Dear Gnutella fans, Here is the ChangeLog for 1.1.6, which unfortunately I forgot to package!
# v 1.1.6 2015-11-08 [stable] This is a bug-fixing release. This time it focues on Windows, which apparently exercised some bugs in common code with other platforms -- so some fixes are going to benefit to everyone even though the bugs did not manifest themselves as often. Improvements: - Increased default and max for "max_leaves" to account for modern systems. - Pulled latest translations from transifex. - Updated GeoIP databases. Bug Fixes: - Fixed memory leak on the error path of OOB / G2 query hit sending. - Fixed memory leak on the file upload path. - aging_insert(): was not properly handling freeing of keys, causing leaks. - Fixed property description message for "max_leaves". - Windows: no HEAP_NO_SERIALIZE flag on HeapAlloc() since we're multi-threaded. - mingw_valloc(): fixed monitoring of unreserved VM space. - Windows: optimized the inital VM space reservation. - rpc_delay(): avoid int overflowing, which would make delay negative! - node_udp_sr_data_ind(): do not crash if UDP was disabled or port was 0. - Fixed crashes in the omalloc() layer. Under the Hood: - Make sure an initial GMT offset is computed before the time thread starts. - Compact the aging structure on 64-bit platforms. - Fixed possible race between evq_trampoline() and evq_event_discard(). - Added safety assertions: a removed item must belong to the list. - Fixed typos in safety assertions. - EVQ: closed race condition betweeen queues recreated in same thread ID. - Use getrlimit(RLIMIT_AS) to get the available space, not RLIMIT_DATA. - Windows: try harder to compute a relevant break value with sbrk(0). - Windows: added cpp trap for dup(). - Windows: fixed fcntl(F_DUPFD) implementation which was not thread-safe. - Windows: dump symbolic PC on stack overflows, after basic information. - elist_remove(): assert that removed item must be part of some list. - evq_event_discard(): be quiet when removing event for the event queue thread. - evq_event_discard(): correctly handle cancelable events. - evq_trampoline(): factorize removal of triggered item from queue list. - evq_trampoline(): fixed logging message to use copied value. - getgateway(): on Linux, cache previously computed value in case netlink fails. - getgateway(): on Linux, the netlink socket can return EAGAIN on recv(). - main_command_line(): optimized and simplified! - malloc_init_vtable(): be sure to set G_SLICE=always-malloc on Windows. - mingw_exception(): skip printing of stacktrace if we got a stack overflow. - mingw_getrlimit(): implement a better RLIMIT_DATA query. - mingw_getrlimit(): return different max / current values to account for usage. - mingw_signal(): always record signal handler in our internal list of handlers. - mingw_valloc(): don't complain about non-hinted allocations when crashing! - mingw_valloc(): use s_minicarp() to limit resource consumption. - mingw_valloc(): when nearing out-of-memory, take data segment into account. - mingw_vmm_init(): accelerate initialization, reducing amount of system calls. - mingw_win2posix(): handle 0 to avoid warnings when called with no error. - mingw_win2posix(): mapped ERROR_BAD_EXE_FORMAT to ENOEXEC. - mingw_win2posix(): mapped ERROR_NETNAME_DELETED to EHOSTUNREACH. - mingw_win2posix(): use s_minicarp() instead of s_warning() to see "culprit". - omalloc_allocate(): was releasing spinlock too soon in read-only mode. - omalloc_chunk_allocate_from(): fixed bug when header dissolution was required. - publisher_handle(): simplified code by factorizing shared_file_unref() call. - s_minicarp() and s_minicrit(): carefully avoid stdio when dumping stack. - sdbm_close(): ensure we sync pending dirty data when database is not volatile. - sdbm_name(): return path to the .pag file if no name was explicitly given. - settings_init(): when computing max VM space, use max from RLIMIT_AS, not cur. - socket_create_and_bind(): on Linux systems, force SO_REUSEADDR before bind(). - sort-test: reduce verbosity by default. - thread_suspend_others(): silently suspends itself if concurrently asked to. - udp_sched_write_error(): don't dump a stack trace on send errors we expect. Enjoy! Raphael ------------------------------------------------------------------------------ _______________________________________________ gtk-gnutella-announce mailing list gtk-gnutella-annou...@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/gtk-gnutella-announce ------------------------------------------------------------------------------ _______________________________________________ gtk-gnutella-devel mailing list gtk-gnutella-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/gtk-gnutella-devel
