* fhandler_console.cc (beep): Use a more Windows-generic wav file if the beep
is missing. Use a more foolproof way to find out whether we should be
recreating the missing key.
* registry.h (reg_key::_disposition): New field.
(reg_key::created): New function.
* registry.cc (reg_key::reg_key): Set _disposition to zero by default.
(reg_key::build_key): Fill in _disposition field.
(syslogd_inited): Convert to enum type noting the exact result of
trying to connect to syslog daemon. Use this way throughout.
(connect_syslogd): New static function taking over the task to
connect to syslog socket. Use correct struct sockaddr_un instead of
struct sockaddr.
(try_connect_syslogd): Call connect_syslogd. If write fails on
connection oriented socket, try to reconnect to syslog socket and
try to write again.
of in_port_t and in_addr_t to top of file and use throughout. Use
sa_family_t type where appropriate.
(struct in6_addr): Change `unsigned char' to `uint8_t'.
(struct sockaddr_in6): Add sin6_scope_id member. Add comments.
* include/cygwin/socket.h: Include stdint.h. Move definition of
socklen_t to top of file. Define sa_family_t. Define struct
sockaddr_storage as per SUSv3.
* include/sys/un.h: Include cygwin/socket.h. Use sa_family_t type.
* child_info.h (CURR_CHILD_INFO_MAGIC): Reset.
(child_info::dwProcessId): Delete.
(child_info::straced): New variable.
(child_info::handle_fork): New member function.
* dcrt0.cc (in_forkee): New global variable.
(__cygwin_user_data::forkee): Mark as obsolete.
(do_global_ctors): Use in_forkee rather than user_data->forkee.
(get_cygwin_startup_info): Ditto. Deal with new straced field to allow strace
to deal with children of attached processes.
(initial_env): Accommodate changes to strace::hello.
(child_info_fork::handle_fork): Rename from plain old 'handle_fork'. Move
alloc_stack() call elsewhere.
(dll_crt0_0): Fill out more of user_data. Reference handle_fork via fork_info.
Add some debugging output.
(_dll_crt0): Don't wait for sync thread if sync_startup is invalid. Zero
sync_startup here. Call alloc_stack() here, if appropriate.
(dll_crt0_1): Use in_forkee rather than user_data->forkee.
(dll_crt0): Ditto.
* malloc_wrapper.cc (malloc_init): Ditto.
* dll_init.cc (in_forkee): Remove local static version of this variable.
(dll_list::load_after_fork): Don't set in_forkee here.
* external.cc (cygwin_internal): Use strace method rather than accessing field
directly.
* fhandler.cc (fhandler_base::read): Ditto.
* fhandler_tty.cc (fhandler_tty_common::__acquire_output_mutex): Ditto.
* fork.cc (frok::parent): Invoke strace write_childpid to communicate with
potential strace.
(child_copy): Add more detail to debugging output.
* init.cc (calibration_id): New static variable.
(prime_threads): Set sync_startup to invalid handle if we already know about
thread_func_ix. Use static calibration_id to hold calibration thread id.
* munge_threadfunc (munge_threadfunc): Don't try to debug if we don't find
threadfunc_ix.
(dll_entry): Avoid calling munge_threadfunc and _cygtls::remove on non-cygwin
threads invoked during process startup.
* pinfo.cc (set_myself): Always call strace.hello here regardless of DEBUGGING.
* sigproc.cc (child_info::child_info): Remove spurious handling of dwProcessId.
Set straced as appropriate.
* spawn.cc (spawn_guts): Rename ciresrv to ch. Invoke strace write_childpid to
communicate with potential strace.
* strace.cc: Include child_info.h.
(strace::hello): Remove inited test. Use active() method to test if strace has
been activated. Handle case where we are started before
(mypid): New function.
(strace::vsprntf): Try to deal more intelligently with case where progname may
not be filled out. Put pid in parentheses if it is a windows pid rather than a
cygwin pid. myself has been filled out.
(strace::write_childpid): New function for notifying strace about the creation
of children.
(strace::vprntf): Use strace method rather than accessing field directly.
(strace_printf): Ditto.
(strace::wm): Ditto.
* winsup.h (in_forkee): Declare.
* include/sys/strace.h (strace::write_childpid): Declare new function.
(strace::attached): Define new function.
(strace::active): Ditto.
(strace::active_val): Ditto.
(_STRACE_ON): Delete.
(_STRACE_OFF): Ditto.
(define_strace0): Use strace method rather than accessing field directly.
(strace_printf_wrap): Ditto.
(strace_printf_wrap1): Ditto.
*** cygwin utils changes:
* strace.cc (nprocesses): Make static global.
(quiet): New variable.
(strace_active): Ditto.
(add_child): Increment nprocesses here. Don't add a child if it is already
added (windows bug?). Report on child if not quiet.
(get_child): Just return NULL if child not found.
(remove_child): Report on child if not quiet.
(attach_process): Don't complain if given a windows process. Use windows pid
in error.
(handle_output_debug_string): Issue error if trying to manipulate a process
that we don't know about. Handle _STRACE_CHILD_PID - attach to reported child
when we get this.
(proc_child): Move nprocesses to file scope. Report on exceptions.
(longopts): Implement "--quiet".
(opts): Implement "-q".
(main): Manipulate quiet flag.
* utils.sgml (strace): Add words describing '-q'.
_cygtls::interrupt_now.
(_cygtls::handle_exceptions): Handle STATUS_DATATYPE_MISALIGNMENT as
SIGBUS error. Differ between unmapped memory (SEGV_MAPERR) and access
violation (SEGV_ACCERR) in case of STATUS_ACCESS_VIOLATION. Write
kernel log message in case of uncatched STATUS_ACCESS_VIOLATION.
(_cygtls::interrupt_now): Just call interruptible now instead of
evaluating interruptibility here.
(do_exit): Move minimal_printf...
* pinfo.cc (pinfo::exit): ...into here.
* strace.cc (strace::vprntf): Guarantee output to the console when
system_printf/api_fatal.
* heap.cc (heap_init): Wait a second before issuing an error when
ERROR_INVALID_ADDRESS since this is probably due to a CTRL-C handler sneaking
in, using the memory that we want to use for the heap, and, eventually exiting.
* cygtls.h (sockaddr_in): Use header rather than defining our own structure.
* exceptions.cc (_cygtls::interrupt_setup): Use exact contents of sa_mask
rather than assuming tht current sig should be masked, too.
(_cygtls::call_signal_handler): Use more aggressive locking.
* gendef (_sigbe): Wait until later before releasing incyg.
(_sigreturn): Remove more arguments to accommodate quasi-sa_sigaction support.
(_sigdelayed): Push arguments for sa_sigaction. More work needed here.
* signal.cc (sigaction): Implement SA_NODEFER.
* tlsoffsets.h: Regenerate.
* sigproc.cc (wait_sig): Use default buffer size of Windows 9x complains.
* pinfo.cc (_onreturn::dummy_handle): Remove.
(_onreturn::h): Make this a pointer.
(_onreturn::~_onreturn): Detect whether pointer is NULL rather than value is
NULL.
(_onreturn::_onreturn): Set h to NULL initially.
(_onreturn::no_close_p_handle): Set h to NULL.
(winpids::add): Initialize onreturn with value from p.hProcess immediately.
if we couldn't suspend signals.
* sigproc.cc (sig_send): Set sigCONT event when we see __SIGNOHOLD.
(wait_sig): Remove holding_signals. Create pipe with a buffer which will
theoretically cause blocking if there is nothing reading on the pipe. Wait for
sigCONT at end of loop when we have a __SIGHOLD.
(_onreturn): New helper class.
(winpids:add): Remove copied stuff. Try to put process handle into pinfo in
question and use _onreturn class to control when to close it.
(winpids::release): Remove use of copied array. Free procinfo when hProc is
NULL. Otherwise call release and call CloseHandle on hProc.
* pinfo.h (winpids::copied): Remove throughout class.
Handle this a different way.
* external.cc (cygwin_internal): Accommodate extra hook_or_detect_cygwin
argument.
* hookapi.cc (cygwin_internal): Fill in subsys variable with the subsystem of
the executable.
* spawn.cc (av::iscui): New variable.
(spawn_guts): Hide window when we don't have a console and this isn't
NT/XP/2003.
(av::fixup): Set iscui flag.
* winsup.h (hook_or_detect_cygwin): Accommodate extra argument.
* fhandler_console.cc (set_console_state_for_spawn): Ditto for the definition.
Only set invisible console for non-cygwin process. Remove debugging leftover.
* spawn.cc (spawn_guts): Pass argument denoting whether this is a cygwin
process to set_console_state_for_spawn and only call this function when
exec'ing.
(fhandler_console::invisible_console): Declare new variable.
(fhandler_console::need_invisible): Ditto.
(fhandler_console::has_a): Ditto.
* fhandler_console.cc (set_console_state_for_spawn): Eliminate return value.
Set up an invisible console if necessary prior to spawning.
(fhandler_console::invisible_console): Define.
* fhandler_tty.cc (fhandler_tty_slave::open): Use
fhandler_console::invisible_console to setup an invisible console.
* pipe.cc (fhandler_pipe::fixup_in_child): Define new function from
fixup_after_exec.
(fhandler_pipe::fixup_after_exec): Use fixup_in_child when appropriate.
(fhandler_pipe::fixup_after_fork): Ditto.
* spawn.cc (handle): Reorganize and modernize a little.
(spawn_guts): Rely on set_console_state_for_spawn to set the console into the
right state but don't create the process with "detached" flag if we have no
controlling tty since that confuses 'cmd'.
* dtable.cc (dtable::stdio_init): Don't set console as controlling terminal if
we have an invisible console.
* sigproc.cc (child_info::sync): Use correct name in ForceCloseHandle1.
* cygheap.cc (cygheap_fixup_in_child): Accommodate new child_copy arguments.
* dcrt0.cc (dll_data_start): Move definition here from fork.
(dll_data_end): Ditto.
(dll_bss_start): Ditto.
(dll_bss_end): Ditto.
(handle_fork): New function. Called when forked to deal with fork issues and
copy data to this process from the parent.
(dll_crt0_0): Call handle_fork when _PROC_FORK condition.
(dll_crt0): Don't copy user_data when we've forked. Don't zero first element
of main_environment ever.
(cygwin_dll_init): Ditto.
* fork.cc (child_copy): Rename from fork_copy and change arguments so that each
pair of things to copy gets its own descriptor.
(frok::child): Remove fixup_mmaps_after_fork call here. Move to handle_fork in
dcrt0.cc.
(frok::parent): Use child_copy rather than fork_copy and accommodate changes in
parameters.
* exceptions.cc (setup_handler): Delay test of whether we're locked until after
GetThreadContext has been called since there are apparently cases where
SuspendThread does not cause an immediate thread suspension.
Christopher Faylor
Corinna Vinschen