experimental {,r,s,d}allocm() API, which provides a superset 931 of the functionality available via malloc(), calloc(), posix_memalign(), 932 realloc(), malloc_usable_size(), and free(). These functions can be used to 933 allocate/reallocate aligned zeroed memory, ask for optional extra memory 934 during reallocation, prevent object movement during reallocation, etc. 935 - Replace JEMALLOC_OPTIONS/JEMALLOC_PROF_PREFIX with MALLOC_CONF, which is 936 more human-readable, and more flexible. For example: 937 JEMALLOC_OPTIONS=AJP 938 is now: 939 MALLOC_CONF=abort:true,fill:true,stats_print:true 940 - Port to Apple OS X. Sponsored by Mozilla. 941 - Make it possible for the application to control thread-->arena mappings via 942 the "thread.arena" mallctl. 943 - Add compile-time support for all TLS-related functionality via pthreads TSD. 944 This is mainly of interest for OS X, which does not support TLS, but has a 945 TSD implementation with similar performance. 946 - Override memalign() and valloc() if they are provided by the system. 947 - Add the "arenas.purge" mallctl, which can be used to synchronously purge all 948 dirty unused pages. 949 - Make cumulative heap profiling data optional, so that it is possible to 950 limit the amount of memory consumed by heap profiling data structures. 951 - Add per thread allocation counters that can be accessed via the 952 "thread.allocated" and "thread.deallocated" mallctls. 953 954 Incompatible changes: 955 - Remove JEMALLOC_OPTIONS and malloc_options (see MALLOC_CONF above). 956 - Increase default backtrace depth from 4 to 128 for heap profiling. 957 - Disable interval-based profile dumps by default. 958 959 Bug fixes: 960 - Remove bad assertions in fork handler functions. These assertions could 961 cause aborts for some combinations of configure settings. 962 - Fix strerror_r() usage to deal with non-standard semantics in GNU libc. 963 - Fix leak context reporting. This bug tended to cause the number of contexts 964 to be underreported (though the reported number of objects and bytes were 965 correct). 966 - Fix a realloc() bug for large in-place growing reallocation. This bug could 967 cause memory corruption, but it was hard to trigger. 968 - Fix an allocation bug for small allocations that could be triggered if 969 multiple threads raced to create a new run of backing pages. 970 - Enhance the heap profiler to trigger samples based on usable size, rather 971 than request size. 972 - Fix a heap profiling bug due to sometimes losing track of requested object 973 size for sampled objects. 974 975* 1.0.3 (August 12, 2010) 976 977 Bug fixes: 978 - Fix the libunwind-based implementation of stack backtracing (used for heap 979 profiling). This bug could cause zero-length backtraces to be reported. 980 - Add a missing mutex unlock in library initialization code. If multiple 981 threads raced to initialize malloc, some of them could end up permanently 982 blocked. 983 984* 1.0.2 (May 11, 2010) 985 986 Bug fixes: 987 - Fix junk filling of large objects, which could cause memory corruption. 988 - Add MAP_NORESERVE support for chunk mapping, because otherwise virtual 989 memory limits could cause swap file configuration to fail. Contributed by 990 Jordan DeLong. 991 992* 1.0.1 (April 14, 2010) 993 994 Bug fixes: 995 - Fix compilation when --enable-fill is specified. 996 - Fix threads-related profiling bugs that affected accuracy and caused memory 997 to be leaked during thread exit. 998 - Fix dirty page purging race conditions that could cause crashes. 999 - Fix crash in tcache flushing code during thread destruction. 1000 1001* 1.0.0 (April 11, 2010) 1002 1003 This release focuses on speed and run-time introspection. Numerous 1004 algorithmic improvements make this release substantially faster than its 1005 predecessors. 1006 1007 New features: 1008 - Implement autoconf-based configuration system. 1009 - Add mallctl*(), for the purposes of introspection and run-time 1010 configuration. 1011 - Make it possible for the application to manually flush a thread's cache, via 1012 the "tcache.flush" mallctl. 1013 - Base maximum dirty page count on proportion of active memory. 1014 - Compute various additional run-time statistics, including per size class 1015 statistics for large objects. 1016 - Expose malloc_stats_print(), which can be called repeatedly by the 1017 application. 1018 - Simplify the malloc_message() signature to only take one string argument, 1019 and incorporate an opaque data pointer argument for use by the application 1020 in combination with malloc_stats_print(). 1021 - Add support for allocation backed by one or more swap files, and allow the 1022 application to disable over-commit if swap files are in use. 1023 - Implement allocation profiling and leak checking. 1024 1025 Removed features: 1026 - Remove the dynamic arena rebalancing code, since thread-specific caching 1027 reduces its utility. 1028 1029 Bug fixes: 1030 - Modify chunk allocation to work when address space layout randomization 1031 (ASLR) is in use. 1032 - Fix thread cleanup bugs related to TLS destruction. 1033 - Handle 0-size allocation requests in posix_memalign(). 1034 - Fix a chunk leak. The leaked chunks were never touched, so this impacted 1035 virtual memory usage, but not physical memory usage. 1036 1037* linux_2008082[78]a (August 27/28, 2008) 1038 1039 These snapshot releases are the simple result of incorporating Linux-specific 1040 support into the FreeBSD malloc sources. 1041 1042-------------------------------------------------------------------------------- 1043vim:filetype=text:textwidth=80 1044