summaryrefslogtreecommitdiff
path: root/ChangeLog
diff options
context:
space:
mode:
authorIvan Maidanski <ivmai@mail.ru>2022-04-25 10:00:32 +0300
committerIvan Maidanski <ivmai@mail.ru>2022-04-25 12:01:15 +0300
commit40f1ed85b3f13d2a28dbf51f24a6323e738d6a8b (patch)
tree8ec8a367682fd0b14d1d0bde57ec759d115ffaaf /ChangeLog
parentc3b58c0f45a886dc5f6fbc54591c40cd2753950b (diff)
downloadbdwgc-40f1ed85b3f13d2a28dbf51f24a6323e738d6a8b.tar.gz
Update ChangeLog file
Diffstat (limited to 'ChangeLog')
-rw-r--r--ChangeLog22
1 files changed, 22 insertions, 0 deletions
diff --git a/ChangeLog b/ChangeLog
index 18101d33..d76b8c7d 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -7,15 +7,19 @@
* Add check-deps goal to Makefiles to build the tests but not execute them
* Add missing undef GC_NO_THREAD_REDIRECTS in threaded tests
* Adjust comments in block_unmap_inner
+* Adjust indentation in C preprocessor directives (reformatting)
* Adjust printing on amount of unmapped memory after GC
* Allow disable USE_PROC_FOR_LIBRARIES even if malloc is redirected on Linux
* Allow mprotect-based incremental GC on AIX and for all Linux targets
* Allow testing of retry-signals functionality
* Allow to build with forced WRAP_MARK_SOME
+* Allow to compile tests with custom NTHREADS set to zero
* Allow to skip heap expansion in GC_init
* Allow to unmap memory block right in GC cycle where block is freed
+* Allow to use same signal for thread suspend and restart (default on E2K)
* Always set only one bit past end in GC_set_hdr_marks
* Assert SET_HDR does not write to GC_all_nils
+* Assert that disappearing links are accessible during registration and GC
* Avoid unmap of memory allocated during GC init regardless of threshold
* Better document GC_generate_random_heap/valid_address API functions
* Build extra/gc.c in digimars.mak
@@ -24,26 +28,33 @@
* Check pointer tag in all mark procedures (E2K)
* Declare GC_register_stackbottom symbol as internal on IA64 (refactoring)
* Define GC_save_regs_ret_val variable only if used on IA64 (refactoring)
+* Define and use SIGNAL_BASED_STOP_WORLD macro internally (refactoring)
* Define marker_[b]sp only if needed (refactoring)
* Disable H/W tag checking in mark procedures by default (E2K)
* Disable memory unmapping explicitly in gcconfig.h for Emscripten
* Discard sections with pseudo-paths except heap in register_map_entries
* Do not compile pthread_start.c on Cygwin even if build shared libs (CMake)
* Do not compile pthread_stop_world.c on Darwin (CMake)
+* Do not define THREAD_SANITIZER if GC is built w/o threads support
* Do not put atomic_ops[_sysdeps].o to libgc.a in Makefile.direct
* Do not treat warnings as errors by default in digimars.mak
+* Document GC_ENABLE_SUSPEND_THREAD better in configure
+* Document GC_with_callee_saves_pushed better
* Document argument of GC_is_marked in gc_mark.h
* Eliminate 'checking if 1U<<12 is less than zero' cppcheck FP warning
* Eliminate 'n obtained from untrusted source' code defect FP in test_cpp
* Enable prefetch loop in GC_mark_from on E2K
* Enable without_libatomic_ops for targets requiring atomic_ops.c (CMake)
+* Ensure GC lock is held in add_leaked/smashed, check_heap_proc (refactoring)
* Ensure GC_n_set_marks() does not count extra bits in hb_marks
* Ensure a collection occurs between adjacent heap expansions
* Ensure every test prints a message on success
* Fail threadkeytest on pthread_create error in main loop
* Fix digimars.mak to build 'all' by default
+* Force pthread_mutex_lock usage for GC lock if ThreadSanitizer
* Group NEED_FIND_LIMIT definitions in gcconfig.h (refactoring)
* Increase number of iterations in leak and middle tests
+* Initialize GC_retry_signals to true in GC_stop_init (refactoring)
* Make abort message unique in write_fault_handler
* Make all functions static in msvc_dbg.c except for backtrace[_symbols]
* Mention CMake-based build in README.md and README.QUICK
@@ -52,12 +63,14 @@
* Move public header files to include/gc in source tree
* Name all tests consistently (refactoring)
* New CMake option to link with external libatomic_ops (system library)
+* Output finalization information by GC_dump
* Pass ATOMIC_OPS_LIBS to tests using AO primitives directly (Automake)
* Prevent requesting blocks past max_heapsize in GC_collect_or_expand
* Print in-use heap statistic on heap growth
* Print iteration number on thread creation or join failure in tests
* Print objs_in_block as zero for large objects in GC_print_block_list
* Print thread number on marker thread creation failure
+* Print value of links in GC_dump_finalization_links
* Put gc_badalc/cpp.cpp instead of .cc ones in CMake script (MSVC/BCC/WCC)
* Really ignore bad GC_INITIAL/MAXIMUM_HEAP_SIZE values
* Reduce number of iterations in disclaim and cpp tests
@@ -66,6 +79,7 @@
* Relax required dependencies for atomicopstest (CMake)
* Remove GC_copyright symbol
* Remove GC_min/max() and simplify heap bounds update in expand_hp_inner
+* Remove GC_world_is_stopped variable (refactoring)
* Remove IF_IA64 macro in pthread_stop_world (refactoring)
* Remove STACK_GROWS_DOWN explicit definitions in gcconfig.h (refactoring)
* Remove Symbian makefile
@@ -77,6 +91,7 @@
* Remove obsolete AC_PROG_LIBTOOL in configure.ac (refactoring)
* Remove redundant 'private' prefix in include directives in include/private
* Remove specific DATASTART definition for ancient OpenBSD/m68k
+* Remove thread_blocked field in GC_thread (refactoring)
* Rename MAP_LEN internal macro to OBJ_MAP_LEN (refactoring)
* Rename README.QUICK to LICENSE and install it by default
* Rename README.rs6000 to README.aix
@@ -87,11 +102,18 @@
* Roundup size passed to GC_expand_hp
* State explicitly that license is MIT-style
* Support CFLAGS_EXTRA in rare build scripts (NT/WCC_MAKEFILE, digimars.mak)
+* Support ThreadSanitizer (GCC)
* Support malloc redirection on E2K
+* Test manual self-suspend in gctest
* Treat GC_arrays as a part of internal memory when printing its size
* Try find libatomic_ops installed by cmake if with_libatomic_ops (CMake)
+* Update next_random_no state using AO primitive
+* Use AO primitive in GC_noop1 instead of no_sanitize attribute
+* Use AO primitives in GC_RAND_NEXT instead of no_sanitize attribute
* Use GC_printf in cpptest consistently (refactoring)
+* Use single-argument suspend_handler for E2K, HP_PA, IA64, M68K
* Warn if heap has grown while GC was disabled
+* Workaround 'data access w/o lock' code defect FP in GC_incr_bytes_freed
== [8.2.2] (not released yet) ==