Fix deadlocks in write and suspend handlers if AO test-and-set is emulated

This could be tested with -D AO_USE_PTHREAD_DEFS passed to CFLAGS.

* configure.ac (AO_TRYLINK_CFLAGS): New variable.
* configure.ac [$with_libatomic_ops!=none && $need_atomic_ops_asm!=true]
(BASE_ATOMIC_OPS_EMULATED): New AC_DEFINE (defined in case of failure
of AC_TRY_LINK of a code snippet calling AO_test_and_set_acquire,
AO_CLEAR, AO_compiler_barrier, AO_store, AO_load, AO_char_store,
AO_char_load, AO_store_release, AO_load_acquire); use AO_TRYLINK_CFLAGS;
add comment.
* include/private/gcconfig.h [BASE_ATOMIC_OPS_EMULATED] (MPROTECT_VDB):
Undefine.
* mark.c [AO_CLEAR] (GC_noop6): Do not call AO_compiler_barrier() if
BASE_ATOMIC_OPS_EMULATED.
* misc.c [!GC_DISABLE_INCREMENTAL] (GC_init, GC_enable_incremental):
Do not set GC_manual_vdb if BASE_ATOMIC_OPS_EMULATED.
* pthread_stop_world.c [!GC_OPENBSD_UTHREADS && !NACL]
(ao_load_acquire_async, ao_load_async, ao_store_release_async,
ao_store_async): New macro; undefine it after the usage.
* pthread_stop_world.c [!GC_OPENBSD_UTHREADS && !NACL]
(GC_store_stack_ptr): Use ao_store_async() instead of AO_store().
* pthread_stop_world.c [!GC_OPENBSD_UTHREADS && !NACL]
(GC_suspend_handler_inner): Use ao_load[_acquire]_async() and
ao_store_release_async() instead of AO_load[_acquire]() and
AO_store_release(), respectively.
* pthread_stop_world.c [!GC_OPENBSD_UTHREADS && !NACL
&& GC_ENABLE_SUSPEND_THREAD] (suspend_self_inner): Use
ao_load_acquire_async() instead of AO_load_acquire().

Use macro to operate on a flag residing in GC_stop_count
(code refactoring of commit 295a2f24e)

* pthread_stop_world.c [!GC_OPENBSD_UTHREADS && !NACL]
(THREAD_RESTARTED): New macro.
* pthread_stop_world.c [!GC_OPENBSD_UTHREADS && !NACL] (GC_stop_count):
Refine comment.
* pthread_stop_world.c [!GC_OPENBSD_UTHREADS && !NACL]
(GC_suspend_handler_inner, GC_stop_world, GC_restart_all): Use
THREAD_RESTARTED.

Fix tests linkage with internal atomic_ops.o

* tests/tests.am (test_ldadd): Add $(nodist_libgc_la_OBJECTS).

Update AUTHORS file (add Fabrice Fontaine)

Fix linkage with a system libatomic_ops shared library

Issue #247 (bdwgc).

When bdwgc is linked with the external libatomic_ops, bdw-gc.pc must
contain the needed dynamic libraries (such as -latomic_ops) otherwise
build of applications could fail on the link stage on some hosts:
* libgc.so: undefined reference to 'AO_fetch_compare_and_swap_emulation'
* libgc.so: undefined reference to 'AO_store_full_emulation'

So, this commit sets ATOMIC_OPS_LIBS to "-latomic_ops" when a system
atomic_ops library is used and uses ATOMIC_OPS_LIBS in bdw-gc.pc.in.

* bdw-gc.pc.in (Libs): Add @ATOMIC_OPS_LIBS@.
* configure.ac [$with_libatomic_ops!=no && $with_libatomic_ops!=none
&& $THREADS!=none] (ATOMIC_OPS_LIBS): Set to -latomic_ops; do AC_SUBST.

Fix test_cpp assertion violation in find-leak mode

* tests/test_cpp.cc (C.Test, D.Test, F.Test): Do not fail on
nFreed>=0.8*nAllocated[F] assertion violation if GC_get_find_leak() returns
non-zero; replace .8*nAllocated[F] to (nAllocated[F]/5)*4 to avoid
floating-point operations.

Guard against potential buffer overflow in CORD_next and CORD_pos_fetch

* cord/cordbscs.c (CORD__pos_fetch, CORD__next): Call ABORT() if
CORD_pos_valid() returns false for the argument (do not compute pe
in this case).

Fix 'collecting from unknown thread' abort in leak-finding mode for Win32
(fix of commit 8e6460052)

* include/private/gc_priv.h [THREADS] (GC_in_thread_creation): Declare
not only for pthreads.
* misc.c [!DONT_USE_ATEXIT && THREADS] (GC_exit_check): Set
GC_in_thread_creation to true before calling GC_gcollect for windows
threads as well.
* win32_threads.c (GC_in_thread_creation): Change STATIC to GC_INNER.

Fix 'too wide non-owner permissions are set for resource' code defect

* misc.c [(UNIX_LIKE && !GC_ANDROID_LOG || CYGWIN32 || SYMBIAN)
&& !SMALL_CONFIG] (GC_init): Pass 0644 permissions mode instead of 0666
to open (create) a log file.
* os_dep.c [MMAP_SUPPORTED && !USE_MMAP_ANON && SYMBIAN]
(GC_unix_mmap_get_mem): Pass 0644 mode instead of 0666 to open "zero"
file.

Fix hbp overflow in GC_install_counts
(a cherry-pick of commits 6a32bb4, 36a6d79 from 'master')

Issue #245 (bdwgc).

The overflow resulted in an infinite loop in GC_install_counts on Win32.

* headers.c (GC_install_counts): If hbp+=BOTTOM_SZ overflow is expected
then break the first loop; add comment.

Fix marks and hb_n_marks consistency when disclaim returns true

* reclaim.c [ENABLE_DISCLAIM] (GC_disclaim_and_reclaim): When a
disclaim callback returns 1 to protect an object from being reclaimed,
mark it to skip it on repeated scans within the cycle. In particular,
this fixes sz*hhdr->hb_n_marks<=HBLKSIZE assertion failure due to
excessive increments of hb_n_marks.

Fix lock assertion violation in get_index if GC_ALWAYS_MULTITHREADED
(fix of commit d5c65315b)

* misc.c [GC_ASSERTIONS && GC_ALWAYS_MULTITHREADED] (GC_init): Move up
LOCK() call to precede GC_expand_hp_inner() call.

Fix 'set_pht_entry_from_index_concurrent value is unknown' cppcheck warning
(fix of commit cdc201f40)

* blacklst.c [PARALLEL_MARK && THREAD_SANITIZER]
(backlist_set_pht_entry_from_index): Assume
set_pht_entry_from_index_concurrent is always defined.
* include/private/gc_priv.h [!THREADS || !AO_HAVE_or]
(set_pht_entry_from_index_concurrent): Define (to
set_pht_entry_from_index).
* include/private/gc_priv.h [THREADS && AO_HAVE_or]
(GC_acquire_dirty_lock, GC_release_dirty_lock): Define to empty.
* os_dep.c [!GC_DISABLE_INCREMENTAL && (!THREADS || AO_HAVE_or)]
(async_set_pht_entry_from_index): Define to
set_pht_entry_from_index_concurrent.

Update AUTHORS file

Update documentation about arm64 ABI in gcconfig.h
(fix of commit b399b13b9)

Issue #242 (bdwgc).

This minor patch makes the info on available arm64/aarch64 ABIs more
consistent.

* include/private/gcconfig.h (mach_type_known): Refine comment about
AARCH64.

Add paths to filenames mentioned in the copyright section in README

* README.md (Copyright & Warranty): Add the path to each mentioned file.

.gitignore: Ignore disclaim_weakmap_test executable file

Update ChangeLog file (v8.0 changes)

Update ChangeLog file (v7.6 changes)
(a cherry-pick of commit 16cbc85d from 'release-7_6')

Update ChangeLog file (v7.2 - v7.4 changes only)
(a cherry-pick of commit 4bcd0c33 from 'release-7_4')

Fix memory allocation on GCF (Linux/x64)

Issue #225 (bdwgc).

GCF (Google Cloud Function) seems to have mmap("/dev/zero") operation
unsupported.  The workaround is to use mmap(MAP_ANONYMOUS).

* include/private/gcconfig.h [X86_64 && LINUX && __GLIBC__
&& !__UCLIBC__] (USE_MMAP_ANON): Define macro; add comment.

Fix 'USE_MUNMAP macro redefinition' error for NaCl
(fix of commit 1aabce0f9)

* include/private/gcconfig.h [I386 && NACL] (USE_MUNMAP): Remove macro
(as it is defined in configure/CMake by default since gc v8.0.0).

Remove extra USE_MMAP definition for Interix
(code refactoring)

* include/private/gcconfig.h [I386 && INTERIX && !USE_MMAP] (USE_MMAP):
Remove macro definition (because in gc v8.0.0+ it is defined later in this
file).

Call real pthread_sigmask instead of its wrapper in start_mark_threads
(fix of commit 95947bdc5)

The wrapper itself does not do anything useful in this case.

* pthread_support.c [PARALLEL_MARK && !NO_MARKER_SPECIAL_SIGMASK]
(GC_start_mark_threads_inner): Call REAL_FUNC(pthread_sigmask) instead
of pthread_sigmask.

Use AO_or in async_set_pht_entry_from_index if available

* blacklst.c (backlist_set_pht_entry_from_index): New macro; move
comment from gc_priv.h.
* blacklst.c (GC_add_to_black_list_normal, GC_add_to_black_list_stack):
Use backlist_set_pht_entry_from_index() instead of
set_pht_entry_from_index_concurrent().
* include/private/gc_priv.h (set_pht_entry_from_index): Add outermost
parentheses; cast result to void.
* include/private/gc_priv.h [THREADS]
(set_pht_entry_from_index_concurrent): Define only if AO_HAVE_or;
update comment.
* include/private/gc_priv.h [THREADS] (GC_acquire_dirty_lock,
GC_release_dirty_lock): Define to no-op if
set_pht_entry_from_index_concurrent is defined.
* include/private/gc_priv.h [THREADS
&& set_pht_entry_from_index_concurrent] (GC_fault_handler_lock): Do not
declare.
* os_dep.c [!GC_DISABLE_INCREMENTAL && THREADS
&& set_pht_entry_from_index_concurrent]
(async_set_pht_entry_from_index): Redirect to
set_pht_entry_from_index_concurrent; do not define
GC_fault_handler_lock variable.
* os_dep.c [!GC_DISABLE_INCREMENTAL && THREADS
&& !set_pht_entry_from_index_concurrent && AO_HAVE_test_and_set_acquire]
(async_set_pht_entry_from_index): Remove GC_ATTR_NO_SANITIZE_THREAD
(because AO_HAVE_or should be defined in case of TSan).

Fix concurrent bitmap update in GC_dirty
(a cherry-pick of commits 7305e5619, 0c0e4cd0b, d6db3f005 from 'master')

Issue #235 (bdwgc).

* darwin_stop_world.c [!GC_NO_THREADS_DISCOVERY]
(GC_suspend_thread_list): Call GC_acquire_dirty_lock and
GC_release_dirty_lock around thread_suspend().
* darwin_stop_world.c (GC_stop_world): Likewise.
* include/private/gc_priv.h (clear_pht_entry_from_index,
set_pht_entry_from_index_safe): Remove unused macro.
* include/private/gc_priv.h [THREADS] (GC_acquire_dirty_lock,
GC_release_dirty_lock): Define new macro; move comment from
win32_threads.c.
* include/private/gc_priv.h [THREADS && !GC_DISABLE_INCREMENTAL]
(GC_fault_handler_lock): Declare global variable (even if not
MPROTECT_VDB).
* os_dep.c [!GC_DISABLE_INCREMENTAL] (async_set_pht_entry_from_index):
Define even if not MPROTECT_VDB; reformat comments.
* os_dep.c [!GC_DISABLE_INCREMENTAL && AO_HAVE_test_and_set_acquire]
(GC_fault_handler_lock): Likewise.
* os_dep.c [THREADS && AO_HAVE_test_and_set_acquire]
(async_set_pht_entry_from_index): Use GC_acquire_dirty_lock and
GC_release_dirty_lock; remove wrong comment.
* os_dep.c [THREADS && !AO_HAVE_test_and_set_acquire]
(currently_updating, async_set_pht_entry_from_index): Remove incorrect
implementation.
* os_dep.c [!GC_DISABLE_INCREMENTAL] (GC_dirty_inner): Call
async_set_pht_entry_from_index instead of set_pht_entry_from_index;
remove FIXME.
* pthread_stop_world.c [GC_ENABLE_SUSPEND_THREAD] (GC_suspend_thread):
If GC_manual_vdb then call GC_acquire_dirty_lock and
GC_release_dirty_lock around a block of RAISE_SIGNAL() and sem_wait().
* pthread_stop_world.c [GC_OPENBSD_UTHREADS] (GC_suspend_all): Call
GC_acquire_dirty_lock and GC_release_dirty_lock around
pthread_suspend_np().
* pthread_stop_world.c [NACL] (GC_suspend_all): If GC_manual_vdb then
call GC_acquire_dirty_lock() and GC_release_dirty_lock() around the
code which ensures parking of threads.
* pthread_stop_world.c [!GC_OPENBSD_UTHREADS] (GC_suspend_all): Add
comment.
* pthread_stop_world.c [!GC_OPENBSD_UTHREADS && !NACL] (GC_stop_world):
If GC_manual_vdb then call GC_acquire_dirty_lock and
GC_release_dirty_lock around a block of GC_suspend_all() and
suspend_restart_barrier(); add comment.
* pthread_support.c [CAN_HANDLE_FORK] (fork_prepare_proc): Call
GC_acquire_dirty_lock().
* pthread_support.c [CAN_HANDLE_FORK] (fork_parent_proc,
fork_child_proc): Call GC_release_dirty_lock().
* win32_threads.c (GC_suspend): Use GC_acquire_dirty_lock and
GC_release_dirty_lock (regardless of MPROTECT_VDB).

Default to a single-threaded build for Nintendo, Orbis, Sony PSP targets
(fix of commits ce1f87461, 97182297e, d5dad58d3, cd63cf1e8)

* include/gc_config_macros.h [!GC_THREADS] (GC_THREADS): Do not check
SN_TARGET_ORBIS and SN_TARGET_PSP2 macros.
* include/private/gcconfig.h [X86_64 && SN_TARGET_ORBIS]: Do not
include pthread.h (should be included from gc_locks.h).
* include/private/gcconfig.h [NN_PLATFORM_CTR || NINTENDO_SWITCH
|| SN_TARGET_ORBIS || SN_TARGET_PS3 || SN_TARGET_PSP2] (THREADS):
Define only if GC_THREADS is defined.

Test marking of finalizer closure object in disclaim_test

* tests/disclaim_test.c (PTR_HASH): New macro.
* tests/disclaim_test.c (pair_dct): Change assertion condition about cd
(it should be equal to the hash value of p).
* tests/disclaim_test.c (pair_new): Replace static fc variable to local
pfc one which is assigned a pointer to atomic object; set pfc->cd to
the hash value of p.

Fix unneeded end_stubborn_change/ptr_store_and_dirty in disclaim_test
(fix of commits b52c140, 263ffbb)

* tests/disclaim_test.c (pair_dct): Add my_assert that cd is null;
store null to p->car (instead of storing cd and calling
GC_end_stubborn_change/GC_reachable_here).

Fix marking of finalizer closure object
(a cherry-pick of commits 21312a008, ea9d6dd6a from 'master')

* fnlz_mlc.c [ENABLE_DISCLAIM]: Include dbg_mlc.h.
* fnlz_mlc.c [ENABLE_DISCLAIM] (GC_init_finalized_malloc): Call
GC_register_displacement_inner(FINALIZER_CLOSURE_FLAG) and
GC_register_displacement_inner(sizeof(oh)+FINALIZER_CLOSURE_FLAG); add
comment.
* fnlz_mlc.c [ENABLE_DISCLAIM] (GC_finalized_malloc): Call GC_dirty and
REACHABLE_AFTER_DIRTY after storing fclos.

Refine documentation in gc_disclaim.h
(a cherry-pick of commits 8ce6ef666, 4158701ae from 'master')

* include/gc_disclaim.h (GC_disclaim_proc, GC_register_disclaim_proc,
GC_finalized_malloc): Refine comment.
* include/gc_disclaim.h (GC_register_disclaim_proc,
GC_finalized_malloc): Refine comment about the protection of the
finalizer closure object from collection.

Specify register_disclaim_proc and finalized_malloc argument as non-null

Also, a check is added that the actual finalizer closure pointer does
not have FINALIZER_CLOSURE_FLAG set.

* fnlz_mlc.c (GC_register_disclaim_proc): Add assertion that proc is
non-null.
* fnlz_mlc.c (GC_finalized_malloc): Add assertions that fclos argument
is non-null and that fclos itself does not have FINALIZER_CLOSURE_FLAG
set.
* include/gc_disclaim.h (GC_register_disclaim_proc,
GC_finalized_malloc): Add GC_ATTR_NONNULL for the pointer argument.

Fix comment typo in CMakeLists.txt

* CMakeLists.txt (CMAKE_USE_PTHREADS_INIT): Fix typo ("lib and includes")
in comment.

Fix GC_VSNPRINTF in cordprnt for DJGPP and MS VC for WinCE
(fix of commit c10a1d900)

GC-internal DJGPP and MSWINCE macros are not available in cord.

* cord/cordprnt.c (GC_VSNPRINTF): Test __DJGPP__ instead of DJGPP macro.
* cord/cordprnt.c [_MSC_VER] (GC_VSNPRINTF): Test _WIN32_WCE instead
of MSWINCE macro.

Fix missing curses.h in cord/de when compiling manually (MS VC, MinGW)

* cord/tests/de.c [__MINGW32__ || __NT__ || _WIN32] (WIN32): Define.

Eliminate 'different const qualifiers' MS VC warnings in cordbscs
(fix of commit 2e7daad62)

* cord/cordbscs.c (CORD_cat_char_star, CORD_cat): Cast the first
argument of GC_PTR_STORE_AND_DIRTY() to void*.

AppVeyor CI: Tailor config for release-8_0

Use libatomic_ops-7.6.x.

Travis CI: Tailor config for release-8_0

Remove scanning by Coverity and Codecov, use libatomic_ops-7.6.x.

[8.0.0]

Bump gc version to 8.0.0 (experimental release)

* ChangeLog (8.0.0): Set release date.
* Makefile.am (LIBGC_VER_INFO): Increment revision (change version info
of libgc.so to 5:1:4).
* Makefile.am (LIBGCCPP_VER_INFO): Increment current and age (change
version info of libgccpp.so to 5:0:4).
* README.md: Bump major version (set minor and micro to 0).
* configure.ac (AC_INIT): Likewise.
* include/gc_version.h (GC_TMP_VERSION_MAJOR, GC_TMP_VERSION_MINOR,
GC_TMP_VERSION_MICRO): Likewise.
* README.md: Remove build status badges (Travis, AppVeyor, Codecov,
Coveralls, Coverity, LGTM).
* cord/cord.am (LIBCORD_VER_INFO): Increment current and age (change
version info of libcord.so to 5:0:4).

Update ChangeLog file

Really fix 'potential unsafe sign check of a bitwise operation' code defect
(fix of commit af00c4d)

* mark.c [!OS2] (GC_mark_from): Replace (signed_word)v>=0 to (v&SIGNB)==0
(anyway, the compiler generates the same code).

Add Codecov.io badge to README

Travis CI: Send code coverage reports to Codecov as well

Fix comment typos in backgraph.c, de.c, gcconfig.h

* ChangeLog (5.0alpha6): Fix typo ("DATASTART").
* include/private/gcconfig.h: Likewise.
* backgraph.c: Fix typo in comment ("If").
* cord/tests/de.c: Fix typo in comment ("Unbounded").

Replace some FIXME items with TODO ones
(code refactoring)

* alloc.c (GC_maybe_gc, GC_collect_a_little_inner,
GC_finish_collection): Replace FIXME item with TODO one.
* gcj_mlc.c (maybe_finalize): Likewise.
* include/gc.h [!GC_DEBUG || !__GNUC__] (GC_PTR_ADD): Likewise.
* include/private/gcconfig.h [X86_64 && LINUX && __GLIBC__
&& !__UCLIBC__] (GETCONTEXT_FPU_EXCMASK_BUG): Likewise.
* mach_dep.c (GC_with_callee_saves_pushed): Likewise.
* mark_rts.c (GC_push_current_stack): Likewise.
* misc.c [MSWIN32 || MSWINCE] (WRITE): Likewise.
* os_dep.c [ECOS] (ecos_gc_memory): Likewise.
* os_dep.c (GC_get_stack_base, GC_page_was_ever_dirty): Likewise.
* reclaim.c (GC_reclaim_clear): Likewise.
* include/private/gcconfig.h [OPENBSD && m68k]: Remove FIXME item.

Eliminate 'unused variable old_maps_size' compiler warning in get_maps
(fix of commit b3eee14)

* os_dep.c [NEED_PROC_MAPS] (GC_get_maps): Do not declare old_maps_size
local variable unless THREADS.

Remove MIN_WORDS macro from gc_priv.h
(code refactoring)

* include/private/gc_priv.h (MIN_WORDS): Move macro to test.c; remove
FIXME.
* tests/test.c: Update comment for include gc_priv.h.

Workaround 'bad pointer arithmetic' false waring in check_annotated_obj

* dbg_mlc.c [!SHORT_DBG_HDRS] (GC_check_annotated_obj): Change
(word*)ptr+ofs expressions to &((word*)ptr)[ofs].

Eliminate 'potential unsafe sign check of a bitwise operation' code defect

* mark.c [!OS2] (GC_mark_from): Cast mark_stack_top-mark_stack to word
first, and cast bitwise OR operation result to signed_word, so that not
to apply bitwise OR operation to the value of ptrdiff_t.

Eliminate 'comparison is always false' code defect in get_maps

In case of a single-threaded build, the comparison is always false
because maps_size>=0 and old_maps_size<=0.

* os_dep.c [NEED_PROC_MAPS] (GC_get_maps): Do not evaluate
maps_size<old_maps_size inside while expression if THREADS; remove
comment.

Eliminate 'poor global variable name' code defect in typd_mlc
(fix of commit a825a2d)

* typd_mlc.c (ld, ad, sd): Remove unused global variables.

Add LGTM.com code quality badges to README.md

Update AUTHORS file (add Demyan Kimitsa)

Fix start_world not resuming all threads on Darwin

Issue #231 (bdwgc).

This happens due mach thread ports are released in GC_stop_world as
a result in some cases the system creates new one and GC_start_world
will not find the thread to resume; in turn, the threads will stay
suspended.  When this happens to GDC threads, the application is
terminated as the message loop is halt.

The high-level description of the change: thread mach port is not
deallocated when thread is saved in GC_mach_threads as it causes thread
not being resumed as thread port object can be changed; threads are now
being suspended without check of suspend_count (as it can be suspended
by client); the logic of resume is changed, now the primary cycle
iterates threads to resume in GC_mach_threads.

* darwin_stop_world.c (struct GC_mach_thread): Rename already_suspended
filed to suspended.
* darwin_stop_world.c [!GC_NO_THREADS_DISCOVERY]
(GC_suspend_thread_list): Add my_task argument; remove info an outCount
local variables; call mach_port_deallocate uless threaded is added to
GC_mach_threads; do not call thread_info; update the relevant comments.
* darwin_stop_world.c [!GC_NO_THREADS_DISCOVERY] (GC_stop_world):
Remove i local variable; do not call mach_port_deallocate for prevlist
items; add comment.
* darwin_stop_world.c (GC_thread_resume): Call WARN instead of ABORT
if thread_resume failed.
* darwin_stop_world.c [!GC_NO_THREADS_DISCOVERY] (GC_start_world):
Initialize j to listcount; iterate over GC_mach_threads in the outer
loop; if thread is suspended then find it in act_list; if found then
resume thread; call mach_port_deallocate for act_list items.

Fix indentation of 'define GC_DATAEND' in gc.h
(fix of commit 4e24d219d)

* include/gc.h [__CYGWIN__ && !__x86_64__] (GC_DATAEND): Fix
indentation.

Travis CI: Fix missing semicolon at the end of GNU_DOWNLOAD_SITE assignment
(fix of commit a9690ac)

Update ChangeLog file (add v7.6.8 release date)

Travis CI: Put GNU S/W download base URL in a dedicated variable
(code refactoring)

Update ChangeLog file

Remove redundant checking of CPPCHECK macro in gc_cpp.h
(fix of commit 5e51e8d)

Issue #229 (bdwgc).

* include/gc_cpp.h [(_MSC_VER || __DMC__ || (__CYGWIN32__ || __CYGWIN__
|| __MINGW32__) && !GC_BUILD && !GC_NOT_DLL) && __cplusplus>201103L
&& GC_OPERATOR_NEW_ARRAY] (operator delete[](void*,size_t)): Define
even in case of CPPCHECK.

Fix delete operator redirection if gc_cpp is built as .dll (Cygwin, MinGW)

Issue #229 (bdwgc).

* include/gc_cpp.h [(__CYGWIN32__ || __CYGWIN__ || __MINGW32__)
&& !GC_BUILD && !GC_NOT_DLL && GC_OPERATOR_NEW_ARRAY] (operator new[],
operator delete[]): Define inline function.
* include/gc_cpp.h [(__CYGWIN32__ || __CYGWIN__ || __MINGW32__)
&& !GC_BUILD && !GC_NOT_DLL] (operator new, operator delete): Likewise.
* include/gc_cpp.h [_MSC_VER || __DMC__ || (__CYGWIN32__ || __CYGWIN__
|| __MINGW32__) && !GC_BUILD && !GC_NOT_DLL && __cplusplus>201103L]
(operator delete(void*,size_t)): Likewise.
* include/gc_cpp.h [_MSC_VER || __DMC__ || (__CYGWIN32__ || __CYGWIN__
|| __MINGW32__) && !GC_BUILD && !GC_NOT_DLL && __cplusplus>201103L
&& GC_OPERATOR_NEW_ARRAY && !CPPCHECK]
(operator delete[](void*,size_t)): Likewise.
* include/gc_cpp.h (operator new(size_t, int, const char*, int),
operator new[](size_t, int, const char*, int)): Do not define
for __DMC__.

Fix 'undefined reference to GC_incremental' linker error in pthread_start
(fix of commit 73d30d2)

GC_incremental is declared as static in case of NO_DEBUGGING.

* pthread_start.c [GC_PTHREADS && !GC_WIN32_THREADS]
(GC_inner_start_routine): Call GC_end_stubborn_change instead of
GC_dirty; add comment.

Use include gc.h with the angle brackets in the man page synopsis

* doc/gc.man (.TH): Update the date.
* doc/gc.man (SYNOPSIS): Replace include "gc.h" to include <gc.h>.

Travis CI: Use TEST_MANUAL_VDB instead of MANUAL_VDB

New API to turn on manual VDB at runtime

Manual VDB is now enabled by GC_set_manual_vdb_allowed(1) if called
before entering the GC incremental mode.

* alloc.c (GC_allocobj): Expand TRUE_INCREMENTAL macro.
* darwin_stop_world.c [MPROTECT_VDB] (GC_stop_world, GC_start_world):
Use GC_auto_incremental instead of GC_incremental.
* mark.c (alloc_mark_stack): Likewise.
* mark.c [PROC_VDB] (GC_push_all): Likewise.
* mark.c [!NEED_FIXUP_POINTER && THREADS && MPROTECT_VDB]
(GC_push_all_stack): Likewise.
* pthread_support.c [CAN_HANDLE_FORK && GC_DARWIN_THREADS
&& MPROTECT_VDB] (GC_atfork_prepare): Likewise.
* win32_threads.c [MPROTECT_VDB && !CYGWIN32]
(GC_register_my_thread_inner): Likewise.
* win32_threads.c [MPROTECT_VDB] (UNPROTECT_THREAD): Likewise.
* doc/gcdescr.md (Generational Collection and Dirty Bits): Update
documentation for the manual VDB.
* include/gc.h (GC_end_stubborn_change): Update comment.
* mark_rts.c (GC_push_all_stack_partially_eager): Likewise.
* include/gc.h (GC_set_manual_vdb_allowed, GC_get_manual_vdb_allowed):
New public function.
* include/private/gc_priv.h (GC_grungy_pages, GC_dirty_pages): Define
for all VDB modes.
* include/private/gc_priv.h (GC_auto_incremental, GC_manual_vdb):
Define.
* include/private/gc_priv.h [!GC_DISABLE_INCREMENTAL] (GC_dirty): Use
GC_manual_vdb instead of GC_incremental.
* include/private/gcconfig.h (GWW_VDB, MPROTECT_VDB, PCR_VDB): Do not
undefine if MANUAL_VDB.
* mallocx.c (GC_generic_malloc_many): Always allocate a single object
(and call GC_dirty_inner/REACHABLE_AFTER_DIRTY) if GC_manual_vdb.
* misc.c [!CAN_HANDLE_FORK && DARWIN && MPROTECT_VDB && !THREADS
&& !SMALL_CONFIG] (GC_set_handle_fork): Do not ABORT if GC_manual_vdb.
* misc.c [!SMALL_CONFIG] (manual_vdb_allowed): New static variable.
* misc.c [!SMALL_CONFIG] (GC_set_manual_vdb_allowed,
GC_get_manual_vdb_allowed): Implement.
* misc.c [!CHECKSUMS && !SMALL_CONFIG] (GC_init,
GC_enable_incremental): Set GC_manual_vdb and GC_incremental to true
if manual_vdb_allowed; do not call GC_dirty_init if manual_vdb_allowed.
* os_dep.c: Update comment about MANUAL_VDB.
* os_dep.c [MANUAL_VDB] (GC_dirty_init,
async_set_pht_entry_from_index): Remove.
* os_dep.c [!GC_DISABLE_INCREMENTAL] (GC_manual_vdb): Define global
variable.
* os_dep.c [!GC_DISABLE_INCREMENTAL] (GC_dirty_inner): Define
regardless of the VDB mode; add FIXME.
* os_dep.c [!GC_DISABLE_INCREMENTAL] (GC_read_dirty, GC_page_was_dirty,
GC_remove_protection): Implement for the case of GC_manual_vdb is true;
do not depend on MANUAL_VDB.
* tests/disclaim_test.c [TEST_MANUAL_VDB] (main): Call
GC_set_manual_vdb_allowed(1) before GC_INIT.
* tests/staticrootslib.c [TEST_MANUAL_VDB] (libsrl_init): Likewise.
* tests/test_cpp.cc [TEST_MANUAL_VDB] (main): Likewise.
* tests/test.c (INIT_MANUAL_VDB_ALLOWED): New macro.
* tests/test.c (GC_COND_INIT): Invoke INIT_MANUAL_VDB_ALLOWED (before
GC_OPT_INIT).
* tests/test.c [!SMALL_CONFIG] (main): Call GC_get_manual_vdb_allowed.

Update ChangeLog file (add v7.4.14 release date)

Update ChangeLog file (add v7.2l release date)

Update AUTHORS file

Initial single-threaded support of Interix subsystem

* include/private/gc_priv.h [__INTERIX] (_ALL_SOURCE): Define macro.
* include/private/gcconfig.h [__INTERIX] (I386, INTERIX,
mach_type_known): Likewise.
* include/private/gcconfig.h [I386 && INTERIX] (OS_TYPE, DATASTART,
DATAEND, STACKBOTTOM, USE_MMAP, USE_MMAP_ANON): Likewise.
* include/private/gcconfig.h [I386 && INTERIX] (_data_start__,
_bss_end__): Declare extern variable.

Update ChangeLog file

Update ChangeLog file (v7.6 changes)

Update ChangeLog file (v7.2 - v7.4 changes only)

Consistently use AS_HELP_STRING for configure new options
(code refactoring)

* configure.ac (thread-local-alloc, threads-discovery, mmap,
dynamic-loading, register-main-static-data, gcov): Replace deprecated
AC_HELP_STRING to AS_HELP_STRING.

Fix incorrect roots order after root removal in remove_roots_subregion
(fix of commits 38d194a, e849b45)

Issue #218 (bdwgc).

* mark_rts.c [USE_PROC_FOR_LIBRARIES] (swap_static_roots): New inline
function.
* mark_rts.c [USE_PROC_FOR_LIBRARIES] (GC_remove_roots_subregion):
Use swap_static_roots (after GC_add_roots_inner); if there is
a temporary root after GC_remove_root_at_pos then swap it with the
appropriate non-temporary one to ensure all temporary roots are
grouped at the end.

Fix the build for Emscripten
(part of commit 0f9ef2d from Unity-Technologies/bdwgc)

Issue #173 (bdwgc).

* mark_rts.c [!CPPCHECK && __GNUC__>=4] (GC_approx_sp): Do not use
__builtin_frame_address if STACK_NOT_SCANNED.

Fix the build for N3DS
(part of commit 0f9ef2d from Unity-Technologies/bdwgc)

Issue #173 (bdwgc).

* include/gc.h [GC_BUILD && NN_PLATFORM_CTR] (GC_on_thread_event_proc):
Define type.
* include/private/gc_locks.h [NN_PLATFORM_CTR || NINTENDO_SWITCH]
(GC_lock, GC_unlock): Declare function.
* include/private/gc_locks.h [NN_PLATFORM_CTR || NINTENDO_SWITCH]
(UNCOND_LOCK, UNCOND_UNLOCK): Define macro.
* include/private/gc_priv.h [THREADS && !SN_TARGET_ORBIS
&& !SN_TARGET_PSP2]: Do not include gc_atomic_ops.h if NN_PLATFORM_CTR.
* pthread_support.c [USE_SPIN_LOCK || !NO_PTHREAD_TRYLOCK] (GC_pause):
Call GC_noop1 instead of AO_compiler_barrier unless AO_CLEAR defined.

Update AUTHORS file (add Josh Peterson)

Fix the build for Orbis and PSP2
(part of commit 0f9ef2d from Unity-Technologies/bdwgc)

Issue #173 (bdwgc).

* include/private/pthread_stop_world.h [!GC_OPENBSD_UTHREADS && !NACL]
(thread_stop_info): Do not define last_stop_count if SN_TARGET_ORBIS
or SN_TARGET_PSP2.
* include/private/pthread_stop_world.h [!NACL && SN_TARGET_ORBIS]
(thread_stop_info): Declare registers field.

Fix compilation of darwin_stop_world for iOS 8+
(part of commit c6e7e36 from Unity-Technologies/bdwgc)

Issue #173 (bdwgc).

* darwin_stop_world.c [ARM32 && ARM_THREAD_STATE32]
(GC_stack_range_for): Set state to unified_state instead of
unified_state.ts_32.

Fix more compilation issues for XboxOne
(part of commit c6e7e36 from Unity-Technologies/bdwgc)

Issue #173 (bdwgc).

* include/private/gc_priv.h [!CPPCHECK && !PCR && MSWIN_XBOX1
&& !DebugBreak] (DebugBreak): Define to __debugbreak.
* misc.c [MSWIN32] (GC_win32_MessageBoxA): Do not define if MSWIN_XBOX1.
* misc.c [!PCR && !SMALL_CONFIG && MSWIN32] (GC_default_on_abort): Do
not call GC_win32_MessageBoxA() if MSWINRT_FLAVOR or MSWIN_XBOX1.

Allow register_main_static_data disabling in CMake script

* CMakeLists.txt (enable_register_main_static_data): New OPTION (on by
default).
* CMakeLists.txt [!enable_register_main_static_data]: Define
GC_DONT_REGISTER_MAIN_STATIC_DATA macro.

Allow dynamic_loading disabling in CMake script

* CMakeLists.txt (enable_dynamic_loading): New OPTION (on by default).
* CMakeLists.txt [!enable_dynamic_loading]: Define
IGNORE_DYNAMIC_LOADING macro.

Allow mmap enabling in CMake script

* CMakeLists.txt (enable_mmap): New OPTION (off by default).
* CMakeLists.txt [!enable_munmap && enable_mmap]: Define USE_MMAP macro.

Allow threads_discovery disabling in CMake script

* CMakeLists.txt (enable_threads_discovery): New OPTION (on by default).
* CMakeLists.txt [!enable_threads_discovery]: Define
GC_NO_THREADS_DISCOVERY macro.

Fix wrong expression for ENDIF in CMakeLists.txt
(fix of commit 1aabce0f9)

* CMakeLists.txt [enable_checksums]: Remove enable_threads for ENDIF.

Define DYNAMIC_LOADING for Darwin unless IGNORE_DYNAMIC_LOADING

DYNAMIC_LOADING macro is now controlled by
configure --disable-dynamic-loading option.

* include/private/gcconfig.h [DARWIN && (I386 || AARCH64 || ARM32
|| X86_64)] (DYNAMIC_LOADING): Define even if
GC_DONT_REGISTER_MAIN_STATIC_DATA.

New configure --disable-register-main-static-data option
(part of commit 6be3fb4 from Unity-Technologies/bdwgc)

Issue #173 (bdwgc).

* configure.ac (register-main-static-data): New AC_ARG_ENABLE.
* configure.ac [$enable_register_main_static_data=no]
(GC_DONT_REGISTER_MAIN_STATIC_DATA): Define AC macro.

New configure --disable-dynamic-loading option
(part of commit 6be3fb4 from Unity-Technologies/bdwgc)

Issue #173 (bdwgc).

* configure.ac (dynamic-loading): New AC_ARG_ENABLE.
* configure.ac [$enable_dynamic_loading=no] (IGNORE_DYNAMIC_LOADING):
Define AC macro.

New configure --enable-mmap option
(part of commit 6be3fb4 from Unity-Technologies/bdwgc)

Issue #173 (bdwgc).

* configure.ac (mmap): New AC_ARG_ENABLE (store the result to
gc_use_mmap).

Adjust formatting of configure help messages and config.h comments

* configure.ac (getcontext, execinfo.h, setjmp.h, GC_BUILTIN_ATOMIC):
Add trailing dot in AC_DEFINE comment message.
* configure.ac (gcj-support, java-finalization, atomic-uncollectable,
disclaim): Remove trailing dot in AC_HELP_STRING help message.
* configure.ac (gcj-support, sigrt-signals, java-finalization,
atomic-uncollectable, redirect-malloc, disclaim, large-config,
handle-fork, munmap, checksums, werror, single-obj-compilation,
gcov, docs): Start AC_HELP_STRING help message with a lowercase letter.
* configure.ac (werror): Wrap help message into AS_HELP_STRING.

Update AUTHORS file (add Michael DeRoy)

New configure --disable-threads-discovery option
(part of commit 7b2813c from Unity-Technologies/bdwgc)

Issue #173 (bdwgc).

* configure.ac (threads-discovery): New AC_ARG_ENABLE.
* configure.ac [$enable_threads_discovery=no]
(GC_NO_THREADS_DISCOVERY): Define macro.

Travis CI: Output content of disclaim_bench.log

Export stop/start_world_external only for multi-threaded builds
(fix of commit bb91f03)

Issue #173 (bdwgc).

Also, update GC_world_stopped value in GC_stop_world_external and
GC_start_world_external.

* include/gc.h (GC_stop_world_external, GC_start_world_external): Do not
declare unless GC_THREADS; add comment.
* misc.c (GC_stop_world_external, GC_start_world_external): Do not
define unless THREADS; add assertion that GC is initialized.
* misc.c [THREADS && THREAD_LOCAL_ALLOC] (GC_stop_world_external): Set
GC_world_stopped to true after STOP_WORLD; add assertion that the
world is not stopped.
* misc.c [THREADS && THREAD_LOCAL_ALLOC] (GC_start_world_external): Set
GC_world_stopped to false before START_WORLD; add assertion that the
world is stopped.
* tests/test.c [GC_PTHREADS && CPPCHECK] (main): Add UNTESTED for
GC_stop_world_external and GC_start_world_external.

Fix UNTESTED for multi-threaded API functions in gctest
(fix of commits c5d4ef9c1, 857bb8ad6)

* tests/test.c [!PCR && !GC_WIN32_THREADS && !GC_PTHREADS && CPPCHECK
&& THREADS] (main): Remove UNTESTED for GC_allow_register_threads,
GC_get_on_thread_event, GC_register_altstack, GC_set_on_thread_event.
* tests/test.c [GC_PTHREADS && CPPCHECK] (main): Add UNTESTED for
GC_allow_register_threads, GC_get_on_thread_event,
GC_register_altstack, GC_set_on_thread_event.

New API to stop and start the GC world externally
(part of commits 39e14be, 05e3377 from Unity-Technologies/bdwgc)

Issue #173 (bdwgc).

* include/gc.h (GC_stop_world_external, GC_start_world_external):
Declare new API function.
* misc.c (GC_stop_world_external, GC_start_world_external): Implement.

Update AUTHORS file (add Jonas Echterhoff)

Fix the BDWGC build on XboxOne
(fix of commit d16debf)

Issue #173 (bdwgc).

This fixes durango_get_mem() definition.

* os_dep.c [MSWIN_XBOX1] (durango_get_mem): Define even if
MMAP_SUPPORTED is not defined.

Fix suspend_thread_list in case of thread_suspend is interrupted
(fix of commit e8cdc7c)

* darwin_stop_world.c [!GC_NO_THREADS_DISCOVERY]
(GC_suspend_thread_list): Retry thread_suspend if failed with
KERN_ABORTED.

Perform thread_suspend in loop as it may be interrupted (Darwin)

* darwin_stop_world.c (GC_stop_world): Retry thread_suspend() on
KERN_ABORTED.