moo/lib/std.c

/*
 * $Id$
 *
    Copyright (c) 2014-2019 Chung, Hyung-Hwan. All rights reserved.

    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions
    are met:
    1. Redistributions of source code must retain the above copyright
       notice, this list of conditions and the following disclaimer.
    2. Redistributions in binary form must reproduce the above copyright
       notice, this list of conditions and the following disclaimer in the
       documentation and/or other materials provided with the distribution.

    THIS SOFTWARE IS PROVIDED BY THE AUTHOR "AS IS" AND ANY EXPRESS OR
    IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
    OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
    IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
    INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
    NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
    THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#if !defined(_GNU_SOURCE)
#	define _GNU_SOURCE
#endif

#include <moo-std.h>
#include "moo-prv.h"

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#if !defined(__DOS__) && !defined(EMSCRIPTEN) && defined(HAVE_PTHREAD) && defined(HAVE_STRERROR_R)
#	define USE_THREAD
#endif

#if defined(_WIN32)
#	if !defined(_WIN32_WINNT)
#		define _WIN32_WINNT 0x0400
#	endif
#	define WIN32_LEAN_AND_MEAN

#	include <windows.h>
#	if !(defined(__BORLANDC__) && (__BORLANDC__ <= 0x0520))
#		include <psapi.h>
#	endif
#	include <tchar.h>
#	include <time.h>
#	include <io.h>
#	include <fcntl.h>
#	include <signal.h>
#	include <errno.h>
#	if defined(MOO_HAVE_CFG_H) && defined(MOO_ENABLE_LIBLTDL)
#		include <ltdl.h>
#		define USE_LTDL
#	else
#		define USE_WIN_DLL
#	endif

#	include "poll-msw.h"
#	define USE_POLL
#	define XPOLLIN POLLIN
#	define XPOLLOUT POLLOUT
#	define XPOLLERR POLLERR
#	define XPOLLHUP POLLHUP

#if !defined(SIZE_T)
#	define SIZE_T unsigned long int
#endif

#elif defined(__OS2__)
#	define INCL_DOSMODULEMGR
#	define INCL_DOSPROCESS
#	define INCL_DOSSEMAPHORES
#	define INCL_DOSEXCEPTIONS
#	define INCL_DOSMISC
#	define INCL_DOSDATETIME
#	define INCL_DOSFILEMGR
#	define INCL_DOSERRORS
#	include <os2.h>
#	include <time.h>
#	include <fcntl.h>
#	include <io.h>
#	include <errno.h>

#	include <types.h> /* some types for socket.h */
#	include <sys/socket.h> /* for socketpair */
#	include <sys/time.h>
#	include <sys/ioctl.h> /* FIONBIO */
#   include <nerrno.h> /* for SOCEXXX error codes */

#	define BSD_SELECT
#	if defined(TCPV40HDRS)
#	include <sys/select.h>
#	include <sys/un.h> /* for sockaddr_un */
#	else
#	include <unistd.h>
#	endif

#	define USE_SELECT
	/* fake XPOLLXXX values */
#	define XPOLLIN  (1 << 0)
#	define XPOLLOUT (1 << 1)
#	define XPOLLERR (1 << 2)
#	define XPOLLHUP (1 << 3)

#elif defined(__DOS__)
#	include <dos.h>
#	include <time.h>
#	include <io.h>
#	include <signal.h>
#	include <errno.h>
#	include <fcntl.h>
#	include <conio.h> /* inp, outp */

#	if defined(_INTELC32_)
#		define DOS_EXIT 0x4C
#		include <i32.h>
#		include <stk.h>
#	else
#		include <dosfunc.h>
#	endif

	/* fake XPOLLXXX values */
#	define XPOLLIN  (1 << 0)
#	define XPOLLOUT (1 << 1)
#	define XPOLLERR (1 << 2)
#	define XPOLLHUP (1 << 3)

#elif defined(macintosh)
#	include <Types.h>
#	include <OSUtils.h>
#	include <Timer.h>

#	include <MacErrors.h>
#	include <Process.h>
#	include <Dialogs.h>
#	include <TextUtils.h>

	/* TODO: a lot to do */

#elif defined(vms) || defined(__vms)
#	define __NEW_STARLET 1
#	include <starlet.h> /* (SYS$...) */
#	include <ssdef.h> /* (SS$...) */
#	include <lib$routines.h> /* (lib$...) */

	/* TODO: a lot to do */

#else
#	include <sys/types.h>
#	include <unistd.h>
#	include <fcntl.h>
#	include <errno.h>

#	if defined(MOO_ENABLE_LIBLTDL)
#		include <ltdl.h>
#		define USE_LTDL
#	elif defined(HAVE_DLFCN_H)
#		include <dlfcn.h>
#		define USE_DLFCN
#	elif defined(__APPLE__) || defined(__MACOSX__)
#		define USE_MACH_O_DYLD
#		include <mach-o/dyld.h>
#	else
#		error UNSUPPORTED DYNAMIC LINKER
#	endif

#	if defined(HAVE_TIME_H)
#		include <time.h>
#	endif
#	if defined(HAVE_SYS_TIME_H)
#		include <sys/time.h>
#	endif
#	if defined(HAVE_SIGNAL_H)
#		include <signal.h>
#	endif
#	if defined(HAVE_SYS_MMAN_H)
#		include <sys/mman.h>
#	endif

#	if defined(USE_THREAD)
#		include <pthread.h>
#		include <sched.h>
#	endif

#	if defined(HAVE_SYS_DEVPOLL_H)
		/* solaris */
#		include <sys/devpoll.h>
#		define USE_DEVPOLL
#		define XPOLLIN POLLIN
#		define XPOLLOUT POLLOUT
#		define XPOLLERR POLLERR
#		define XPOLLHUP POLLHUP
#	elif defined(HAVE_SYS_EVENT_H) && defined(HAVE_KQUEUE)
		/* netbsd, openbsd, etc */
#		include <sys/event.h>
#		define USE_KQUEUE
		/* fake XPOLLXXX values */
#		define XPOLLIN  (1 << 0)
#		define XPOLLOUT (1 << 1)
#		define XPOLLERR (1 << 2)
#		define XPOLLHUP (1 << 3)
#	elif defined(HAVE_SYS_EPOLL_H) && defined(HAVE_EPOLL_CREATE)
		/* linux */
#		include <sys/epoll.h>
#		define USE_EPOLL
#		define XPOLLIN EPOLLIN
#		define XPOLLOUT EPOLLOUT
#		define XPOLLERR EPOLLERR
#		define XPOLLHUP EPOLLHUP
#	elif defined(HAVE_POLL_H)
#		include <poll.h>
#		define USE_POLL
#		define XPOLLIN POLLIN
#		define XPOLLOUT POLLOUT
#		define XPOLLERR POLLERR
#		define XPOLLHUP POLLHUP
#	else
#		define USE_SELECT
		/* fake XPOLLXXX values */
#		define XPOLLIN  (1 << 0)
#		define XPOLLOUT (1 << 1)
#		define XPOLLERR (1 << 2)
#		define XPOLLHUP (1 << 3)
#	endif

#endif

#if !defined(MOO_DEFAULT_PFMODDIR)
#	define MOO_DEFAULT_PFMODDIR ""
#endif

#if !defined(MOO_DEFAULT_PFMODPREFIX)
#	if defined(_WIN32)
#		define MOO_DEFAULT_PFMODPREFIX "moo-"
#	elif defined(__OS2__)
#		define MOO_DEFAULT_PFMODPREFIX "moo"
#	elif defined(__DOS__)
#		define MOO_DEFAULT_PFMODPREFIX "moo"
#	else
#		define MOO_DEFAULT_PFMODPREFIX "libmoo-"
#	endif
#endif

#if !defined(MOO_DEFAULT_PFMODPOSTFIX)
#	if defined(_WIN32)
#		define MOO_DEFAULT_PFMODPOSTFIX ""
#	elif defined(__OS2__)
#		define MOO_DEFAULT_PFMODPOSTFIX ""
#	elif defined(__DOS__)
#		define MOO_DEFAULT_PFMODPOSTFIX ""
#	else
#		if defined(USE_DLFCN)
#			define MOO_DEFAULT_PFMODPOSTFIX ".so"
#		elif defined(USE_MACH_O_DYLD)
#			define MOO_DEFAULT_PFMODPOSTFIX ".dylib"
#		else
#			define MOO_DEFAULT_PFMODPOSTFIX ""
#		endif
#	endif
#endif

#if defined(USE_THREAD)
#	define MUTEX_INIT(x) pthread_mutex_init((x), MOO_NULL)
#	define MUTEX_DESTROY(x) pthread_mutex_destroy(x)
#	define MUTEX_LOCK(x) pthread_mutex_lock(x)
#	define MUTEX_UNLOCK(x) pthread_mutex_unlock(x)
#else
#	define MUTEX_INIT(x)
#	define MUTEX_DESTROY(x)
#	define MUTEX_LOCK(x)
#	define MUTEX_UNLOCK(x)
#endif

typedef struct bb_t bb_t;
struct bb_t
{
	char buf[MOO_IOBUF_CAPA];
	moo_oow_t pos;
	moo_oow_t len;

	FILE* fp;
	moo_bch_t* fn;
	moo_cmgr_t* cmgr;
};

#if defined(USE_SELECT)
struct select_fd_t
{
	int fd;
	int events;
};
#endif

typedef struct xtn_t xtn_t;
struct xtn_t
{
	moo_t* next;
	moo_t* prev;

	int vm_running;
	int rcv_tick;

	moo_cmgr_t* input_cmgr;
	moo_cmgr_t* log_cmgr;

	struct
	{
		int fd;
		int fd_flag; /* bitwise OR'ed fo logfd_flag_t bits */

		struct
		{
			moo_bch_t buf[4096];
			moo_oow_t len;
		} out;
	} log;

	const moo_iostd_t* in;

	#if defined(_WIN32)
	HANDLE waitable_timer;
	DWORD tc_last;
	DWORD tc_overflow;
	#elif defined(__OS2__)
	ULONG tc_last;
	moo_ntime_t tc_last_ret;
	#elif defined(__DOS__)
	clock_t tc_last;
	moo_ntime_t tc_last_ret;
	#endif

	#if defined(USE_DEVPOLL)
	int ep; /* /dev/poll */
	#elif defined(USE_KQUEUE)
	int ep; /* kqueue */
	#elif defined(USE_EPOLL)
	int ep; /* epoll */
	#elif defined(USE_POLL)
	/* nothing */
	#elif defined(USE_SELECT)
	/* nothing */
	#endif

	#if defined(USE_THREAD)
	struct
	{
		int p[2]; /* pipe for signaling */
		pthread_t thr;
		int up;
		int abort;
	} iothr;
	#endif

	struct
	{
		int p[2]; /* pipe for signaling */
	} sigfd;

	struct
	{
	#if defined(USE_DEVPOLL)
		/*TODO: make it dynamically changeable depending on the number of
		 *      file descriptors added */
		struct pollfd buf[64]; /* buffer for reading events */
	#elif defined(USE_KQUEUE)
		struct
		{
			moo_oow_t* ptr;
			moo_oow_t capa;
		} reg;
		struct kevent buf[64];
	#elif defined(USE_EPOLL)
		/*TODO: make it dynamically changeable depending on the number of
		 *      file descriptors added */
		struct epoll_event buf[64]; /* buffer for reading events */
	#elif defined(USE_POLL)
		struct
		{
			struct pollfd* ptr;
			moo_oow_t capa;
			moo_oow_t len;
		#if defined(USE_THREAD)
			pthread_mutex_t pmtx;
		#endif
		} reg; /* registry */
		struct pollfd* buf;

	#elif defined(USE_SELECT)
		struct
		{
			fd_set rfds;
			fd_set wfds;
			int maxfd;
		#if defined(USE_THREAD)
			pthread_mutex_t smtx;
		#endif
		} reg;

		struct select_fd_t buf[FD_SETSIZE];
	#endif

		moo_oow_t len;

	#if defined(USE_THREAD)
		pthread_mutex_t mtx;
		pthread_cond_t cnd;
		pthread_cond_t cnd2;
	#endif

		int halting;
	} ev;
};

#define GET_XTN(moo) ((xtn_t*)((moo_uint8_t*)moo_getxtn(moo) - MOO_SIZEOF(xtn_t)))

static moo_t* g_moo = MOO_NULL;

/* ========================================================================= */

static void* sys_alloc (moo_mmgr_t* mmgr, moo_oow_t size)
{
	return malloc(size);
}

static void* sys_realloc (moo_mmgr_t* mmgr, void* ptr, moo_oow_t size)
{
	return realloc(ptr, size);
}

static void sys_free (moo_mmgr_t* mmgr, void* ptr)
{
	free (ptr);
}

static moo_mmgr_t sys_mmgr =
{
	sys_alloc,
	sys_realloc,
	sys_free,
	MOO_NULL
};

/* ========================================================================= */

#if defined(_WIN32) || defined(__OS2__) || defined(__DOS__)
#	define IS_PATH_SEP(c) ((c) == '/' || (c) == '\\')
#else
#	define IS_PATH_SEP(c) ((c) == '/')
#endif

/* TODO: handle path with a drive letter or in the UNC notation */
#define IS_PATH_ABSOLUTE(x) IS_PATH_SEP(x[0])

static const moo_bch_t* get_base_name (const moo_bch_t* path)
{
	const moo_bch_t* p, * last = MOO_NULL;

	for (p = path; *p != '\0'; p++)
	{
		if (IS_PATH_SEP(*p)) last = p;
	}

	return (last == MOO_NULL)? path: (last + 1);
}

static MOO_INLINE moo_ooi_t open_input (moo_t* moo, moo_ioarg_t* arg)
{
	xtn_t* xtn = GET_XTN(moo);
	bb_t* bb = MOO_NULL;
	moo_bch_t* at;

/* TOOD: support predefined include directory as well */
	if (arg->includer)
	{
		/* includee */
		moo_oow_t ucslen, bcslen, parlen;
		const moo_bch_t* fn, * fb;

	#if defined(MOO_OOCH_IS_UCH)
		if (moo_convootobcstr(moo, arg->name, &ucslen, MOO_NULL, &bcslen) <= -1) goto oops;
	#else
		bcslen = moo_count_bcstr(arg->name);
	#endif

		fn = ((bb_t*)arg->includer->handle)->fn;

		fb = get_base_name(fn);
		parlen = fb - fn;

		bb = moo_callocmem(moo, MOO_SIZEOF(*bb) + (MOO_SIZEOF(moo_bch_t) * (parlen + bcslen + 1)));
		if (!bb) goto oops;

		bb->fn = (moo_bch_t*)(bb + 1);
		moo_copy_bchars (bb->fn, fn, parlen);
	#if defined(MOO_OOCH_IS_UCH)
		moo_convootobcstr (moo, arg->name, &ucslen, &bb->fn[parlen], &bcslen);
	#else
		moo_copy_bcstr (&bb->fn[parlen], bcslen + 1, arg->name);
	#endif
	}
	else
	{
		/* main stream */
		moo_oow_t ucslen, bcslen;

		switch (xtn->in->type)
		{
		#if defined(MOO_OOCH_IS_BCH)
			case MOO_IOSTD_FILE:
				MOO_ASSERT (moo, &xtn->in->u.fileb.path == &xtn->in->u.file.path);
		#endif
			case MOO_IOSTD_FILEB:
				bcslen = moo_count_bcstr(xtn->in->u.fileb.path);

				bb = moo_callocmem(moo, MOO_SIZEOF(*bb) + (MOO_SIZEOF(moo_bch_t) * (bcslen + 1)));
				if (!bb) goto oops;

				bb->fn = (moo_bch_t*)(bb + 1);
				moo_copy_bcstr (bb->fn, bcslen + 1, xtn->in->u.fileb.path);

				break;

		#if defined(MOO_OOCH_IS_UCH)
			case MOO_IOSTD_FILE:
				MOO_ASSERT (moo, &xtn->in->u.fileu.path == &xtn->in->u.file.path);
		#endif
			case MOO_IOSTD_FILEU:
				if (moo_convutobcstr(moo, xtn->in->u.fileu.path, &ucslen, MOO_NULL, &bcslen) <= -1)
				{
					moo_seterrbfmt (moo, moo_geterrnum(moo), "unable to convert encoding of path %ls", xtn->in->u.fileu.path);
					goto oops;
				}

				bb = moo_callocmem(moo, MOO_SIZEOF(*bb) + (MOO_SIZEOF(moo_bch_t) * (bcslen + 1)));
				if (!bb) goto oops;

				bb->fn = (moo_bch_t*)(bb + 1);
				bcslen += 1;
				moo_convutobcstr (moo, xtn->in->u.fileu.path, &ucslen, bb->fn, &bcslen);

				break;

			default:
				moo_seterrbfmt (moo, MOO_EINVAL, "unsupported standard input type - %d", (int)xtn->in->type);
				goto oops;
		}
	}

	/* the path name can be appened with @encoding */
	at = moo_find_bchar_in_bcstr(bb->fn, '@');
	if (at && at[1] != '\0')
	{
		*at = '\0';
		bb->cmgr = moo_get_cmgr_by_bcstr(at + 1);
		if (!bb->cmgr)
		{
			moo_seterrbfmt (moo, MOO_EINVAL, "unsupported charset - %hs", at + 1);
			goto oops;
		}
	}
	else
	{
		bb->cmgr = arg->includer? ((bb_t*)arg->includer->handle)->cmgr: xtn->in->cmgr;
		if (!bb->cmgr) bb->cmgr = xtn->input_cmgr;
	}

/* TODO: support _wfopen or the like */
#if defined(_WIN32) || defined(__OS2__) || defined(__DOS__)
	bb->fp = fopen(bb->fn, "rb");
#else
	bb->fp = fopen(bb->fn, "r");
#endif
	if (!bb->fp)
	{
		moo_seterrbfmtwithsyserr (moo, 0, errno, "unable to open file %hs", bb->fn);
		goto oops;
	}

	arg->handle = bb;
	if (!arg->name)
	{
		/* for the top-level stream, i simply change the arg->name field
		 * so that the compiler knows the file to be read */
		moo_oocs_t io_name;

		io_name.ptr = moo_dupbtooocstr(moo, get_base_name(bb->fn), &io_name.len);
		if (!io_name.ptr) goto oops;

		arg->name = moo_addcioname(moo, &io_name);
		moo_freemem (moo, io_name.ptr);

		if (!arg->name) goto oops;
	}

	return 0;

oops:
	if (bb)
	{
		if (bb->fp) fclose (bb->fp);
		moo_freemem (moo, bb);
	}
	return -1;
}

static MOO_INLINE moo_ooi_t close_input (moo_t* moo, moo_ioarg_t* arg)
{
	bb_t* bb;

	bb = (bb_t*)arg->handle;
	MOO_ASSERT (moo, bb != MOO_NULL && bb->fp != MOO_NULL);

	fclose (bb->fp);
	moo_freemem (moo, bb);

	arg->handle = MOO_NULL;
	return 0;
}

static MOO_INLINE moo_ooi_t read_input (moo_t* moo, moo_ioarg_t* arg)
{
	bb_t* bb;
	moo_oow_t bcslen, ucslen, remlen;
	int x;

	bb = (bb_t*)arg->handle;
	MOO_ASSERT (moo, bb != MOO_NULL && bb->fp != MOO_NULL);
	do
	{
		x = fgetc(bb->fp);
		if (x == EOF)
		{
			if (ferror((FILE*)bb->fp))
			{
				moo_seterrbfmtwithsyserr (moo, 0, errno, "unable to read file %hs", bb->fn);
				return -1;
			}
			break;
		}

		bb->buf[bb->len++] = x;
	}
	while (bb->len < MOO_COUNTOF(bb->buf) && x != '\r' && x != '\n');

#if defined(MOO_OOCH_IS_UCH)
	bcslen = bb->len;
	ucslen = MOO_COUNTOF(arg->buf);
	/*x = moo_convbtooochars(moo, bb->buf, &bcslen, arg->buf, &ucslen);*/
	x = moo_conv_bchars_to_uchars_with_cmgr(bb->buf, &bcslen, arg->buf, &ucslen, bb->cmgr, 0);
	if (x <= -1 /*&& ucslen <= 0 */)
	{
		moo_seterrbfmt (moo, (x == -2)? MOO_EBUFFULL: MOO_EECERR, "unable to convert source input - %.*hs", bcslen, bb->buf);
		return -1;
	}

	/* if ucslen is greater than 0, i see that some characters have been
	 * converted properly */
#else
	bcslen = (bb->len < MOO_COUNTOF(arg->buf))? bb->len: MOO_COUNTOF(arg->buf);
	ucslen = bcslen;
	moo_copy_bchars (arg->buf, bb->buf, bcslen);
#endif

	remlen = bb->len - bcslen;
	if (remlen > 0) MOO_MEMMOVE (bb->buf, &bb->buf[bcslen], remlen);
	bb->len = remlen;
	return ucslen;
}

static moo_ooi_t input_handler (moo_t* moo, moo_iocmd_t cmd, moo_ioarg_t* arg)
{
	switch (cmd)
	{
		case MOO_IO_OPEN:
			return open_input(moo, arg);

		case MOO_IO_CLOSE:
			return close_input(moo, arg);

		case MOO_IO_READ:
			return read_input(moo, arg);

		default:
			moo_seterrnum (moo, MOO_EINTERN);
			return -1;
	}
}

/* ========================================================================= */


static int get_huge_page_size (moo_t* moo, moo_oow_t* page_size)
{
	FILE* fp;
	char buf[256];

	fp = fopen("/proc/meminfo", "r");
	if (!fp) return -1;

	while (!feof(fp))
	{
		if (fgets(buf, sizeof(buf) - 1, fp) == NULL) goto oops;

		if (strncmp(buf, "Hugepagesize: ", 13) == 0)
		{
			unsigned long int tmp;
			tmp = strtoul(&buf[13], NULL, 10);
			if (tmp == MOO_TYPE_MAX(unsigned long int) && errno == ERANGE) goto oops;

			*page_size = tmp * 1024; /* KBytes to Bytes */
			fclose (fp);
			return 0;
		}
	}

oops:
	fclose (fp);
	return -1;
}

static void* alloc_heap (moo_t* moo, moo_oow_t* size)
{
#if defined(_WIN32)
	moo_oow_t* ptr;
	moo_oow_t req_size, align, aligned_size;
	HINSTANCE k32;
	SIZE_T (*k32_GetLargePageMinimum) (void);
	HANDLE token = MOO_NULL;
	TOKEN_PRIVILEGES new_state, prev_state;
	TOKEN_PRIVILEGES* prev_state_ptr;
	DWORD prev_state_reqsize = 0;
	int token_adjusted = 0;

	align = 2 * 1024 * 1024; /* default 2MB */

	k32 = LoadLibrary(TEXT("kernel32.dll"));
	if (k32)
	{
		k32_GetLargePageMinimum = (SIZE_T(*)(void))GetProcAddress (k32, "GetLargePageMinimum");
		if (k32_GetLargePageMinimum) align = k32_GetLargePageMinimum();
		FreeLibrary (k32);
	}
	/* the standard page size shouldn't help. so let me comment out this part.
	else
	{
		SYSTEM_INFO si;
		GetSystemInfo (&si);
		align = si.dwPageSize;
	}*/

	req_size = MOO_SIZEOF(moo_oow_t) + size;
	aligned_size = MOO_ALIGN(req_size, align);

	if (!OpenProcessToken(GetCurrentProcess(), TOKEN_ADJUST_PRIVILEGES | TOKEN_QUERY, &token)) goto oops;
	if (!LookupPrivilegeValue(MOO_NULL, TEXT("SeLockMemoryPrivilege"), &new_state.Privileges[0].Luid)) goto oops;
	new_state.PrivilegeCount = 1;
	new_state.Privileges[0].Attributes = SE_PRIVILEGE_ENABLED;

	prev_state_ptr = &prev_state;
	if (!AdjustTokenPrivileges(token, FALSE, &new_state, MOO_SIZEOF(prev_state), prev_state_ptr, &prev_state_reqsize) || GetLastError() != ERROR_SUCCESS)
	{
		if (prev_state_reqsize >= MOO_SIZEOF(prev_state))
		{
			/* GetLastError() == ERROR_INSUFFICIENT_BUFFER */
			prev_state_ptr = (TOKEN_PRIVILEGES*)HeapAlloc(GetProcessHeap(), 0, prev_state_reqsize);
			if (!prev_state_ptr) goto oops;
			if (!AdjustTokenPrivileges(token, FALSE, &new_state, prev_state_reqsize, prev_state_ptr, &prev_state_reqsize) || GetLastError() != ERROR_SUCCESS) goto oops;
		}
		else goto oops;
	}
	token_adjusted = 1;

#if !defined(MEM_LARGE_PAGES)
#	define MEM_LARGE_PAGES (0x20000000)
#endif
	ptr = VirtualAlloc(MOO_NULL, aligned_size, MEM_COMMIT | MEM_RESERVE | MEM_LARGE_PAGES, PAGE_READWRITE);
	if (!ptr)
	{
		SYSTEM_INFO si;
		GetSystemInfo (&si);
		align = si.dwPageSize;
		aligned_size = MOO_ALIGN(req_size, align);
		ptr = VirtualAlloc(MOO_NULL, aligned_size, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
		if (!ptr) goto oops;
	}

	AdjustTokenPrivileges (token, FALSE, prev_state_ptr, 0, MOO_NULL, 0);
	CloseHandle (token);
	if (prev_state_ptr && prev_state_ptr != &prev_state) HeapFree (GetProcessHeap(), 0, prev_state_ptr);

	*size = aligned_size;
	return ptr;

oops:
	moo_seterrwithsyserr (moo, 1, GetLastError());
	if (token)
	{
		if (token_adjusted) AdjustTokenPrivileges (token, FALSE, prev_state_ptr, 0, MOO_NULL, 0);
		CloseHandle (token);
	}
	if (prev_state_ptr && prev_state_ptr != &prev_state) HeapFree (GetProcessHeap(), 0, prev_state_ptr);
	return MOO_NULL;

#elif defined(HAVE_MMAP) && defined(HAVE_MUNMAP) && defined(MAP_ANONYMOUS)
	/* It's called via moo_makeheap() when MOO creates a GC heap.
	 * The heap is large in size. I can use a different memory allocation
	 * function instead of an ordinary malloc.
	 * upon failure, it doesn't require to set error information as moo_makeheap()
	 * set the error number to MOO_EOOMEM. */

#if !defined(MAP_HUGETLB) && (defined(__amd64__) || defined(__x86_64__))
#	define MAP_HUGETLB 0x40000
#endif

	moo_oow_t* ptr;
	int flags;
	moo_oow_t req_size, align, aligned_size;

	req_size = MOO_SIZEOF(moo_oow_t) + *size;
	flags = MAP_PRIVATE | MAP_ANONYMOUS;

	#if defined(MAP_UNINITIALIZED)
	flags |= MAP_UNINITIALIZED;
	#endif

	#if defined(MAP_HUGETLB)
	if (get_huge_page_size(moo, &align) <= -1) align = 2 * 1024 * 1024; /* default to 2MB */
	if (req_size > align / 2)
	{
		/* if the requested size is large enough, attempt HUGETLB */
		flags |= MAP_HUGETLB;
	}
	else
	{
		align = sysconf(_SC_PAGESIZE);
	}
	#else
	align = sysconf(_SC_PAGESIZE);
	#endif

	aligned_size = MOO_ALIGN_POW2(req_size, align);
	ptr = (moo_oow_t*)mmap(NULL, aligned_size, PROT_READ | PROT_WRITE, flags, -1, 0);
	#if defined(MAP_HUGETLB)
	if (ptr == MAP_FAILED && (flags & MAP_HUGETLB))
	{
		flags &= ~MAP_HUGETLB;
		align = sysconf(_SC_PAGESIZE);
		aligned_size = MOO_ALIGN_POW2(req_size, align);
		ptr = (moo_oow_t*)mmap(NULL, aligned_size, PROT_READ | PROT_WRITE, flags, -1, 0);
		if (ptr == MAP_FAILED)
		{
			moo_seterrwithsyserr (moo, 0, errno);
			return MOO_NULL;
		}
	}
	#else
	if (ptr == MAP_FAILED)
	{
		moo_seterrwithsyserr (moo, 0, errno);
		return MOO_NULL;
	}
	#endif

	*ptr = aligned_size;
	*size = aligned_size - MOO_SIZEOF(moo_oow_t);
	return (void*)(ptr + 1);

#else
	return MOO_MMGR_ALLOC(moo->_mmgr, *size);
#endif
}

static void free_heap (moo_t* moo, void* ptr)
{
#if defined(_WIN32)
	VirtualFree (ptr, 0, MEM_RELEASE); /* release the entire region */

#elif defined(HAVE_MMAP) && defined(HAVE_MUNMAP)
	moo_oow_t* actual_ptr;
	actual_ptr = (moo_oow_t*)ptr - 1;
	munmap (actual_ptr, *actual_ptr);
#else
	MOO_MMGR_FREE(moo->_mmgr, ptr);
#endif
}

/* ========================================================================= */

enum logfd_flag_t
{
	LOGFD_TTY = (1 << 0),
	LOGFD_OPENED_HERE = (1 << 1)
};

#if defined(EMSCRIPTEN)
EM_JS(int, write_all, (int, const moo_bch_t* ptr, moo_oow_t len), {
	// UTF8ToString() doesn't handle a null byte in the middle of an array.
	// Use the heap memory and pass the right portion to UTF8Decoder.
	//console.log ("%s", UTF8ToString(ptr, len));
	console.log ("%s", UTF8Decoder.decode(HEAPU8.subarray(ptr, ptr + len)));
	return 0;
});

#else

static int write_all (int fd, const moo_bch_t* ptr, moo_oow_t len)
{
#if defined(EMSCRIPTEN)
	EM_ASM_ ({
		// UTF8ToString() doesn't handle a null byte in the middle of an array.
		// Use the heap memory and pass the right portion to UTF8Decoder.
		//console.log ("%s", UTF8ToString($0, $1));
		console.log ("%s", UTF8Decoder.decode(HEAPU8.subarray($0, $0 + $1)));
	}, ptr, len);
#else
	while (len > 0)
	{
		moo_ooi_t wr;

		wr = write(fd, ptr, len);

		if (wr <= -1)
		{
		#if defined(EAGAIN) && defined(EWOULDBLOCK) && (EAGAIN == EWOULDBLOCK)
			if (errno == EAGAIN) continue;
		#else
			#if defined(EAGAIN)
			if (errno == EAGAIN) continue;
			#elif defined(EWOULDBLOCK)
			if (errno == EWOULDBLOCK) continue;
			#endif
		#endif

		#if defined(EINTR)
			/* TODO: would this interfere with non-blocking nature of this VM? */
			if (errno == EINTR) continue;
		#endif
			return -1;
		}

		ptr += wr;
		len -= wr;
	}
#endif

	return 0;
}
#endif

static int write_log (moo_t* moo, int fd, const moo_bch_t* ptr, moo_oow_t len)
{
	xtn_t* xtn = GET_XTN(moo);

	while (len > 0)
	{
		if (xtn->log.out.len > 0)
		{
			moo_oow_t rcapa, cplen;

			rcapa = MOO_COUNTOF(xtn->log.out.buf) - xtn->log.out.len;
			cplen = (len >= rcapa)? rcapa: len;

			MOO_MEMCPY (&xtn->log.out.buf[xtn->log.out.len], ptr, cplen);
			xtn->log.out.len += cplen;
			ptr += cplen;
			len -= cplen;

			if (xtn->log.out.len >= MOO_COUNTOF(xtn->log.out.buf))
			{
				int n;
				n = write_all(fd, xtn->log.out.buf, xtn->log.out.len);
				xtn->log.out.len = 0;
				if (n <= -1) return -1;
			}
		}
		else
		{
			moo_oow_t rcapa;

			rcapa = MOO_COUNTOF(xtn->log.out.buf);
			if (len >= rcapa)
			{
				if (write_all(fd, ptr, rcapa) <= -1) return -1;
				ptr += rcapa;
				len -= rcapa;
			}
			else
			{
				MOO_MEMCPY (xtn->log.out.buf, ptr, len);
				xtn->log.out.len += len;
				ptr += len;
				len -= len;
			}
		}
	}

	return 0;
}

static void flush_log (moo_t* moo, int fd)
{
	xtn_t* xtn = GET_XTN(moo);
	if (xtn->log.out.len > 0)
	{
		write_all (fd, xtn->log.out.buf, xtn->log.out.len);
		xtn->log.out.len = 0;
	}
}

static void log_write (moo_t* moo, moo_bitmask_t mask, const moo_ooch_t* msg, moo_oow_t len)
{
	xtn_t* xtn = GET_XTN(moo);
	moo_bch_t buf[256];
	moo_oow_t ucslen, bcslen, msgidx;
	int n, logfd;

	if (mask & MOO_LOG_STDERR) logfd = 2;
	else if (mask & MOO_LOG_STDOUT) logfd = 1;
	else
	{
		logfd = xtn->log.fd;
#if !defined(EMSCRIPTEN)
		if (logfd <= -1) return;
#endif
	}

/* TODO: beautify the log message.
 *       do classification based on mask. */
	if (!(mask & (MOO_LOG_STDOUT | MOO_LOG_STDERR)))
	{
	#if defined(_WIN32)
		TIME_ZONE_INFORMATION tzi;
		SYSTEMTIME now;
	#else
		time_t now;
		struct tm tm, *tmp;
	#endif
	#if defined(MOO_OOCH_IS_UCH)
		char ts[32 * MOO_BCSIZE_MAX];
	#else
		char ts[32];
	#endif
		moo_oow_t tslen;

	#if defined(_WIN32)
		/* %z for strftime() in win32 seems to produce a long non-numeric timezone name.
		 * i don't use strftime() for time formatting. */
		GetLocalTime (&now);
		if (GetTimeZoneInformation(&tzi) != TIME_ZONE_ID_INVALID)
		{
			tzi.Bias = -tzi.Bias;
			tslen = sprintf(ts, "%04d-%02d-%02d %02d:%02d:%02d %+02.2d%02.2d ",
				(int)now.wYear, (int)now.wMonth, (int)now.wDay,
				(int)now.wHour, (int)now.wMinute, (int)now.wSecond,
				(int)(tzi.Bias / 60), (int)(tzi.Bias % 60));
		}
		else
		{
			tslen = sprintf(ts, "%04d-%02d-%02d %02d:%02d:%02d ",
				(int)now.wYear, (int)now.wMonth, (int)now.wDay,
				(int)now.wHour, (int)now.wMinute, (int)now.wSecond);
		}
	#elif defined(__OS2__)
		now = time(MOO_NULL);

		#if defined(__WATCOMC__)
		tmp = _localtime(&now, &tm);
		#else
		tmp = localtime(&now);
		#endif

		#if defined(__BORLANDC__)
		/* the borland compiler doesn't handle %z properly - it showed 00 all the time */
		tslen = strftime(ts, sizeof(ts), "%Y-%m-%d %H:%M:%S %Z ", tmp);
		#else
		tslen = strftime(ts, sizeof(ts), "%Y-%m-%d %H:%M:%S %z ", tmp);
		#endif
		if (tslen == 0)
		{
			tslen = sprintf(ts, "%04d-%02d-%02d %02d:%02d:%02d ", tmp->tm_year + 1900, tmp->tm_mon + 1, tmp->tm_mday, tmp->tm_hour, tmp->tm_min, tmp->tm_sec);
		}

	#elif defined(__DOS__)
		now = time(MOO_NULL);
		tmp = localtime(&now);
		/* since i know that %z/%Z is not available in strftime, i switch to sprintf immediately */
		tslen = sprintf(ts, "%04d-%02d-%02d %02d:%02d:%02d ", tmp->tm_year + 1900, tmp->tm_mon + 1, tmp->tm_mday, tmp->tm_hour, tmp->tm_min, tmp->tm_sec);
	#else
		now = time(MOO_NULL);
		#if defined(HAVE_LOCALTIME_R)
		tmp = localtime_r(&now, &tm);
		#else
		tmp = localtime(&now);
		#endif

		#if defined(HAVE_STRFTIME_SMALL_Z)
		tslen = strftime(ts, sizeof(ts), "%Y-%m-%d %H:%M:%S %z ", tmp);
		#else
		tslen = strftime(ts, sizeof(ts), "%Y-%m-%d %H:%M:%S %Z ", tmp);
		#endif
		if (tslen == 0)
		{
			tslen = sprintf(ts, "%04d-%02d-%02d %02d:%02d:%02d ", tmp->tm_year + 1900, tmp->tm_mon + 1, tmp->tm_mday, tmp->tm_hour, tmp->tm_min, tmp->tm_sec);
		}
	#endif

	#if defined(MOO_OOCH_IS_UCH)
		if (moo_getcmgr(moo) != xtn->log_cmgr)
		{
			moo_uch_t tsu[32];
			moo_oow_t tsulen;

			/* the timestamp is likely to contain simple ascii characters only.
			 * conversion is not likely to fail regardless of encodings.
			 * so i don't check errors here */
			tsulen = MOO_COUNTOF(tsu);
			moo_convbtooochars (moo, ts, &tslen, tsu, &tsulen);
			tslen = MOO_COUNTOF(ts);
			moo_conv_uchars_to_bchars_with_cmgr (tsu, &tsulen, ts, &tslen, xtn->log_cmgr);
		}
	#endif

		write_log (moo, logfd, ts, tslen);
	}

	if (logfd == xtn->log.fd && (xtn->log.fd_flag & LOGFD_TTY))
	{
		if (mask & MOO_LOG_FATAL) write_log (moo, logfd, "\x1B[1;31m", 7);
		else if (mask & MOO_LOG_ERROR) write_log (moo, logfd, "\x1B[1;32m", 7);
		else if (mask & MOO_LOG_WARN) write_log (moo, logfd, "\x1B[1;33m", 7);
	}

#if defined(MOO_OOCH_IS_UCH)
	msgidx = 0;
	while (len > 0)
	{
		ucslen = len;
		bcslen = MOO_COUNTOF(buf);

		/*n = moo_convootobchars(moo, &msg[msgidx], &ucslen, buf, &bcslen);*/
		n = moo_conv_uchars_to_bchars_with_cmgr(&msg[msgidx], &ucslen, buf, &bcslen, xtn->log_cmgr);
		if (n == 0 || n == -2)
		{
			/* n = 0:
			 *   converted all successfully
			 * n == -2:
			 *    buffer not sufficient. not all got converted yet.
			 *    write what have been converted this round. */

			MOO_ASSERT (moo, ucslen > 0); /* if this fails, the buffer size must be increased */

			/* attempt to write all converted characters */
			if (write_log(moo, logfd, buf, bcslen) <= -1) break;

			if (n == 0) break;
			else
			{
				msgidx += ucslen;
				len -= ucslen;
			}
		}
		else if (n <= -1)
		{
			/* conversion error but i just stop here but don't treat it as a hard error. */
			break;
		}
	}
#else
	write_log (moo, logfd, msg, len);
#endif

	if (logfd == xtn->log.fd && (xtn->log.fd_flag & LOGFD_TTY))
	{
		if (mask & (MOO_LOG_FATAL | MOO_LOG_ERROR | MOO_LOG_WARN)) write_log (moo, logfd, "\x1B[0m", 4);
	}

	flush_log (moo, logfd);
}

/* ========================================================================= */

static moo_errnum_t errno_to_errnum (int errcode)
{
	switch (errcode)
	{
		case ENOMEM: return MOO_ESYSMEM;
		case EINVAL: return MOO_EINVAL;

	#if defined(EBUSY)
		case EBUSY: return MOO_EBUSY;
	#endif
		case EACCES: return MOO_EACCES;
	#if defined(EPERM)
		case EPERM: return MOO_EPERM;
	#endif
	#if defined(ENOTDIR)
		case ENOTDIR: return MOO_ENOTDIR;
	#endif
		case ENOENT: return MOO_ENOENT;
	#if defined(EEXIST)
		case EEXIST: return MOO_EEXIST;
	#endif
	#if defined(EINTR)
		case EINTR:  return MOO_EINTR;
	#endif

	#if defined(EPIPE)
		case EPIPE:  return MOO_EPIPE;
	#endif

	#if defined(EAGAIN) && defined(EWOULDBLOCK) && (EAGAIN != EWOULDBLOCK)
		case EAGAIN:
		case EWOULDBLOCK: return MOO_EAGAIN;
	#elif defined(EAGAIN)
		case EAGAIN: return MOO_EAGAIN;
	#elif defined(EWOULDBLOCK)
		case EWOULDBLOCK: return MOO_EAGAIN;
	#endif

	#if defined(EBADF)
		case EBADF: return MOO_EBADHND;
	#endif

	#if defined(EIO)
		case EIO: return MOO_EIOERR;
	#endif

		default: return MOO_ESYSERR;
	}
}

#if defined(_WIN32)
static moo_errnum_t winerr_to_errnum (DWORD errcode)
{
	switch (errcode)
	{
		case ERROR_NOT_ENOUGH_MEMORY:
		case ERROR_OUTOFMEMORY:
			return MOO_ESYSMEM;

		case ERROR_INVALID_PARAMETER:
		case ERROR_INVALID_NAME:
			return MOO_EINVAL;

		case ERROR_INVALID_HANDLE:
			return MOO_EBADHND;

		case ERROR_ACCESS_DENIED:
		case ERROR_SHARING_VIOLATION:
			return MOO_EACCES;

	#if defined(ERROR_IO_PRIVILEGE_FAILED)
		case ERROR_IO_PRIVILEGE_FAILED:
	#endif
		case ERROR_PRIVILEGE_NOT_HELD:
			return MOO_EPERM;

		case ERROR_FILE_NOT_FOUND:
		case ERROR_PATH_NOT_FOUND:
			return MOO_ENOENT;

		case ERROR_ALREADY_EXISTS:
		case ERROR_FILE_EXISTS:
			return MOO_EEXIST;

		case ERROR_BROKEN_PIPE:
			return MOO_EPIPE;

		default:
			return MOO_ESYSERR;
	}
}
#endif

#if defined(__OS2__)
static moo_errnum_t os2err_to_errnum (APIRET errcode)
{
	/* APIRET e */
	switch (errcode)
	{
		case ERROR_NOT_ENOUGH_MEMORY:
			return MOO_ESYSMEM;

		case ERROR_INVALID_PARAMETER:
		case ERROR_INVALID_NAME:
			return MOO_EINVAL;

		case ERROR_INVALID_HANDLE:
			return MOO_EBADHND;

		case ERROR_ACCESS_DENIED:
		case ERROR_SHARING_VIOLATION:
			return MOO_EACCES;

		case ERROR_FILE_NOT_FOUND:
		case ERROR_PATH_NOT_FOUND:
			return MOO_ENOENT;

		case ERROR_ALREADY_EXISTS:
			return MOO_EEXIST;

		/*TODO: add more mappings */
		default:
			return MOO_ESYSERR;
	}
}
#if !defined(TCPV40HDRS)
static moo_errnum_t os2sockerr_to_errnum (int errcode)
{
	switch (errcode)
	{
		case SOCEPERM:  return MOO_EPERM;
		case SOCENOENT: return MOO_ENOENT;
		case SOCEINTR:  return MOO_EINTR;
		case SOCEACCES: return MOO_EACCES;
		case SOCEINVAL: return MOO_EINVAL;
		case SOCENOMEM: return MOO_ESYSMEM;
		case SOCEPIPE:  return MOO_EPIPE;
		default: return MOO_ESYSERR;
	}
}
#endif /* TCPV40HDRS */
#endif /* __OS2__ */

#if defined(macintosh)
static moo_errnum_t macerr_to_errnum (int errcode)
{
	switch (e)
	{
		case notEnoughMemoryErr:
			return MOO_ESYSMEM;
		case paramErr:
			return MOO_EINVAL;

		case qErr: /* queue element not found during deletion */
		case fnfErr: /* file not found */
		case dirNFErr: /* direcotry not found */
		case resNotFound: /* resource not found */
		case resFNotFound: /* resource file not found */
		case nbpNotFound: /* name not found on remove */
			return MOO_ENOENT;

		/*TODO: add more mappings */
		default:
			return MOO_ESYSERR;
	}
}
#endif

static moo_errnum_t syserrstrb (moo_t* moo, int syserr_type, int syserr_code, moo_bch_t* buf, moo_oow_t len)
{
	switch (syserr_type)
	{
		case 2:
		#if defined(__OS2__)
			#if defined(TCPV40HDRS)
			if (buf)
			{
				char tmp[64];
				sprintf (tmp, "socket error %d", (int)syserr_code);
				moo_copy_bcstr (buf, len, tmp);
			}
			return MOO_ESYSERR;
			#else
			/* sock_strerror() available in tcpip32.dll only */
			if (buf) moo_copy_bcstr (buf, len, sock_strerror(syserr_code));
			return os2sockerr_to_errnum(syserr_code);
			#endif
		#endif
			/* fall thru for other platforms */

		case 1:
		#if defined(_WIN32)
			if (buf)
			{
				DWORD rc;
				rc = FormatMessageA (
					FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS,
					NULL, syserr_code, MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT),
					buf, len, MOO_NULL
				);
				while (rc > 0 && buf[rc - 1] == '\r' || buf[rc - 1] == '\n') buf[--rc] = '\0';
			}
			return winerr_to_errnum(syserr_code);
		#elif defined(__OS2__)
			if (buf)
			{
				char tmp[64];
				sprintf (tmp, "system error %d", (int)syserr_code);
				hcl_copy_bcstr (buf, len, tmp);
			}
			return os2err_to_errnum(syserr_code);
		#elif defined(macintosh)
			/* TODO: convert code to string */
			if (buf) moo_copy_bcstr (buf, len, "system error");
			return macerr_to_errnum(syserr_code);
		#else
			/* in other systems, errno is still the native system error code.
			 * fall thru */
		#endif

		case 0:
		#if defined(HAVE_STRERROR_R)
			if (buf) strerror_r (syserr_code, buf, len);
		#else
			/* this is not thread safe */
			if (buf) moo_copy_bcstr (buf, len, strerror(syserr_code));
		#endif
			return errno_to_errnum(syserr_code);
	}

	if (buf) moo_copy_bcstr (buf, len, "system error");
	return MOO_ESYSERR;
}

/* ========================================================================= */

#if defined(MOO_BUILD_RELEASE)

static void assert_fail (moo_t* moo, const moo_bch_t* expr, const moo_bch_t* file, moo_oow_t line)
{
	/* do nothing */
}

#else /* defined(MOO_BUILD_RELEASE) */

#if defined(MOO_ENABLE_LIBUNWIND)
#include <libunwind.h>
static void backtrace_stack_frames (moo_t* moo)
{
	unw_cursor_t cursor;
	unw_context_t context;
	int n;

	unw_getcontext(&context);
	unw_init_local(&cursor, &context);

	moo_logbfmt (moo, MOO_LOG_UNTYPED | MOO_LOG_DEBUG, "[BACKTRACE]\n");
	for (n = 0; unw_step(&cursor) > 0; n++)
	{
		unw_word_t ip, sp, off;
		char symbol[256];

		unw_get_reg (&cursor, UNW_REG_IP, &ip);
		unw_get_reg (&cursor, UNW_REG_SP, &sp);

		if (unw_get_proc_name(&cursor, symbol, MOO_COUNTOF(symbol), &off))
		{
			moo_copy_bcstr (symbol, MOO_COUNTOF(symbol), "<unknown>");
		}

		moo_logbfmt (moo, MOO_LOG_UNTYPED | MOO_LOG_DEBUG,
			"#%02d ip=0x%*p sp=0x%*p %hs+0x%zu\n",
			n, MOO_SIZEOF(void*) * 2, (void*)ip, MOO_SIZEOF(void*) * 2, (void*)sp, symbol, (moo_oow_t)off);
	}
}
#elif defined(HAVE_BACKTRACE)
#include <execinfo.h>
static void backtrace_stack_frames (moo_t* moo)
{
	void* btarray[128];
	moo_oow_t btsize;
	char** btsyms;

	btsize = backtrace(btarray, MOO_COUNTOF(btarray));
	btsyms = backtrace_symbols(btarray, btsize);
	if (btsyms)
	{
		moo_oow_t i;
		moo_logbfmt (moo, MOO_LOG_UNTYPED | MOO_LOG_DEBUG, "[BACKTRACE]\n");

		for (i = 0; i < btsize; i++)
		{
			moo_logbfmt (moo, MOO_LOG_UNTYPED | MOO_LOG_DEBUG, "  %hs\n", btsyms[i]);
		}
		free (btsyms);
	}
}
#else
static void backtrace_stack_frames (moo_t* moo)
{
	/* do nothing. not supported */
}
#endif /* defined(MOO_ENABLE_LIBUNWIND) */

static void assert_fail (moo_t* moo, const moo_bch_t* expr, const moo_bch_t* file, moo_oow_t line)
{
	moo_logbfmt (moo, MOO_LOG_UNTYPED | MOO_LOG_FATAL, "ASSERTION FAILURE: %hs at %hs:%zu\n", expr, file, line);
	backtrace_stack_frames (moo);

#if defined(_WIN32)
	ExitProcess (249);
#elif defined(__OS2__)
	DosExit (EXIT_PROCESS, 249);
#elif defined(__DOS__)
	{
		union REGS regs;
		regs.h.ah = DOS_EXIT;
		regs.h.al = 249;
		intdos (&regs, &regs);
	}
#elif defined(vms) || defined(__vms)
	lib$stop (SS$_ABORT); /* use SS$_OPCCUS instead? */
	/* this won't be reached since lib$stop() terminates the process */
	sys$exit (SS$_ABORT); /* this condition code can be shown with
	                       * 'show symbol $status' from the command-line. */
#elif defined(macintosh)

	ExitToShell ();

#else

	kill (getpid(), SIGABRT);
	_exit (1);
#endif
}

#endif /* defined(MOO_BUILD_RELEASE) */

/* ========================================================================= */

static void vm_gettime (moo_t* moo, moo_ntime_t* now)
{
#if defined(_WIN32)

	#if defined(_WIN64) || (defined(_WIN32_WINNT) && (_WIN32_WINNT >= 0x0600))
	moo_uint64_t bigsec, bigmsec;
	bigmsec = GetTickCount64();
	#else
	xtn_t* xtn = GET_XTN(moo);
	moo_uint64_t bigsec, bigmsec;
	DWORD msec;

	msec = GetTickCount(); /* this can sustain for 49.7 days */
	if (msec < xtn->tc_last)
	{
		/* i assume the difference is never bigger than 49.7 days */
		/*diff = (MOO_TYPE_MAX(DWORD) - xtn->tc_last) + 1 + msec;*/
		xtn->tc_overflow++;
		bigmsec = ((moo_uint64_t)MOO_TYPE_MAX(DWORD) * xtn->tc_overflow) + msec;
	}
	else bigmsec = msec;
	xtn->tc_last = msec;
	#endif

	bigsec = MOO_MSEC_TO_SEC(bigmsec);
	bigmsec -= MOO_SEC_TO_MSEC(bigsec);
	MOO_INIT_NTIME(now, bigsec, MOO_MSEC_TO_NSEC(bigmsec));

#elif defined(__OS2__)
	xtn_t* xtn = GET_XTN(moo);
	ULONG msec, elapsed;
	moo_ntime_t et;

/* TODO: use DosTmrQueryTime() and DosTmrQueryFreq()? */
	DosQuerySysInfo (QSV_MS_COUNT, QSV_MS_COUNT, &msec, MOO_SIZEOF(msec)); /* milliseconds */

	elapsed = (msec < xtn->tc_last)? (MOO_TYPE_MAX(ULONG) - xtn->tc_last + msec + 1): (msec - xtn->tc_last);
	xtn->tc_last = msec;

	et.sec = MOO_MSEC_TO_SEC(elapsed);
	msec = elapsed - MOO_SEC_TO_MSEC(et.sec);
	et.nsec = MOO_MSEC_TO_NSEC(msec);

	MOO_ADD_NTIME (&xtn->tc_last_ret , &xtn->tc_last_ret, &et);
	*now = xtn->tc_last_ret;

#elif defined(__DOS__) && (defined(_INTELC32_) || defined(__WATCOMC__))
	xtn_t* xtn = GET_XTN(moo);
	clock_t c, elapsed;
	moo_ntime_t et;

	c = clock();
	elapsed = (c < xtn->tc_last)? (MOO_TYPE_MAX(clock_t) - xtn->tc_last + c + 1): (c - xtn->tc_last);
	xtn->tc_last = c;

	et.sec = elapsed / CLOCKS_PER_SEC;
	#if (CLOCKS_PER_SEC == 100)
		et.nsec = MOO_MSEC_TO_NSEC((elapsed % CLOCKS_PER_SEC) * 10);
	#elif (CLOCKS_PER_SEC == 1000)
		et.nsec = MOO_MSEC_TO_NSEC(elapsed % CLOCKS_PER_SEC);
	#elif (CLOCKS_PER_SEC == 1000000L)
		et.nsec = MOO_USEC_TO_NSEC(elapsed % CLOCKS_PER_SEC);
	#elif (CLOCKS_PER_SEC == 1000000000L)
		et.nsec = (elapsed % CLOCKS_PER_SEC);
	#else
	#	error UNSUPPORTED CLOCKS_PER_SEC
	#endif

	MOO_ADD_NTIME (&xtn->tc_last_ret , &xtn->tc_last_ret, &et);
	*now = xtn->tc_last_ret;

#elif defined(macintosh)
	UnsignedWide tick;
	moo_uint64_t tick64;
	Microseconds (&tick);
	tick64 = *(moo_uint64_t*)&tick;
	MOO_INIT_NTIME (now, MOO_USEC_TO_SEC(tick64), MOO_USEC_TO_NSEC(tick64));
#elif defined(HAVE_CLOCK_GETTIME) && defined(CLOCK_MONOTONIC)
	struct timespec ts;
	clock_gettime (CLOCK_MONOTONIC, &ts);
	MOO_INIT_NTIME(now, ts.tv_sec, ts.tv_nsec);
#elif defined(HAVE_CLOCK_GETTIME) && defined(CLOCK_REALTIME)
	struct timespec ts;
	clock_gettime (CLOCK_REALTIME, &ts);
	MOO_INIT_NTIME(now, ts.tv_sec, ts.tv_nsec);
#else
	struct timeval tv;
	gettimeofday (&tv, MOO_NULL);
	MOO_INIT_NTIME(now, tv.tv_sec, MOO_USEC_TO_NSEC(tv.tv_usec));
#endif
}

/* ========================================================================= */

static int _add_poll_fd (moo_t* moo, int fd, int event_mask)
{
#if defined(USE_DEVPOLL)
	xtn_t* xtn = GET_XTN(moo);
	struct pollfd ev;

	MOO_ASSERT (moo, xtn->ep >= 0);
	ev.fd = fd;
	ev.events = event_mask;
	ev.revents = 0;
	if (write(xtn->ep, &ev, MOO_SIZEOF(ev)) != MOO_SIZEOF(ev))
	{
		moo_seterrwithsyserr (moo, 0, errno);
		MOO_DEBUG2 (moo, "Cannot add file descriptor %d to devpoll - %hs\n", fd, strerror(errno));
		return -1;
	}

	return 0;

#elif defined(USE_KQUEUE)
	xtn_t* xtn = GET_XTN(moo);
	struct kevent ev;
	moo_oow_t rindex, roffset;
	moo_oow_t rv = 0;

	rindex = (moo_oow_t)fd / (MOO_BITSOF(moo_oow_t) >> 1);
	roffset = ((moo_oow_t)fd << 1) % MOO_BITSOF(moo_oow_t);

	if (rindex >= xtn->ev.reg.capa)
	{
		moo_oow_t* tmp;
		moo_oow_t newcapa;

		MOO_STATIC_ASSERT (MOO_SIZEOF(*tmp) == MOO_SIZEOF(*xtn->ev.reg.ptr));

		newcapa = rindex + 1;
		newcapa = MOO_ALIGN_POW2(newcapa, 16);

		tmp = (moo_oow_t*)moo_reallocmem(moo, xtn->ev.reg.ptr, newcapa * MOO_SIZEOF(*tmp));
		if (!tmp)
		{
			const moo_ooch_t* oldmsg = moo_backuperrmsg(moo);
			moo_seterrbfmt (moo, MOO_ESYSERR, "unable to add file descriptor %d to kqueue - %js", fd, oldmsg);
			MOO_DEBUG1 (moo, "%js", moo_geterrmsg(moo));
			return -1;
		}

		MOO_MEMSET (&tmp[xtn->ev.reg.capa], 0, newcapa - xtn->ev.reg.capa);
		xtn->ev.reg.ptr = tmp;
		xtn->ev.reg.capa = newcapa;
	}

	if (event_mask & XPOLLIN)
	{
		/*EV_SET (&ev, fd, EVFILT_READ, EV_ADD | EV_CLEAR, 0, 0, 0);*/
		MOO_MEMSET (&ev, 0, MOO_SIZEOF(ev));
		ev.ident = fd;
		ev.flags = EV_ADD;
	#if defined(USE_THREAD)
		ev.flags |= EV_CLEAR; /* EV_CLEAR for edge trigger? */
	#endif
		ev.filter = EVFILT_READ;
		if (kevent(xtn->ep, &ev, 1, MOO_NULL, 0, MOO_NULL) == -1)
		{
			moo_seterrwithsyserr (moo, 0, errno);
			MOO_DEBUG2 (moo, "Cannot add file descriptor %d to kqueue for read - %hs\n", fd, strerror(errno));
			return -1;
		}

		rv |= 1;
	}
	if (event_mask & XPOLLOUT)
	{
		/*EV_SET (&ev, fd, EVFILT_WRITE, EV_ADD | EV_CLEAR, 0, 0, 0);*/
		MOO_MEMSET (&ev, 0, MOO_SIZEOF(ev));
		ev.ident = fd;
		ev.flags = EV_ADD;
	#if defined(USE_THREAD)
		ev.flags |= EV_CLEAR; /* EV_CLEAR for edge trigger? */
	#endif
		ev.filter = EVFILT_WRITE;
		if (kevent(xtn->ep, &ev, 1, MOO_NULL, 0, MOO_NULL) == -1)
		{
			moo_seterrwithsyserr (moo, 0, errno);
			MOO_DEBUG2 (moo, "Cannot add file descriptor %d to kqueue for write - %hs\n", fd, strerror(errno));

			if (event_mask & XPOLLIN)
			{
				MOO_MEMSET (&ev, 0, MOO_SIZEOF(ev));
				ev.ident = fd;
				ev.flags = EV_DELETE;
				ev.filter = EVFILT_READ;
				kevent(xtn->ep, &ev, 1, MOO_NULL, 0, MOO_NULL);
			}
			return -1;
		}

		rv |= 2;
	}

	MOO_SETBITS (moo_oow_t, xtn->ev.reg.ptr[rindex], roffset, 2, rv);
	return 0;

#elif defined(USE_EPOLL)
	xtn_t* xtn = GET_XTN(moo);
	struct epoll_event ev;

	MOO_ASSERT (moo, xtn->ep >= 0);
	MOO_MEMSET (&ev, 0, MOO_SIZEOF(ev));
	ev.events = event_mask;
	#if defined(USE_THREAD) && defined(EPOLLET)
	/* epoll_wait may return again if the worker thread consumes events.
	 * switch to level-trigger. */
	/* TODO: verify if EPOLLLET is desired */
	ev.events |= EPOLLET/*  | EPOLLRDHUP | EPOLLHUP */;
	#endif
	/*ev.data.ptr = (void*)event_data;*/
	ev.data.fd = fd;
	if (epoll_ctl(xtn->ep, EPOLL_CTL_ADD, fd, &ev) == -1)
	{
		moo_seterrwithsyserr (moo, 0, errno);
		MOO_DEBUG2 (moo, "Cannot add file descriptor %d to epoll - %hs\n", fd, strerror(errno));
		return -1;
	}
	return 0;

#elif defined(USE_POLL)
	xtn_t* xtn = GET_XTN(moo);

	MUTEX_LOCK (&xtn->ev.reg.pmtx);
	if (xtn->ev.reg.len >= xtn->ev.reg.capa)
	{
		struct pollfd* tmp, * tmp2;
		moo_oow_t newcapa;

		newcapa = MOO_ALIGN_POW2 (xtn->ev.reg.len + 1, 256);
		tmp = (struct pollfd*)moo_reallocmem(moo, xtn->ev.reg.ptr, newcapa * MOO_SIZEOF(*tmp));
		tmp2 = (struct pollfd*)moo_reallocmem(moo, xtn->ev.buf, newcapa * MOO_SIZEOF(*tmp2));
		if (!tmp || !tmp2)
		{
			MOO_DEBUG2 (moo, "Cannot add file descriptor %d to poll - %hs\n", fd, strerror(errno));
			MUTEX_UNLOCK (&xtn->ev.reg.pmtx);
			if (tmp) moo_freemem (moo, tmp);
			return -1;
		}

		xtn->ev.reg.ptr = tmp;
		xtn->ev.reg.capa = newcapa;

		xtn->ev.buf = tmp2;
	}

	xtn->ev.reg.ptr[xtn->ev.reg.len].fd = fd;
	xtn->ev.reg.ptr[xtn->ev.reg.len].events = event_mask;
	xtn->ev.reg.ptr[xtn->ev.reg.len].revents = 0;
	xtn->ev.reg.len++;
	MUTEX_UNLOCK (&xtn->ev.reg.pmtx);

	return 0;

#elif defined(USE_SELECT)
	xtn_t* xtn = GET_XTN(moo);

	MUTEX_LOCK (&xtn->ev.reg.smtx);
	if (event_mask & XPOLLIN)
	{
		FD_SET (fd, &xtn->ev.reg.rfds);
		if (fd > xtn->ev.reg.maxfd) xtn->ev.reg.maxfd = fd;
	}
	if (event_mask & XPOLLOUT)
	{
		FD_SET (fd, &xtn->ev.reg.wfds);
		if (fd > xtn->ev.reg.maxfd) xtn->ev.reg.maxfd = fd;
	}
	MUTEX_UNLOCK (&xtn->ev.reg.smtx);

	return 0;

#else

	MOO_DEBUG1 (moo, "Cannot add file descriptor %d to poll - not implemented\n", fd);
	moo_seterrnum (moo, MOO_ENOIMPL);
	return -1;
#endif

}

static int _del_poll_fd (moo_t* moo, int fd)
{

#if defined(USE_DEVPOLL)
	xtn_t* xtn = GET_XTN(moo);
	struct pollfd ev;

	MOO_ASSERT (moo, xtn->ep >= 0);
	ev.fd = fd;
	ev.events = POLLREMOVE;
	ev.revents = 0;
	if (write(xtn->ep, &ev, MOO_SIZEOF(ev)) != MOO_SIZEOF(ev))
	{
		moo_seterrwithsyserr (moo, 0, errno);
		MOO_DEBUG2 (moo, "Cannot remove file descriptor %d from devpoll - %hs\n", fd, strerror(errno));
		return -1;
	}

	return 0;

#elif defined(USE_KQUEUE)
	xtn_t* xtn = GET_XTN(moo);
	moo_oow_t rindex, roffset;
	int rv;
	struct kevent ev;

	rindex = (moo_oow_t)fd / (MOO_BITSOF(moo_oow_t) >> 1);
	roffset = ((moo_oow_t)fd << 1) % MOO_BITSOF(moo_oow_t);

	if (rindex >= xtn->ev.reg.capa)
	{
		moo_seterrbfmt (moo, MOO_EINVAL, "unknown file descriptor %d", fd);
		MOO_DEBUG2 (moo, "Cannot remove file descriptor %d from kqueue - %js\n", fd, moo_geterrmsg(moo));
		return -1;
	};

	rv = MOO_GETBITS (moo_oow_t, xtn->ev.reg.ptr[rindex], roffset, 2);

	if (rv & 1)
	{
		/*EV_SET (&ev, fd, EVFILT_READ, EV_DELETE, 0, 0, 0);*/
		MOO_MEMSET (&ev, 0, MOO_SIZEOF(ev));
		ev.ident = fd;
		ev.flags = EV_DELETE;
		ev.filter = EVFILT_READ;
		kevent(xtn->ep, &ev, 1, MOO_NULL, 0, MOO_NULL);
		/* no error check for now */
	}

	if (rv & 2)
	{
		/*EV_SET (&ev, fd, EVFILT_WRITE, EV_DELETE, 0, 0, 0);*/
		MOO_MEMSET (&ev, 0, MOO_SIZEOF(ev));
		ev.ident = fd;
		ev.flags = EV_DELETE;
		ev.filter = EVFILT_WRITE;
		kevent(xtn->ep, &ev, 1, MOO_NULL, 0, MOO_NULL);
		/* no error check for now */
	}

	MOO_SETBITS (moo_oow_t, xtn->ev.reg.ptr[rindex], roffset, 2, 0);
	return 0;

#elif defined(USE_EPOLL)
	xtn_t* xtn = GET_XTN(moo);
	struct epoll_event ev;

	MOO_ASSERT (moo, xtn->ep >= 0);
	MOO_MEMSET (&ev, 0, MOO_SIZEOF(ev));
	if (epoll_ctl(xtn->ep, EPOLL_CTL_DEL, fd, &ev) == -1)
	{
		moo_seterrwithsyserr (moo, 0, errno);
		MOO_DEBUG2 (moo, "Cannot remove file descriptor %d from epoll - %hs\n", fd, strerror(errno));
		return -1;
	}
	return 0;

#elif defined(USE_POLL)
	xtn_t* xtn = GET_XTN(moo);
	moo_oow_t i;

	/* TODO: performance boost. no linear search */
	MUTEX_LOCK (&xtn->ev.reg.pmtx);
	for (i = 0; i < xtn->ev.reg.len; i++)
	{
		if (xtn->ev.reg.ptr[i].fd == fd)
		{
			xtn->ev.reg.len--;
			MOO_MEMMOVE (&xtn->ev.reg.ptr[i], &xtn->ev.reg.ptr[i+1], (xtn->ev.reg.len - i) * MOO_SIZEOF(*xtn->ev.reg.ptr));
			MUTEX_UNLOCK (&xtn->ev.reg.pmtx);
			return 0;
		}
	}
	MUTEX_UNLOCK (&xtn->ev.reg.pmtx);


	MOO_DEBUG1 (moo, "Cannot remove file descriptor %d from poll - not found\n", fd);
	moo_seterrnum (moo, MOO_ENOENT);
	return -1;

#elif defined(USE_SELECT)
	xtn_t* xtn = GET_XTN(moo);

	MUTEX_LOCK (&xtn->ev.reg.smtx);
	FD_CLR (fd, &xtn->ev.reg.rfds);
	FD_CLR (fd, &xtn->ev.reg.wfds);
	if (fd >= xtn->ev.reg.maxfd)
	{
		int i;
		/* TODO: any way to make this search faster or to do without the search like this */
		for (i = fd - 1; i >= 0; i--)
		{
			if (FD_ISSET(i, &xtn->ev.reg.rfds) || FD_ISSET(i, &xtn->ev.reg.wfds)) break;
		}
		xtn->ev.reg.maxfd = i;
	}
	MUTEX_UNLOCK (&xtn->ev.reg.smtx);

	return 0;

#else

	MOO_DEBUG1 (moo, "Cannot remove file descriptor %d from poll - not implemented\n", fd);
	moo_seterrnum (moo, MOO_ENOIMPL);
	return -1;
#endif
}


static int _mod_poll_fd (moo_t* moo, int fd, int event_mask)
{
#if defined(USE_DEVPOLL)

	if (_del_poll_fd (moo, fd) <= -1) return -1;

	if (_add_poll_fd (moo, fd, event_mask) <= -1)
	{
		/* TODO: any good way to rollback successful deletion? */
		return -1;
	}

	return 0;
#elif defined(USE_KQUEUE)
	xtn_t* xtn = GET_XTN(moo);
	moo_oow_t rindex, roffset;
	int rv, newrv = 0;
	struct kevent ev;

	rindex = (moo_oow_t)fd / (MOO_BITSOF(moo_oow_t) >> 1);
	roffset = ((moo_oow_t)fd << 1) % MOO_BITSOF(moo_oow_t);

	if (rindex >= xtn->ev.reg.capa)
	{
		moo_seterrbfmt (moo, MOO_EINVAL, "unknown file descriptor %d", fd);
		MOO_DEBUG2 (moo, "Cannot modify file descriptor %d in kqueue - %js\n", fd, moo_geterrmsg(moo));
		return -1;
	};

	rv = MOO_GETBITS(moo_oow_t, xtn->ev.reg.ptr[rindex], roffset, 2);

	if (rv & 1)
	{
		if (!(event_mask & XPOLLIN))
		{
			MOO_MEMSET (&ev, 0, MOO_SIZEOF(ev));
			ev.ident = fd;
			ev.flags = EV_DELETE;
			ev.filter = EVFILT_READ;
			if (kevent(xtn->ep, &ev, 1, MOO_NULL, 0, MOO_NULL) == -1) goto kqueue_syserr;

			newrv &= ~1;
		}
	}
	else
	{
		if (event_mask & XPOLLIN)
		{
			MOO_MEMSET (&ev, 0, MOO_SIZEOF(ev));
			ev.ident = fd;
			ev.flags = EV_ADD;
		#if defined(USE_THREAD)
			ev.flags |= EV_CLEAR; /* EV_CLEAR for edge trigger? */
		#endif
			ev.filter = EVFILT_READ;
			if (kevent(xtn->ep, &ev, 1, MOO_NULL, 0, MOO_NULL) == -1) goto kqueue_syserr;

			newrv |= 1;
		}
	}

	if (rv & 2)
	{
		if (!(event_mask & XPOLLOUT))
		{
			MOO_MEMSET (&ev, 0, MOO_SIZEOF(ev));
			ev.ident = fd;
			ev.flags = EV_DELETE;
			ev.filter = EVFILT_WRITE;
			/* there is no operation rollback for the (rv & 1) case.
			 * the rollback action may fail again even if i try it */
			if (kevent(xtn->ep, &ev, 1, MOO_NULL, 0, MOO_NULL) == -1) goto kqueue_syserr;

			newrv &= ~2;
		}
	}
	else
	{
		if (event_mask & XPOLLOUT)
		{
			MOO_MEMSET (&ev, 0, MOO_SIZEOF(ev));
			ev.ident = fd;
			ev.flags = EV_ADD;
		#if defined(USE_THREAD)
			ev.flags |= EV_CLEAR; /* EV_CLEAR for edge trigger? */
		#endif
			ev.filter = EVFILT_WRITE;

			/* there is no operation rollback for the (rv & 1) case.
			 * the rollback action may fail again even if i try it */
			if (kevent(xtn->ep, &ev, 1, MOO_NULL, 0, MOO_NULL) == -1) goto kqueue_syserr;

			newrv |= 2;
		}
	}

	MOO_SETBITS (moo_oow_t, xtn->ev.reg.ptr[rindex], roffset, 2, newrv);
	return 0;

kqueue_syserr:
	moo_seterrwithsyserr (moo, 0, errno);
	MOO_DEBUG2 (moo, "Cannot modify file descriptor %d in kqueue - %hs\n", fd, strerror(errno));
	return -1;

#elif defined(USE_EPOLL)
	xtn_t* xtn = GET_XTN(moo);
	struct epoll_event ev;

	MOO_ASSERT (moo, xtn->ep >= 0);
	MOO_MEMSET (&ev, 0, MOO_SIZEOF(ev));
	ev.events = event_mask;
	#if defined(USE_THREAD) && defined(EPOLLET)
	/* epoll_wait may return again if the worker thread consumes events.
	 * switch to level-trigger. */
	/* TODO: verify if EPOLLLET is desired */
	ev.events |= EPOLLET;
	#endif
	ev.data.fd = fd;
	if (epoll_ctl(xtn->ep, EPOLL_CTL_MOD, fd, &ev) == -1)
	{
		moo_seterrwithsyserr (moo, 0, errno);
		MOO_DEBUG2 (moo, "Cannot modify file descriptor %d in epoll - %hs\n", fd, strerror(errno));
		return -1;
	}

	return 0;

#elif defined(USE_POLL)

	xtn_t* xtn = GET_XTN(moo);
	moo_oow_t i;

	MUTEX_LOCK (&xtn->ev.reg.pmtx);
	for (i = 0; i < xtn->ev.reg.len; i++)
	{
		if (xtn->ev.reg.ptr[i].fd == fd)
		{
			MOO_MEMMOVE (&xtn->ev.reg.ptr[i], &xtn->ev.reg.ptr[i+1], (xtn->ev.reg.len - i - 1) * MOO_SIZEOF(*xtn->ev.reg.ptr));
			xtn->ev.reg.ptr[i].fd = fd;
			xtn->ev.reg.ptr[i].events = event_mask;
			xtn->ev.reg.ptr[i].revents = 0;
			MUTEX_UNLOCK (&xtn->ev.reg.pmtx);

			return 0;
		}
	}
	MUTEX_UNLOCK (&xtn->ev.reg.pmtx);

	MOO_DEBUG1 (moo, "Cannot modify file descriptor %d in poll - not found\n", fd);
	moo_seterrnum (moo, MOO_ENOENT);
	return -1;

#elif defined(USE_SELECT)

	xtn_t* xtn = GET_XTN(moo);

	MUTEX_LOCK (&xtn->ev.reg.smtx);
	MOO_ASSERT (moo, fd <= xtn->ev.reg.maxfd);

	if (event_mask & XPOLLIN)
		FD_SET (fd, &xtn->ev.reg.rfds);
	else
		FD_CLR (fd, &xtn->ev.reg.rfds);

	if (event_mask & XPOLLOUT)
		FD_SET (fd, &xtn->ev.reg.wfds);
	else
		FD_CLR (fd, &xtn->ev.reg.wfds);
	MUTEX_UNLOCK (&xtn->ev.reg.smtx);

	return 0;

#else
	MOO_DEBUG1 (moo, "Cannot modify file descriptor %d in poll - not implemented\n", fd);
	moo_seterrnum (moo, MOO_ENOIMPL);
	return -1;
#endif
}

static int vm_muxadd (moo_t* moo, moo_ooi_t io_handle, moo_ooi_t mask)
{
	int event_mask;

	event_mask = 0;
	if (mask & MOO_SEMAPHORE_IO_MASK_INPUT) event_mask |= XPOLLIN;
	if (mask & MOO_SEMAPHORE_IO_MASK_OUTPUT) event_mask |= XPOLLOUT;

	if (event_mask == 0)
	{
		MOO_DEBUG2 (moo, "<vm_muxadd> Invalid semaphore mask %zd on handle %zd\n", mask, io_handle);
		moo_seterrbfmt (moo, MOO_EINVAL, "invalid semaphore mask %zd on handle %zd", mask, io_handle);
		return -1;
	}

	return _add_poll_fd(moo, io_handle, event_mask);
}

static int vm_muxmod (moo_t* moo, moo_ooi_t io_handle, moo_ooi_t mask)
{
	int event_mask;

	event_mask = 0;
	if (mask & MOO_SEMAPHORE_IO_MASK_INPUT) event_mask |= XPOLLIN;
	if (mask & MOO_SEMAPHORE_IO_MASK_OUTPUT) event_mask |= XPOLLOUT;

	if (event_mask == 0)
	{
		MOO_DEBUG2 (moo, "<vm_muxadd> Invalid semaphore mask %zd on handle %zd\n", mask, io_handle);
		moo_seterrbfmt (moo, MOO_EINVAL, "invalid semaphore mask %zd on handle %zd", mask, io_handle);
		return -1;
	}

	return _mod_poll_fd(moo, io_handle, event_mask);
}

static int vm_muxdel (moo_t* moo, moo_ooi_t io_handle)
{
	return _del_poll_fd(moo, io_handle);
}

#if defined(USE_THREAD)
static void* iothr_main (void* arg)
{
	moo_t* moo = (moo_t*)arg;
	xtn_t* xtn = GET_XTN(moo);

	/*while (!moo->abort_req)*/
	while (!xtn->iothr.abort)
	{
		if (xtn->ev.len <= 0) /* TODO: no mutex needed for this check? */
		{
			int n;
		#if defined(USE_DEVPOLL)
			struct dvpoll dvp;
		#elif defined(USE_KQUEUE)
			struct timespec ts;
		#elif defined(USE_POLL)
			moo_oow_t nfds;
		#elif defined(USE_SELECT)
			struct timeval tv;
			fd_set rfds;
			fd_set wfds;
			int maxfd;
		#endif

		poll_for_event:

		#if defined(USE_DEVPOLL)
			dvp.dp_timeout = 10000; /* milliseconds */
			dvp.dp_fds = xtn->ev.buf;
			dvp.dp_nfds = MOO_COUNTOF(xtn->ev.buf);
			n = ioctl (xtn->ep, DP_POLL, &dvp);
		#elif defined(USE_KQUEUE)
			ts.tv_sec = 10;
			ts.tv_nsec = 0;
			n = kevent(xtn->ep, MOO_NULL, 0, xtn->ev.buf, MOO_COUNTOF(xtn->ev.buf), &ts);
			/* n == 0: timeout
			 * n == -1: error */
		#elif defined(USE_EPOLL)
			n = epoll_wait(xtn->ep, xtn->ev.buf, MOO_COUNTOF(xtn->ev.buf), 10000); /* TODO: make this timeout value in the io thread */
		#elif defined(USE_POLL)
			MUTEX_LOCK (&xtn->ev.reg.pmtx);
			MOO_MEMCPY (xtn->ev.buf, xtn->ev.reg.ptr, xtn->ev.reg.len * MOO_SIZEOF(*xtn->ev.buf));
			nfds = xtn->ev.reg.len;
			MUTEX_UNLOCK (&xtn->ev.reg.pmtx);
			n = poll(xtn->ev.buf, nfds, 10000);
			if (n > 0)
			{
				/* compact the return buffer as poll() doesn't */
				moo_oow_t i, j;
				for (i = 0, j = 0; i < nfds && j < n; i++)
				{
					if (xtn->ev.buf[i].revents)
					{
						if (j != i) xtn->ev.buf[j] = xtn->ev.buf[i];
						j++;
					}
				}
				n = j;
			}
		#elif defined(USE_SELECT)
			tv.tv_sec = 10;
			tv.tv_usec = 0;
			MUTEX_LOCK (&xtn->ev.reg.smtx);
			maxfd = xtn->ev.reg.maxfd;
			MOO_MEMCPY (&rfds, &xtn->ev.reg.rfds, MOO_SIZEOF(rfds));
			MOO_MEMCPY (&wfds, &xtn->ev.reg.wfds, MOO_SIZEOF(wfds));
			MUTEX_UNLOCK (&xtn->ev.reg.smtx);
			n = select (maxfd + 1, &rfds, &wfds, MOO_NULL, &tv);
			if (n > 0)
			{
				int fd, count = 0;
				for (fd = 0;  fd <= maxfd; fd++)
				{
					int events = 0;
					if (FD_ISSET(fd, &rfds)) events |= XPOLLIN;
					if (FD_ISSET(fd, &wfds)) events |= XPOLLOUT;

					if (events)
					{
						MOO_ASSERT (moo, count < MOO_COUNTOF(xtn->ev.buf));
						xtn->ev.buf[count].fd = fd;
						xtn->ev.buf[count].events = events;
						count++;
					}
				}

				n = count;
				MOO_ASSERT (moo, n > 0);
			}
		#endif

			pthread_mutex_lock (&xtn->ev.mtx);
			if (n <= -1)
			{
				/* TODO: don't use MOO_DEBUG2. it's not thread safe... */
				/* the following call has a race-condition issue when called in this separate thread */
				/*MOO_DEBUG2 (moo, "Warning: multiplexer wait failure - %d, %hs\n", errno, strerror(errno));*/
			}
			else if (n > 0)
			{
				xtn->ev.len = n;
			}
			pthread_cond_signal (&xtn->ev.cnd2);
			pthread_mutex_unlock (&xtn->ev.mtx);
		}
		else
		{
			/* the event buffer has not been emptied yet */
			struct timespec ts;

			pthread_mutex_lock (&xtn->ev.mtx);
			if (xtn->ev.len <= 0)
			{
				/* it got emptied between the if check and pthread_mutex_lock() above */
				pthread_mutex_unlock (&xtn->ev.mtx);
				goto poll_for_event;
			}

		#if defined(HAVE_CLOCK_GETTIME) && defined(CLOCK_REALTIME)
			clock_gettime (CLOCK_REALTIME, &ts);
		#else
			{
				struct timeval tv;
				gettimeofday (&tv, MOO_NULL);
				ts.tv_sec = tv.tv_sec;
				ts.tv_nsec = MOO_USEC_TO_NSEC(tv.tv_usec);
			}
		#endif
			ts.tv_sec += 10;
			pthread_cond_timedwait (&xtn->ev.cnd, &xtn->ev.mtx, &ts);
			pthread_mutex_unlock (&xtn->ev.mtx);
		}

		/*sched_yield ();*/
	}

	return MOO_NULL;
}
#endif

static void vm_muxwait (moo_t* moo, const moo_ntime_t* dur, moo_vmprim_muxwait_cb_t muxwcb)
{
	xtn_t* xtn = GET_XTN(moo);

#if defined(USE_THREAD)
	int n;

	/* create a thread if mux wait is started at least once. */
	if (!xtn->iothr.up)
	{
		xtn->iothr.up = 1;
		if (pthread_create(&xtn->iothr.thr, MOO_NULL, iothr_main, moo) != 0)
		{
			MOO_LOG2 (moo, MOO_LOG_WARN, "Warning: pthread_create failure - %d, %hs\n", errno, strerror(errno));
			xtn->iothr.up = 0;
/* TODO: switch to the non-threaded mode? */
		}
	}

	if (xtn->iothr.abort) return;

	if (xtn->ev.len <= 0)
	{
		struct timespec ts;
		moo_ntime_t ns;

		if (!dur) return; /* immediate check is requested. and there is no event */

	#if defined(HAVE_CLOCK_GETTIME) && defined(CLOCK_REALTIME)
		clock_gettime (CLOCK_REALTIME, &ts);
		ns.sec = ts.tv_sec;
		ns.nsec = ts.tv_nsec;
	#else
		{
			struct timeval tv;
			gettimeofday (&tv, MOO_NULL);
			ns.sec = tv.tv_sec;
			ns.nsec = MOO_USEC_TO_NSEC(tv.tv_usec);
		}
	#endif
		MOO_ADD_NTIME (&ns, &ns, dur);
		ts.tv_sec = ns.sec;
		ts.tv_nsec = ns.nsec;

		pthread_mutex_lock (&xtn->ev.mtx);
		if (xtn->ev.len <= 0)
		{
			/* the event buffer is still empty */
			pthread_cond_timedwait (&xtn->ev.cnd2, &xtn->ev.mtx, &ts);
		}
		pthread_mutex_unlock (&xtn->ev.mtx);
	}

	n = xtn->ev.len;

	if (n > 0)
	{
		do
		{
			--n;

		#if defined(USE_DEVPOLL)
			if (xtn->ev.buf[n].fd == xtn->iothr.p[0])
		#elif defined(USE_KQUEUE)
			if (xtn->ev.buf[n].ident == xtn->iothr.p[0])
		#elif defined(USE_EPOLL)
			/*if (xtn->ev.buf[n].data.ptr == (void*)MOO_TYPE_MAX(moo_oow_t))*/
			if (xtn->ev.buf[n].data.fd == xtn->iothr.p[0])
		#elif defined(USE_POLL)
			if (xtn->ev.buf[n].fd == xtn->iothr.p[0])
		#elif defined(USE_SELECT)
			if (xtn->ev.buf[n].fd == xtn->iothr.p[0])
		#else
		#	error UNSUPPORTED
		#endif
			{
				moo_uint8_t u8;
				while (read(xtn->iothr.p[0], &u8, MOO_SIZEOF(u8)) > 0)
				{
					/* consume as much as possible */;
					if (u8 == 'Q') xtn->iothr.abort = 1;
				}
			}
			else if (muxwcb)
			{
				int revents;
				moo_ooi_t mask;

			#if defined(USE_DEVPOLL)
				revents = xtn->ev.buf[n].revents;
			#elif defined(USE_KQUEUE)
				if (xtn->ev.buf[n].filter == EVFILT_READ) mask = MOO_SEMAPHORE_IO_MASK_INPUT;
				else if (xtn->ev.buf[n].filter == EVFILT_WRITE) mask = MOO_SEMAPHORE_IO_MASK_OUTPUT;
				else mask = 0;
				goto call_muxwcb_kqueue;
			#elif defined(USE_EPOLL)
				revents = xtn->ev.buf[n].events;
			#elif defined(USE_POLL)
				revents = xtn->ev.buf[n].revents;
			#elif defined(USE_SELECT)
				revents = xtn->ev.buf[n].events;
			#endif

				mask = 0;
				if (revents & XPOLLIN) mask |= MOO_SEMAPHORE_IO_MASK_INPUT;
				if (revents & XPOLLOUT) mask |= MOO_SEMAPHORE_IO_MASK_OUTPUT;
				if (revents & XPOLLERR) mask |= MOO_SEMAPHORE_IO_MASK_ERROR;
				if (revents & XPOLLHUP) mask |= MOO_SEMAPHORE_IO_MASK_HANGUP;

			#if defined(USE_DEVPOLL)
				muxwcb (moo, xtn->ev.buf[n].fd, mask);
			#elif defined(USE_KQUEUE)
			call_muxwcb_kqueue:
				muxwcb (moo, xtn->ev.buf[n].ident, mask);
			#elif defined(USE_EPOLL)
				muxwcb (moo, xtn->ev.buf[n].data.fd, mask);
			#elif defined(USE_POLL)
				muxwcb (moo, xtn->ev.buf[n].fd, mask);
			#elif defined(USE_SELECT)
				muxwcb (moo, xtn->ev.buf[n].fd, mask);
			#else
			#	error UNSUPPORTED
			#endif
			}
		}
		while (n > 0);

		pthread_mutex_lock (&xtn->ev.mtx);
		xtn->ev.len = 0;
		pthread_cond_signal (&xtn->ev.cnd);
		pthread_mutex_unlock (&xtn->ev.mtx);
	}

#else /* USE_THREAD */
	int n;
	#if defined(USE_DEVPOLL)
	int tmout;
	struct dvpoll dvp;
	#elif defined(USE_KQUEUE)
	struct timespec ts;
	#elif defined(USE_EPOLL)
	int tmout;
	#elif defined(USE_POLL)
	int tmout;
	#elif defined(USE_SELECT)
	struct timeval tv;
	fd_set rfds, wfds;
	int maxfd;
	#endif


	#if defined(USE_DEVPOLL)
	tmout = dur? MOO_SECNSEC_TO_MSEC(dur->sec, dur->nsec): 0;

	dvp.dp_timeout = tmout; /* milliseconds */
	dvp.dp_fds = xtn->ev.buf;
	dvp.dp_nfds = MOO_COUNTOF(xtn->ev.buf);
	n = ioctl(xtn->ep, DP_POLL, &dvp);

	#elif defined(USE_KQUEUE)

	if (dur)
	{
		ts.tv_sec = dur->sec;
		ts.tv_nsec = dur->nsec;
	}
	else
	{
		ts.tv_sec = 0;
		ts.tv_nsec = 0;
	}

	n = kevent(xtn->ep, MOO_NULL, 0, xtn->ev.buf, MOO_COUNTOF(xtn->ev.buf), &ts);
	/* n == 0: timeout
	 * n == -1: error */

	#elif defined(USE_EPOLL)
	tmout = dur? MOO_SECNSEC_TO_MSEC(dur->sec, dur->nsec): 0;
	n = epoll_wait(xtn->ep, xtn->ev.buf, MOO_COUNTOF(xtn->ev.buf), tmout);

	#elif defined(USE_POLL)
	tmout = dur? MOO_SECNSEC_TO_MSEC(dur->sec, dur->nsec): 0;
	MOO_MEMCPY (xtn->ev.buf, xtn->ev.reg.ptr, xtn->ev.reg.len * MOO_SIZEOF(*xtn->ev.buf));
	n = poll(xtn->ev.buf, xtn->ev.reg.len, tmout);
	if (n > 0)
	{
		/* compact the return buffer as poll() doesn't */
		moo_oow_t i, j;
		for (i = 0, j = 0; i < xtn->ev.reg.len && j < n; i++)
		{
			if (xtn->ev.buf[i].revents)
			{
				if (j != i) xtn->ev.buf[j] = xtn->ev.buf[i];
				j++;
			}
		}
		n = j;
	}
	#elif defined(USE_SELECT)
	if (dur)
	{
		tv.tv_sec = dur->sec;
		tv.tv_usec = MOO_NSEC_TO_USEC(dur->nsec);
	}
	else
	{
		tv.tv_sec = 0;
		tv.tv_usec = 0;
	}
	maxfd = xtn->ev.reg.maxfd;
	MOO_MEMCPY (&rfds, &xtn->ev.reg.rfds, MOO_SIZEOF(rfds));
	MOO_MEMCPY (&wfds, &xtn->ev.reg.wfds, MOO_SIZEOF(wfds));
	n = select(maxfd + 1, &rfds, &wfds, MOO_NULL, &tv);
	if (n > 0)
	{
		int fd, count = 0;
		for (fd = 0; fd <= maxfd; fd++)
		{
			int events = 0;
			if (FD_ISSET(fd, &rfds)) events |= XPOLLIN;
			if (FD_ISSET(fd, &wfds)) events |= XPOLLOUT;

			if (events)
			{
				MOO_ASSERT (moo, count < MOO_COUNTOF(xtn->ev.buf));
				xtn->ev.buf[count].fd = fd;
				xtn->ev.buf[count].events = events;
				count++;
			}
		}

		n = count;
		MOO_ASSERT (moo, n > 0);
	}
	#endif

	if (n <= -1)
	{
	#if defined(__OS2__)
		moo_seterrwithsyserr (moo, 2, sock_errno());
		MOO_DEBUG2 (moo, "Warning: multiplexer wait failure - %d, %js\n", sock_errno(), moo_geterrmsg(moo));
	#else
		moo_seterrwithsyserr (moo, 0, errno);
		MOO_DEBUG2 (moo, "Warning: multiplexer wait failure - %d, %js\n", errno, moo_geterrmsg(moo));
	#endif
	}
	else
	{
		xtn->ev.len = n;
	}

	/* the muxwcb must be valid all the time in a non-threaded mode */
	MOO_ASSERT (moo, muxwcb != MOO_NULL);

	while (n > 0)
	{
		int revents;
		moo_ooi_t mask;

		--n;

	#if defined(USE_DEVPOLL)
		revents = xtn->ev.buf[n].revents;
	#elif defined(USE_KQUEUE)
		if (xtn->ev.buf[n].filter == EVFILT_READ) mask = MOO_SEMAPHORE_IO_MASK_INPUT;
		else if (xtn->ev.buf[n].filter == EVFILT_WRITE) mask = MOO_SEMAPHORE_IO_MASK_OUTPUT;
		else mask = 0;
		goto call_muxwcb_kqueue;
	#elif defined(USE_EPOLL)
		revents = xtn->ev.buf[n].events;
	#elif defined(USE_POLL)
		revents = xtn->ev.buf[n].revents;
	#elif defined(USE_SELECT)
		revents = xtn->ev.buf[n].events;
	#else
		revents = 0; /* TODO: fake. unsupported but to compile on such an unsupported system.*/
	#endif

		mask = 0;
		if (revents & XPOLLIN) mask |= MOO_SEMAPHORE_IO_MASK_INPUT;
		if (revents & XPOLLOUT) mask |= MOO_SEMAPHORE_IO_MASK_OUTPUT;
		if (revents & XPOLLERR) mask |= MOO_SEMAPHORE_IO_MASK_ERROR;
		if (revents & XPOLLHUP) mask |= MOO_SEMAPHORE_IO_MASK_HANGUP;

	#if defined(USE_DEVPOLL)
		muxwcb (moo, xtn->ev.buf[n].fd, mask);
	#elif defined(USE_KQUEUE)
	call_muxwcb_kqueue:
		muxwcb (moo, xtn->ev.buf[n].ident, mask);
	#elif defined(USE_EPOLL)
		muxwcb (moo, xtn->ev.buf[n].data.fd, mask);
	#elif defined(USE_POLL)
		muxwcb (moo, xtn->ev.buf[n].fd, mask);
	#elif defined(USE_SELECT)
		muxwcb (moo, xtn->ev.buf[n].fd, mask);
	#endif
	}

	xtn->ev.len = 0;
#endif  /* USE_THREAD */
}

#if defined(__DOS__)
#	if defined(_INTELC32_)
	void _halt_cpu (void);
#	elif defined(__WATCOMC__)
	void _halt_cpu (void);
#	pragma aux _halt_cpu = "hlt"
#	endif
#endif

/* ========================================================================= */

static int vm_sleep (moo_t* moo, const moo_ntime_t* dur)
{
	xtn_t* xtn = GET_XTN(moo);

	if (MOO_UNLIKELY(xtn->ev.halting)) return xtn->ev.halting;

#if defined(_WIN32)
	if (xtn->waitable_timer)
	{
		LARGE_INTEGER li;
		li.QuadPart = -(MOO_SECNSEC_TO_NSEC(dur->sec, dur->nsec) / 100); /* in 100 nanoseconds */
		if(SetWaitableTimer(xtn->waitable_timer, &li, 0, MOO_NULL, MOO_NULL, FALSE) == FALSE) goto normal_sleep;
		WaitForSingleObject(xtn->waitable_timer, INFINITE);
	}
	else
	{
	normal_sleep:
		/* fallback to normal Sleep() */
		Sleep (MOO_SECNSEC_TO_MSEC(dur->sec,dur->nsec));
	}
#elif defined(__OS2__)

	/* TODO: in gui mode, this is not a desirable method???
	 *       this must be made event-driven coupled with the main event loop */
	DosSleep (MOO_SECNSEC_TO_MSEC(dur->sec,dur->nsec));

#elif defined(macintosh)

	/* TODO: ... */
#	error NOT IMPLEMENTED

#elif defined(__DOS__) && (defined(_INTELC32_) || defined(__WATCOMC__))
	{
		clock_t c;

		c = clock ();
		c += dur->sec * CLOCKS_PER_SEC;

		#if (CLOCKS_PER_SEC == 100)
			c += MOO_NSEC_TO_MSEC(dur->nsec) / 10;
		#elif (CLOCKS_PER_SEC == 1000)
			c += MOO_NSEC_TO_MSEC(dur->nsec);
		#elif (CLOCKS_PER_SEC == 1000000L)
			c += MOO_NSEC_TO_USEC(dur->nsec);
		#elif (CLOCKS_PER_SEC == 1000000000L)
			c += dur->nsec;
		#else
		#	error UNSUPPORTED CLOCKS_PER_SEC
		#endif

	/* TODO: handle clock overvlow */
	/* TODO: check if there is abortion request or interrupt */
		while (c > clock())
		{
			_halt_cpu();
		}
	}
#elif defined(USE_THREAD)
	/* the sleep callback is called only if there is no IO semaphore
	 * waiting. so i can safely call vm_muxwait() without a muxwait callback
	 * when USE_THREAD is true */
	vm_muxwait (moo, dur, MOO_NULL);

#elif defined(HAVE_NANOSLEEP)
	{
		struct timespec ts;
		ts.tv_sec = dur->sec;
		ts.tv_nsec = dur->nsec;
		nanosleep (&ts, MOO_NULL);
	}

#elif defined(HAVE_USLEEP)
	usleep (MOO_SECNSEC_TO_USEC(dur->sec, dur->nsec));

#else
#	error UNSUPPORTED SLEEP
#endif

	return 0;
}

static moo_ooi_t vm_getsigfd (moo_t* moo)
{
	xtn_t* xtn = GET_XTN(moo);
	return xtn->sigfd.p[0];
}

static int vm_getsig (moo_t* moo, moo_uint8_t* u8)
{
	xtn_t* xtn = GET_XTN(moo);
#if defined(_WIN32)
	/* TODO: can i make the pipe non-block in win32? */
	DWORD navail;
	if (PeekNamedPipe(_get_osfhandle(xtn->sigfd.p[0]), MOO_NULL, 0, MOO_NULL, &navail, MOO_NULL) == 0)
	{
		moo_seterrwithsyserr (moo, 1, GetLastError());
		return -1;
	}
	if (navail <= 0) return 0;
#endif
	if (read(xtn->sigfd.p[0], u8, MOO_SIZEOF(*u8)) == -1)
	{
		if (errno == EINTR || errno == EWOULDBLOCK || errno == EAGAIN) return 0;
		moo_seterrwithsyserr (moo, 0, errno);
		return -1;
	}

	return 1;
}

static int vm_setsig (moo_t* moo, moo_uint8_t u8)
{
	xtn_t* xtn = GET_XTN(moo);
	if (write(xtn->sigfd.p[1], &u8, MOO_SIZEOF(u8)) == -1)
	{
		if (errno == EINTR || errno == EWOULDBLOCK || errno == EAGAIN) return 0;
		moo_seterrwithsyserr (moo, 0, errno);
		return -1;
	}
	return 1;
}
/* ========================================================================= */

#if defined(HAVE_SIGACTION)

typedef struct sig_state_t sig_state_t;
struct sig_state_t
{
	moo_oow_t handler;
	moo_oow_t old_handler;
	sigset_t  old_sa_mask;
	int       old_sa_flags;
};

typedef void (*sig_handler_t) (int sig);

static sig_state_t g_sig_state[MOO_NSIG];

static void dispatch_siginfo (int sig, siginfo_t* si, void* ctx)
{
	if (g_sig_state[sig].handler != (moo_oow_t)SIG_IGN &&
	    g_sig_state[sig].handler != (moo_oow_t)SIG_DFL)
	{
		((sig_handler_t)g_sig_state[sig].handler) (sig);
	}

	if (g_sig_state[sig].old_handler &&
	    g_sig_state[sig].old_handler != (moo_oow_t)SIG_IGN &&
	    g_sig_state[sig].old_handler != (moo_oow_t)SIG_DFL)
	{
		((void(*)(int, siginfo_t*, void*))g_sig_state[sig].old_handler) (sig, si, ctx);
	}
}

static void dispatch_signal (int sig)
{
	if (g_sig_state[sig].handler != (moo_oow_t)SIG_IGN &&
	    g_sig_state[sig].handler != (moo_oow_t)SIG_DFL)
	{
		((sig_handler_t)g_sig_state[sig].handler) (sig);
	}

	if (g_sig_state[sig].old_handler &&
	    g_sig_state[sig].old_handler != (moo_oow_t)SIG_IGN &&
	    g_sig_state[sig].old_handler != (moo_oow_t)SIG_DFL)
	{
		((sig_handler_t)g_sig_state[sig].old_handler) (sig);
	}
}

static int set_signal_handler (int sig, sig_handler_t handler, int extra_flags)
{
	if (g_sig_state[sig].handler)
	{
		/* already set - allow handler change. ignore extra_flags. */
		if (g_sig_state[sig].handler == (moo_oow_t)handler) return -1;
		g_sig_state[sig].handler = (moo_oow_t)handler;
	}
	else
	{
		struct sigaction sa, oldsa;

		if (sigaction(sig, MOO_NULL, &oldsa) == -1) return -1;

		MOO_MEMSET (&sa, 0, MOO_SIZEOF(sa));
		if (oldsa.sa_flags & SA_SIGINFO)
		{
			sa.sa_sigaction = dispatch_siginfo;
			sa.sa_flags = SA_SIGINFO;
		}
		else
		{
			sa.sa_handler = dispatch_signal;
			sa.sa_flags = 0;
		}
		sa.sa_flags |= extra_flags;
		/*sa.sa_flags |= SA_INTERUPT;
		sa.sa_flags |= SA_RESTART;*/
		sigfillset (&sa.sa_mask); /* block all signals while the handler is being executed */

		if (sigaction(sig, &sa, MOO_NULL) == -1) return -1;

		g_sig_state[sig].handler = (moo_oow_t)handler;
		if (oldsa.sa_flags & SA_SIGINFO)
			g_sig_state[sig].old_handler = (moo_oow_t)oldsa.sa_sigaction;
		else
			g_sig_state[sig].old_handler = (moo_oow_t)oldsa.sa_handler;

		g_sig_state[sig].old_sa_mask = oldsa.sa_mask;
		g_sig_state[sig].old_sa_flags = oldsa.sa_flags;
	}

	return 0;
}

static int unset_signal_handler (int sig)
{
	struct sigaction sa;

	if (!g_sig_state[sig].handler) return -1; /* not set */

	MOO_MEMSET (&sa, 0, MOO_SIZEOF(sa));
	sa.sa_mask = g_sig_state[sig].old_sa_mask;
	sa.sa_flags = g_sig_state[sig].old_sa_flags;

	if (sa.sa_flags & SA_SIGINFO)
	{
		sa.sa_sigaction = (void(*)(int,siginfo_t*,void*))g_sig_state[sig].old_handler;
	}
	else
	{
		sa.sa_handler = (sig_handler_t)g_sig_state[sig].old_handler;
	}

	if (sigaction(sig, &sa, MOO_NULL) == -1) return -1;

	g_sig_state[sig].handler = 0;
	/* keep other fields untouched */

	return 0;
}

static int is_signal_handler_set (int sig)
{
	return !!g_sig_state[sig].handler;
}
#endif
/* ========================================================================= */


static MOO_INLINE void abort_all_moos (int signo)
{
	/* TODO: make this atomic */
	if (g_moo)
	{
		moo_t* moo = g_moo;
		do
		{
			xtn_t* xtn = GET_XTN(moo);
			moo_uint8_t u8;
			/*moo_abortstd (moo);*/
			u8 = signo & 0xFF;
			write (xtn->sigfd.p[1], &u8, MOO_SIZEOF(u8));
			moo = xtn->next;
		}
		while (moo);
	}
	/* TODO: make this atomic */
}

static MOO_INLINE void do_nothing (int unused)
{
}

/*#define MOO_TICKER_INTERVAL_USECS 10000*/ /* microseconds. 0.01 seconds */
#define MOO_TICKER_INTERVAL_USECS 20000 /* microseconds. 0.02 seconds. */

static MOO_INLINE void swproc_all_moos (int unused)
{
	/* TODO: make this atomic */
	if (g_moo)
	{
		moo_t* moo = g_moo;
		do
		{
			xtn_t* xtn = GET_XTN(moo);
			if (xtn->rcv_tick) moo_switchprocess (moo);
			moo = xtn->next;
		}
		while (moo);
	}
	/* TODO: make this atomic */
}

#if defined(_WIN32)

static HANDLE msw_tick_timer = MOO_NULL; /*INVALID_HANDLE_VALUE;*/
static int msw_tick_done = 0;

static DWORD WINAPI msw_wait_for_timer_event (LPVOID ctx)
{
	/* I don't think i need to use the waiting timer for this.
	 * a simple loop with sleep inside should also work as i don't do anything
	 * special except waiting for timer expiry.
	 *   while (!msw_tick_done)
	 *   {
	 *       Sleep (...);
	 *       swproc_all_moos();
	 *   }
	 * but never mind for now. let's do it the hard way.
	 */

	msw_tick_timer = CreateWaitableTimer(MOO_NULL, FALSE, MOO_NULL);
	if (msw_tick_timer)
	{
		LARGE_INTEGER li;

		/* lpDueTime in 100 nanoseconds */
		li.QuadPart = -MOO_USEC_TO_NSEC(MOO_TICKER_INTERVAL_USECS) / 100;

	/*#define MSW_TICKER_MANUAL_RESET */
	#if defined(MSW_TICKER_MANUAL_RESET)
		/* if manual resetting is enabled, the reset is done after
		 * swproc_all_moos has been called. so the interval is the
		 * interval specified plus the time taken in swproc_all_moos. */
		SetWaitableTimer (msw_tick_timer, &li, 0, MOO_NULL, MOO_NULL, FALSE);
	#else
		/* with auto reset, the interval is not affected by time taken
		 * in swproc_all_moos() */
		SetWaitableTimer (msw_tick_timer, &li, MOO_USEC_TO_MSEC(MOO_TICKER_INTERVAL_USECS), MOO_NULL, MOO_NULL, FALSE);
	#endif

		while (!msw_tick_done)
		{
			if (WaitForSingleObject(msw_tick_timer, 100000) == WAIT_OBJECT_0)
			{
				swproc_all_moos (0);
			#if defined(MSW_TICKER_MANUAL_RESET)
				SetWaitableTimer (msw_tick_timer, &li, 0, MOO_NULL, MOO_NULL, FALSE);
			#endif
			}
		}

		CancelWaitableTimer (msw_tick_timer);

		CloseHandle (msw_tick_timer);
		msw_tick_timer = MOO_NULL;
	}

	msw_tick_done = 0;
	/*ExitThread (0);*/
	return 0;
}

static MOO_INLINE void start_ticker (void)
{
	HANDLE thr;

	msw_tick_done = 0;

	thr = CreateThread(MOO_NULL, 0, msw_wait_for_timer_event, MOO_NULL, 0, MOO_NULL);
	if (thr)
	{
		SetThreadPriority (thr, THREAD_PRIORITY_HIGHEST);

		/* MSDN - The thread object remains in the system until the thread has terminated
		 *        and all handles to it have been closed through a call to CloseHandle.
		 * it is safe to close the handle here */
		CloseHandle (thr);
	}
}

static MOO_INLINE void stop_ticker (void)
{
	if (msw_tick_timer) CancelWaitableTimer (msw_tick_timer);
	msw_tick_done = 1;
}

#elif defined(__OS2__)

static HEV os2_tick_sem;
static HTIMER os2_tick_timer;
static int os2_tick_done = 0;

static void EXPENTRY os2_wait_for_timer_event (ULONG x)
{
	APIRET rc;
	ULONG count;

	rc = DosCreateEventSem(MOO_NULL, &os2_tick_sem, DC_SEM_SHARED, FALSE);
	if (rc != NO_ERROR)
	{
		goto done;
	}

	rc = DosStartTimer(MOO_USEC_TO_MSEC(MOO_TICKER_INTERVAL_USECS), (HSEM)os2_tick_sem, &os2_tick_timer);
	if (rc != NO_ERROR)
	{
		DosCloseEventSem (os2_tick_sem);
		goto done;
	}

	while (!os2_tick_done)
	{
		rc = DosWaitEventSem(os2_tick_sem, 5000L);
	#if 0
		swproc_all_moos (0);
		DosResetEventSem (os2_tick_sem, &count);
	#else
		DosResetEventSem (os2_tick_sem, &count);
		swproc_all_moos (0);
	#endif
	}

	DosStopTimer (os2_tick_timer);
	DosCloseEventSem (os2_tick_sem);

done:
	os2_tick_timer = NULL;
	os2_tick_sem = NULL;
	os2_tick_done = 0;
	DosExit (EXIT_THREAD, 0);
}

static MOO_INLINE void start_ticker (void)
{
	static TID tid;
	os2_tick_done = 0;
	DosCreateThread (&tid, os2_wait_for_timer_event, 0, 0, 4096);
	/* TODO: Error check */
}

static MOO_INLINE void stop_ticker (void)
{
	if (os2_tick_sem) DosPostEventSem (os2_tick_sem);
	os2_tick_done = 1;
}

#elif defined(__DOS__) && (defined(_INTELC32_) || defined(__WATCOMC__))

#if defined(_INTELC32_)
static void (*dos_prev_timer_intr_handler) (void);
#pragma interrupt(dos_timer_intr_handler)
static void dos_timer_intr_handler (void)
#else
static void (__interrupt *dos_prev_timer_intr_handler) (void);
static void __interrupt dos_timer_intr_handler (void)
#endif
{
	/*
	_XSTACK* stk = (_XSTACK *)_get_stk_frame();
	r = (unsigned short)stk->eax;
	*/

	/* The timer interrupt (normally) occurs 18.2 times per second. */
	swproc_all_moos (0);
	_chain_intr (dos_prev_timer_intr_handler);
}

static MOO_INLINE void start_ticker (void)
{
	dos_prev_timer_intr_handler = _dos_getvect(0x1C);
	_dos_setvect (0x1C, dos_timer_intr_handler);
}

static MOO_INLINE void stop_ticker (void)
{
	_dos_setvect (0x1C, dos_prev_timer_intr_handler);
}

#elif defined(macintosh)

static TMTask mac_tmtask;
static ProcessSerialNumber mac_psn;

/* milliseconds if positive, microseconds(after negation) if negative */
#define TMTASK_DELAY MOO_USEC_TO_MSEC(MOO_TICKER_INTERVAL_USECS)

static pascal void timer_intr_handler (TMTask* task)
{
	swproc_all_moos (0);
	WakeUpProcess (&mac_psn);
	PrimeTime ((QElem*)&mac_tmtask, TMTASK_DELAY);
}

static MOO_INLINE void start_ticker (void)
{
	GetCurrentProcess (&mac_psn);
	MOO_MEMSET (&mac_tmtask, 0, MOO_SIZEOF(mac_tmtask));
	mac_tmtask.tmAddr = NewTimerProc (timer_intr_handler);
	InsXTime ((QElem*)&mac_tmtask);
	PrimeTime ((QElem*)&mac_tmtask, TMTASK_DELAY);
}

static MOO_INLINE void stop_ticker (void)
{
	RmvTime ((QElem*)&mac_tmtask);
	/*DisposeTimerProc (mac_tmtask.tmAddr);*/
}

#elif defined(HAVE_SETITIMER) && defined(SIGVTALRM) && defined(ITIMER_VIRTUAL)

static MOO_INLINE void start_ticker (void)
{
	if (set_signal_handler(SIGVTALRM, swproc_all_moos, SA_RESTART) >= 0)
	{
		struct itimerval itv;
		itv.it_interval.tv_sec = 0;
		itv.it_interval.tv_usec = MOO_TICKER_INTERVAL_USECS;
		itv.it_value.tv_sec = 0;
		itv.it_value.tv_usec = MOO_TICKER_INTERVAL_USECS;
		if (setitimer(ITIMER_VIRTUAL, &itv, MOO_NULL) == -1)
		{
			/* WSL supports ITIMER_VIRTUAL only as of windows 10.0.18362.413.
			   the following is a fallback which will get */
			unset_signal_handler (SIGVTALRM);

		#if defined(SIGALRM) && defined(ITIMER_REAL)
			if (set_signal_handler(SIGALRM, swproc_all_moos, SA_RESTART) >= 0)
			{
				/* i double the interval as ITIMER_REAL is against the wall clock.
				 * if the underlying system is under heavy load, some signals
				 * will get lost */
				itv.it_interval.tv_sec = 0;
				itv.it_interval.tv_usec = MOO_TICKER_INTERVAL_USECS * 2;
				itv.it_value.tv_sec = 0;
				itv.it_value.tv_usec = MOO_TICKER_INTERVAL_USECS * 2;
				setitimer(ITIMER_REAL, &itv, MOO_NULL);
			}
		#endif
		}
	}
}

static MOO_INLINE void stop_ticker (void)
{
	/* ignore the signal fired by the activated timer.
	 * unsetting the signal may cause the program to terminate(default action) */
	if (is_signal_handler_set(SIGVTALRM) && set_signal_handler(SIGVTALRM, SIG_IGN, 0) >= 0)
	{
		struct itimerval itv;
		itv.it_interval.tv_sec = 0;
		itv.it_interval.tv_usec = 0;
		itv.it_value.tv_sec = 0; /* make setitimer() one-shot only */
		itv.it_value.tv_usec = 0;
		setitimer (ITIMER_VIRTUAL, &itv, MOO_NULL);
	}

	#if defined(SIGALRM) && defined(ITIMER_REAL)
	if (is_signal_handler_set(SIGALRM) && set_signal_handler(SIGALRM, SIG_IGN, 0) >= 0)
	{
		struct itimerval itv;
		itv.it_interval.tv_sec = 0;
		itv.it_interval.tv_usec = 0;
		itv.it_value.tv_sec = 0; /* make setitimer() one-shot only */
		itv.it_value.tv_usec = 0;
		setitimer (ITIMER_REAL, &itv, MOO_NULL);
	}
	#endif
}

#else

static pid_t ticker_pid = -1;

static MOO_INLINE void start_ticker (void)
{
	if (set_signal_handler(SIGALRM, swproc_all_moos, SA_RESTART) >= 0)
	{
		ticker_pid = fork();

		if (ticker_pid <= -1)
		{
			unset_signal_handler (SIGALRM);
		}
		else if (ticker_pid == 0)
		{
			/* child process - actual ticker */
			while (1)
			{
			#if defined(HAVE_NANOSLEEP)
				struct timespec ts;
				ts.tv_sec = 0;
				ts.tv_nsec = MOO_USEC_TO_NSEC(MOO_TICKER_INTERVAL_USECS) * 2;
				nanosleep (&ts, MOO_NULL);
			#elif defined(HAVE_USLEEP)
				usleep (MOO_TICKER_INTERVAL_USECS * 2);

			#else
			#	error UNDEFINED SLEEP
			#endif

				kill (getppid(), SIGALRM);
			}

			_exit (0);
		}

		/* parent just carries on. */
	}
}

static MOO_INLINE void stop_ticker (void)
{
	if (ticker_pid >= 0)
	{
		int wstatus;
		kill (ticker_pid, SIGKILL);
		while (waitpid(ticker_pid, &wstatus, 0) != ticker_pid);
		ticker_pid = -1;

		unset_signal_handler (SIGALRM);
	}
}

#endif

/* ========================================================================= */

#if defined(USE_LTDL)
#	define sys_dl_error() lt_dlerror()
#	define sys_dl_open(x) lt_dlopen(x)
#	define sys_dl_openext(x) lt_dlopenext(x)
#	define sys_dl_close(x) lt_dlclose(x)
#	define sys_dl_getsym(x,n) lt_dlsym(x,n)

#elif defined(USE_DLFCN)
#	define sys_dl_error() dlerror()
#	define sys_dl_open(x) dlopen(x,RTLD_NOW | RTLD_LOCAL)
#	define sys_dl_openext(x) dlopen(x,RTLD_NOW | RTLD_LOCAL)
#	define sys_dl_close(x) dlclose(x)
#	define sys_dl_getsym(x,n) dlsym(x,n)

#elif defined(USE_WIN_DLL)
#	define sys_dl_error() msw_dlerror()
#	define sys_dl_open(x) LoadLibraryExA(x, MOO_NULL, 0)
#	define sys_dl_openext(x) LoadLibraryExA(x, MOO_NULL, 0)
#	define sys_dl_close(x) FreeLibrary(x)
#	define sys_dl_getsym(x,n) GetProcAddress(x,n)

#elif defined(USE_MACH_O_DYLD)
#	define sys_dl_error() mach_dlerror()
#	define sys_dl_open(x) mach_dlopen(x)
#	define sys_dl_openext(x) mach_dlopen(x)
#	define sys_dl_close(x) mach_dlclose(x)
#	define sys_dl_getsym(x,n) mach_dlsym(x,n)
#endif

#if defined(USE_WIN_DLL)

static const char* msw_dlerror (void)
{
	/* TODO: handle wchar_t, moo_ooch_t etc? */
	static char buf[256];
	DWORD rc;

	rc = FormatMessageA (
		FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS,
		NULL, GetLastError(), MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT),
		buf, MOO_COUNTOF(buf), MOO_NULL
	);
	while (rc > 0 && buf[rc - 1] == '\r' || buf[rc - 1] == '\n')
	{
		buf[--rc] = '\0';
	}
	return buf;
}

#elif defined(USE_MACH_O_DYLD)
static const char* mach_dlerror_str = "";

static void* mach_dlopen (const char* path)
{
	NSObjectFileImage image;
	NSObjectFileImageReturnCode rc;
	void* handle;

	mach_dlerror_str = "";
	if ((rc = NSCreateObjectFileImageFromFile(path, &image)) != NSObjectFileImageSuccess)
	{
		switch (rc)
		{
			case NSObjectFileImageFailure:
			case NSObjectFileImageFormat:
				mach_dlerror_str = "unable to crate object file image";
				break;

			case NSObjectFileImageInappropriateFile:
				mach_dlerror_str = "inappropriate file";
				break;

			case NSObjectFileImageArch:
				mach_dlerror_str = "incompatible architecture";
				break;

			case NSObjectFileImageAccess:
				mach_dlerror_str = "inaccessible file";
				break;

			default:
				mach_dlerror_str = "unknown error";
				break;
		}
		return MOO_NULL;
	}
	handle = (void*)NSLinkModule(image, path, NSLINKMODULE_OPTION_PRIVATE | NSLINKMODULE_OPTION_RETURN_ON_ERROR);
	NSDestroyObjectFileImage (image);
	return handle;
}

static MOO_INLINE void mach_dlclose (void* handle)
{
	mach_dlerror_str = "";
	NSUnLinkModule (handle, NSUNLINKMODULE_OPTION_NONE);
}

static MOO_INLINE void* mach_dlsym (void* handle, const char* name)
{
	mach_dlerror_str = "";
	return (void*)NSAddressOfSymbol(NSLookupSymbolInModule(handle, name));
}

static const char* mach_dlerror (void)
{
	int err_no;
	const char* err_file;
	NSLinkEditErrors err;

	if (mach_dlerror_str[0] == '\0')
		NSLinkEditError (&err, &err_no, &err_file, &mach_dlerror_str);

	return mach_dlerror_str;
}
#endif

static void dl_startup (moo_t* moo)
{
#if defined(USE_LTDL)
	lt_dlinit ();
#endif
}

static void dl_cleanup (moo_t* moo)
{
#if defined(USE_LTDL)
	lt_dlexit ();
#endif
}

static void* dl_open (moo_t* moo, const moo_ooch_t* name, int flags)
{
#if defined(USE_LTDL) || defined(USE_DLFCN) || defined(USE_MACH_O_DYLD) || defined(USE_WIN_DLL)
	moo_bch_t stabuf[128], * bufptr;
	moo_oow_t ucslen, bcslen, bufcapa;
	void* handle;

	#if defined(MOO_OOCH_IS_UCH)
	if (moo_convootobcstr(moo, name, &ucslen, MOO_NULL, &bufcapa) <= -1) return MOO_NULL;
	/* +1 for terminating null. but it's not needed because MOO_COUNTOF(MOO_DEFAULT_PFMODPREFIX)
	 * and MOO_COUNTOF(MOO_DEFAULT_PFMODPOSTIFX) include the terminating nulls. Never mind about
	 * the extra 2 characters. */
	#else
	bufcapa = moo_count_bcstr(name);
	#endif
	bufcapa += MOO_COUNTOF(MOO_DEFAULT_PFMODDIR) + MOO_COUNTOF(MOO_DEFAULT_PFMODPREFIX) + MOO_COUNTOF(MOO_DEFAULT_PFMODPOSTFIX) + 1;

	if (bufcapa <= MOO_COUNTOF(stabuf)) bufptr = stabuf;
	else
	{
		bufptr = moo_allocmem(moo, bufcapa * MOO_SIZEOF(*bufptr));
		if (!bufptr) return MOO_NULL;
	}

	if (flags & MOO_VMPRIM_DLOPEN_PFMOD)
	{
		moo_oow_t len, i, xlen, dlen;

		/* opening a primitive function module - mostly libmoo-xxxx.
		 * if PFMODPREFIX is absolute, never use PFMODDIR */
		dlen = IS_PATH_ABSOLUTE(MOO_DEFAULT_PFMODPREFIX)?
			0: moo_copy_bcstr(bufptr, bufcapa, MOO_DEFAULT_PFMODDIR);
		len = moo_copy_bcstr(&bufptr[dlen], bufcapa - dlen, MOO_DEFAULT_PFMODPREFIX);
		len += dlen;

		bcslen = bufcapa - len;
	#if defined(MOO_OOCH_IS_UCH)
		moo_convootobcstr (moo, name, &ucslen, &bufptr[len], &bcslen);
	#else
		bcslen = moo_copy_bcstr(&bufptr[len], bcslen, name);
	#endif

		/* length including the directory, the prefix and the name. but excluding the postfix */
		xlen = len + bcslen;

		for (i = len; i < xlen; i++)
		{
			/* convert a period(.) to a dash(-) */
			if (bufptr[i] == '.') bufptr[i] = '-';
		}

	retry:
		moo_copy_bcstr (&bufptr[xlen], bufcapa - xlen, MOO_DEFAULT_PFMODPOSTFIX);

		/* both prefix and postfix attached. for instance, libmoo-xxx */
		handle = sys_dl_openext(&bufptr[dlen]);
		if (!handle)
		{
			MOO_DEBUG3 (moo, "Unable to open(ext) PFMOD %hs[%js] - %hs\n", &bufptr[dlen], name, sys_dl_error());

			if (dlen > 0)
			{
				handle = sys_dl_openext(&bufptr[0]);
				if (handle) goto pfmod_open_ok;
				MOO_DEBUG3 (moo, "Unable to open(ext) PFMOD %hs[%js] - %hs\n", &bufptr[0], name, sys_dl_error());
			}

			/* try without prefix and postfix */
			bufptr[xlen] = '\0';
			handle = sys_dl_openext(&bufptr[len]);
			if (!handle)
			{
				moo_bch_t* dash;
				const moo_bch_t* dl_errstr;
				dl_errstr = sys_dl_error();
				MOO_DEBUG3 (moo, "Unable to open(ext) PFMOD %hs[%js] - %hs\n", &bufptr[len], name, dl_errstr);
				moo_seterrbfmt (moo, MOO_ESYSERR, "unable to open(ext) PFMOD %js - %hs", name, dl_errstr);
				dash = moo_rfind_bchar(bufptr, moo_count_bcstr(bufptr), '-');
				if (dash)
				{
					/* remove a segment at the back.
					 * [NOTE] a dash contained in the original name before
					 *        period-to-dash transformation may cause extraneous/wrong
					 *        loading reattempts. */
					xlen = dash - bufptr;
					goto retry;
				}
			}
			else
			{
				MOO_DEBUG3 (moo, "OPENED_HERE(ext) PFMOD %hs[%js] handle %p\n", &bufptr[len], name, handle);
			}
		}
		else
		{
		pfmod_open_ok:
			MOO_DEBUG3 (moo, "OPENED_HERE(ext) PFMOD %hs[%js] handle %p\n", &bufptr[dlen], name, handle);
		}
	}
	else
	{
		/* opening a raw shared object without a prefix and/or a postfix */
	#if defined(MOO_OOCH_IS_UCH)
		bcslen = bufcapa;
		moo_convootobcstr (moo, name, &ucslen, bufptr, &bcslen);
	#else
		bcslen = moo_copy_bcstr(bufptr, bufcapa, name);
	#endif

		if (moo_find_bchar(bufptr, bcslen, '.'))
		{
			handle = sys_dl_open(bufptr);
			if (!handle)
			{
				const moo_bch_t* dl_errstr;
				dl_errstr = sys_dl_error();
				MOO_DEBUG2 (moo, "Unable to open DL %hs - %hs\n", bufptr, dl_errstr);
				moo_seterrbfmt (moo, MOO_ESYSERR, "unable to open DL %js - %hs", name, dl_errstr);
			}
			else MOO_DEBUG2 (moo, "OPENED_HERE DL %hs handle %p\n", bufptr, handle);
		}
		else
		{
			handle = sys_dl_openext(bufptr);
			if (!handle)
			{
				const moo_bch_t* dl_errstr;
				dl_errstr = sys_dl_error();
				MOO_DEBUG2 (moo, "Unable to open(ext) DL %hs - %hs\n", bufptr, dl_errstr);
				moo_seterrbfmt (moo, MOO_ESYSERR, "unable to open(ext) DL %js - %hs", name, dl_errstr);
			}
			else MOO_DEBUG2 (moo, "OPENED_HERE(ext) DL %hs handle %p\n", bufptr, handle);
		}
	}

	if (bufptr != stabuf) moo_freemem (moo, bufptr);
	return handle;

#else

/* TODO: support various platforms */
	/* TODO: implemenent this */
	MOO_DEBUG1 (moo, "Dynamic loading not implemented - cannot open %js\n", name);
	moo_seterrbfmt (moo, MOO_ENOIMPL, "dynamic loading not implemented - cannot open %js", name);
	return MOO_NULL;
#endif
}

static void dl_close (moo_t* moo, void* handle)
{
#if defined(USE_LTDL) || defined(USE_DLFCN) || defined(USE_MACH_O_DYLD) || defined(USE_WIN_DLL)
	MOO_DEBUG1 (moo, "Closed DL handle %p\n", handle);
	sys_dl_close (handle);

#else
	/* TODO: implemenent this */
	MOO_DEBUG1 (moo, "Dynamic loading not implemented - cannot close handle %p\n", handle);
#endif
}

static void* dl_getsym (moo_t* moo, void* handle, const moo_ooch_t* name)
{
#if defined(USE_LTDL) || defined(USE_DLFCN) || defined(USE_MACH_O_DYLD) || defined(USE_WIN_DLL)
	moo_bch_t stabuf[64], * bufptr;
	moo_oow_t bufcapa, ucslen, bcslen, i;
	const moo_bch_t* symname;
	void* sym;

	#if defined(MOO_OOCH_IS_UCH)
	if (moo_convootobcstr(moo, name, &ucslen, MOO_NULL, &bcslen) <= -1) return MOO_NULL;
	#else
	bcslen = moo_count_bcstr (name);
	#endif

	if (bcslen >= MOO_COUNTOF(stabuf) - 2)
	{
		bufcapa = bcslen + 3;
		bufptr = moo_allocmem(moo, bufcapa * MOO_SIZEOF(*bufptr));
		if (!bufptr) return MOO_NULL;
	}
	else
	{
		bufcapa = MOO_COUNTOF(stabuf);
		bufptr = stabuf;
	}

	bcslen = bufcapa - 1;
	#if defined(MOO_OOCH_IS_UCH)
	moo_convootobcstr (moo, name, &ucslen, &bufptr[1], &bcslen);
	#else
	bcslen = moo_copy_bcstr(&bufptr[1], bcslen, name);
	#endif

	/* convert a period(.) to an underscore(_) */
	for (i = 1; i <= bcslen; i++) if (bufptr[i] == '.') bufptr[i] = '_';

	symname = &bufptr[1]; /* try the name as it is */
	sym = sys_dl_getsym(handle, symname);
	if (!sym)
	{
		bufptr[0] = '_';
		symname = &bufptr[0]; /* try _name */
		sym = sys_dl_getsym(handle, symname);
		if (!sym)
		{
			bufptr[bcslen + 1] = '_';
			bufptr[bcslen + 2] = '\0';

			symname = &bufptr[1]; /* try name_ */
			sym = sys_dl_getsym(handle, symname);

			if (!sym)
			{
				symname = &bufptr[0]; /* try _name_ */
				sym = sys_dl_getsym(handle, symname);
				if (!sym)
				{
					const moo_bch_t* dl_errstr;
					dl_errstr = sys_dl_error();
					MOO_DEBUG3 (moo, "Unable to get module symbol %js from handle %p - %hs\n", name, handle, dl_errstr);
					moo_seterrbfmt (moo, MOO_ENOENT, "unable to get module symbol %hs - %hs", symname, dl_errstr);
				}
			}
		}
	}

	if (sym) MOO_DEBUG3 (moo, "Loaded module symbol %js from handle %p - %hs\n", name, handle, symname);
	if (bufptr != stabuf) moo_freemem (moo, bufptr);
	return sym;

#else
	/* TODO: IMPLEMENT THIS */
	MOO_DEBUG2 (moo, "Dynamic loading not implemented - Cannot load module symbol %js from handle %p\n", name, handle);
	moo_seterrbfmt (moo, MOO_ENOIMPL, "dynamic loading not implemented - Cannot load module symbol %js from handle %p", name, handle);
	return MOO_NULL;
#endif
}
/* ========================================================================= */

static int open_pipes (moo_t* moo, int p[2])
{
#if defined(_WIN32)
	u_long flags;
#else
	int flags;
#endif

#if defined(_WIN32)
	if (_pipe(p, 256, _O_BINARY | _O_NOINHERIT) == -1)
	{
		moo_seterrbfmtwithsyserr (moo, 0, errno, "unable to create pipes");
		return -1;
	}
#elif defined(__OS2__)
	#if defined(TCPV40HDRS)
	/* neither pipe nor socketpair available */
	if (os2_socket_pair(p) == -1)
	#else
	if (socketpair(AF_LOCAL, SOCK_STREAM, 0, p) == -1)
	#endif
	{
		moo_seterrbfmtwithsyserr (moo, 2, sock_errno(), "unable to create pipes");
		return -1;
	}
#elif defined(HAVE_PIPE2) && defined(O_CLOEXEC) && defined(O_NONBLOCK)
	if (pipe2(p, O_CLOEXEC | O_NONBLOCK) == -1)
	{
		moo_seterrbfmtwithsyserr (moo, 0, errno, "unable to create pipes");
		return -1;
	}
#else
	if (pipe(p) == -1)
	{
		moo_seterrbfmtwithsyserr (moo, 0, errno, "unable to create pipes");
		return -1;
	}
#endif


#if defined(_WIN32)
	flags = 1;
	ioctl (p[0], FIONBIO, &flags);
	ioctl (p[1], FIONBIO, &flags);
#elif defined(__OS2__)
	flags = 1; /* don't block */
	ioctl (p[0], FIONBIO, (char*)&flags, moo_SIZEOF(flags));
	ioctl (p[1], FIONBIO, (char*)&flags, moo_SIZEOF(flags));
#elif defined(HAVE_PIPE2) && defined(O_CLOEXEC) && defined(O_NONBLOCK)
		/* do nothing */
#else
	#if defined(FD_CLOEXEC)
	flags = fcntl(p[0], F_GETFD);
	if (flags >= 0) fcntl (p[0], F_SETFD, flags | FD_CLOEXEC);
	flags = fcntl(p[1], F_GETFD);
	if (flags >= 0) fcntl (p[1], F_SETFD, flags | FD_CLOEXEC);
	#endif
	#if defined(O_NONBLOCK)
	flags = fcntl(p[0], F_GETFL);
	if (flags >= 0) fcntl (p[0], F_SETFL, flags | O_NONBLOCK);
	flags = fcntl(p[1], F_GETFL);
	if (flags >= 0) fcntl (p[1], F_SETFL, flags | O_NONBLOCK);
	#endif
#endif

	return 0;
}

static void close_pipes (moo_t* moo, int p[2])
{
#if defined(_WIN32)
	_close (p[0]);
	_close (p[1]);
#elif defined(__OS2__)
	soclose (p[0]);
	soclose (p[1]);
#else
	close (p[0]);
	close (p[1]);
#endif
	p[0] = -1;
	p[1] = -1;
}

static int vm_startup (moo_t* moo)
{
	xtn_t* xtn = GET_XTN(moo);
	int sigfd_pcount = 0;
	int iothr_pcount = 0, flags;

#if defined(_WIN32)
	xtn->waitable_timer = CreateWaitableTimer(MOO_NULL, TRUE, MOO_NULL);
#endif

#if defined(USE_DEVPOLL)
	xtn->ep = open("/dev/poll", O_RDWR);
	if (xtn->ep == -1)
	{
		moo_seterrwithsyserr (moo, 0, errno);
		MOO_DEBUG1 (moo, "Cannot create devpoll - %hs\n", strerror(errno));
		goto oops;
	}

	#if defined(FD_CLOEXEC)
	flags = fcntl(xtn->ep, F_GETFD);
	if (flags >= 0) fcntl (xtn->ep, F_SETFD, flags | FD_CLOEXEC);
	#endif

#elif defined(USE_KQUEUE)
	#if defined(HAVE_KQUEUE1) && defined(O_CLOEXEC)
	xtn->ep = kqueue1(O_CLOEXEC);
	if (xtn->ep == -1) xtn->ep = kqueue();
	#else
	xtn->ep = kqueue();
	#endif
	if (xtn->ep == -1)
	{
		moo_seterrwithsyserr (moo, 0, errno);
		MOO_DEBUG1 (moo, "Cannot create kqueue - %hs\n", strerror(errno));
		goto oops;
	}

	#if defined(FD_CLOEXEC)
	flags = fcntl(xtn->ep, F_GETFD);
	if (flags >= 0 && !(flags & FD_CLOEXEC)) fcntl (xtn->ep, F_SETFD, flags | FD_CLOEXEC);
	#endif

#elif defined(USE_EPOLL)
	#if defined(HAVE_EPOLL_CREATE1) && defined(EPOLL_CLOEXEC)
	xtn->ep = epoll_create1(EPOLL_CLOEXEC);
	if (xtn->ep == -1) xtn->ep = epoll_create(1024);
	#else
	xtn->ep = epoll_create(1024);
	#endif
	if (xtn->ep == -1)
	{
		moo_seterrwithsyserr (moo, 0, errno);
		MOO_DEBUG1 (moo, "Cannot create epoll - %hs\n", strerror(errno));
		goto oops;
	}

	#if defined(FD_CLOEXEC)
	flags = fcntl(xtn->ep, F_GETFD);
	if (flags >= 0 && !(flags & FD_CLOEXEC)) fcntl (xtn->ep, F_SETFD, flags | FD_CLOEXEC);
	#endif

#elif defined(USE_POLL)

	MUTEX_INIT (&xtn->ev.reg.pmtx);

#elif defined(USE_SELECT)
	FD_ZERO (&xtn->ev.reg.rfds);
	FD_ZERO (&xtn->ev.reg.wfds);
	xtn->ev.reg.maxfd = -1;
	MUTEX_INIT (&xtn->ev.reg.smtx);
#endif /* USE_DEVPOLL */

	if (open_pipes(moo, xtn->sigfd.p) <= -1) goto oops;
	sigfd_pcount = 2;

#if defined(USE_THREAD)
	if (open_pipes(moo, xtn->iothr.p) <= -1) goto oops;
	iothr_pcount = 2;

	if (_add_poll_fd(moo, xtn->iothr.p[0], XPOLLIN) <= -1) goto oops;

	pthread_mutex_init (&xtn->ev.mtx, MOO_NULL);
	pthread_cond_init (&xtn->ev.cnd, MOO_NULL);
	pthread_cond_init (&xtn->ev.cnd2, MOO_NULL);
	xtn->ev.halting = 0;

	xtn->iothr.abort = 0;
	xtn->iothr.up = 0;
	/*pthread_create (&xtn->iothr, MOO_NULL, iothr_main, moo);*/

#endif /* USE_THREAD */

	xtn->vm_running = 1;
	return 0;

oops:
#if defined(USE_THREAD)
	if (iothr_pcount > 0)
	{
		close (xtn->iothr.p[0]);
		close (xtn->iothr.p[1]);
	}
#endif

	if (sigfd_pcount > 0)
	{
		close (xtn->sigfd.p[0]);
		close (xtn->sigfd.p[1]);
	}

#if defined(USE_DEVPOLL) || defined(USE_EPOLL)
	if (xtn->ep >= 0)
	{
		close (xtn->ep);
		xtn->ep = -1;
	}
#endif

	return -1;
}

static void vm_cleanup (moo_t* moo)
{
	xtn_t* xtn = GET_XTN(moo);

	xtn->vm_running = 0;

#if defined(_WIN32)
	if (xtn->waitable_timer)
	{
		CloseHandle (xtn->waitable_timer);
		xtn->waitable_timer = MOO_NULL;
	}
#endif

#if defined(USE_THREAD)
	if (xtn->iothr.up)
	{
		xtn->iothr.abort = 1;
		write (xtn->iothr.p[1], "Q", 1);
		pthread_cond_signal (&xtn->ev.cnd);
		pthread_join (xtn->iothr.thr, MOO_NULL);
		xtn->iothr.up = 0;
	}
	pthread_cond_destroy (&xtn->ev.cnd);
	pthread_cond_destroy (&xtn->ev.cnd2);
	pthread_mutex_destroy (&xtn->ev.mtx);

	_del_poll_fd (moo, xtn->iothr.p[0]);
	close_pipes (moo, xtn->iothr.p);
#endif /* USE_THREAD */

	close_pipes (moo, xtn->sigfd.p);

#if defined(USE_DEVPOLL)
	if (xtn->ep >= 0)
	{
		close (xtn->ep);
		xtn->ep = -1;
	}
	/*destroy_poll_data_space (moo);*/
#elif defined(USE_KQUEUE)
	if (xtn->ep >= 0)
	{
		close (xtn->ep);
		xtn->ep = -1;
	}
#elif defined(USE_EPOLL)
	if (xtn->ep >= 0)
	{
		close (xtn->ep);
		xtn->ep = -1;
	}
#elif defined(USE_POLL)
	if (xtn->ev.reg.ptr)
	{
		moo_freemem (moo, xtn->ev.reg.ptr);
		xtn->ev.reg.ptr = MOO_NULL;
		xtn->ev.reg.len = 0;
		xtn->ev.reg.capa = 0;
	}
	if (xtn->ev.buf)
	{
		moo_freemem (moo, xtn->ev.buf);
		xtn->ev.buf = MOO_NULL;
	}
	/*destroy_poll_data_space (moo);*/
	MUTEX_DESTROY (&xtn->ev.reg.pmtx);
#elif defined(USE_SELECT)
	FD_ZERO (&xtn->ev.reg.rfds);
	FD_ZERO (&xtn->ev.reg.wfds);
	xtn->ev.reg.maxfd = -1;
	MUTEX_DESTROY (&xtn->ev.reg.smtx);
#endif
}


/* ========================================================================= */
#if defined(_WIN32)

static const wchar_t* msw_exception_name (DWORD excode)
{
	switch (excode)
	{
		case EXCEPTION_ACCESS_VIOLATION:          return L"Access violation exception";
		case EXCEPTION_ARRAY_BOUNDS_EXCEEDED:     return L"Array bounds exceeded";
		case EXCEPTION_BREAKPOINT:                return L"Breakpoint";
		case EXCEPTION_DATATYPE_MISALIGNMENT:     return L"Float datatype misalignment";
		case EXCEPTION_FLT_DENORMAL_OPERAND:      return L"Float denormal operand";
		case EXCEPTION_FLT_DIVIDE_BY_ZERO:        return L"Float divide by zero";
		case EXCEPTION_FLT_INEXACT_RESULT:        return L"Float inexact result";
		case EXCEPTION_FLT_OVERFLOW:              return L"Float overflow";
		case EXCEPTION_FLT_STACK_CHECK:           return L"Float stack check";
		case EXCEPTION_FLT_UNDERFLOW:             return L"Float underflow";
		case EXCEPTION_ILLEGAL_INSTRUCTION:       return L"Illegal instruction";
		case EXCEPTION_IN_PAGE_ERROR:             return L"In page error";
		case EXCEPTION_INT_DIVIDE_BY_ZERO:        return L"Integer divide by zero";
		case EXCEPTION_INT_OVERFLOW:              return L"Integer overflow";
		case EXCEPTION_INVALID_DISPOSITION:       return L"Invalid disposition";
		case EXCEPTION_NONCONTINUABLE_EXCEPTION:  return L"Noncontinuable exception";
		case EXCEPTION_PRIV_INSTRUCTION:          return L"Priv instruction";
		case EXCEPTION_SINGLE_STEP:               return L"Single step";
		case EXCEPTION_STACK_OVERFLOW:            return L"Stack overflow";
		default:                                  return L"Unknown exception";
	}
}

static const wchar_t* msw_exception_opname (const ULONG opcode)
{
	switch (opcode)
	{
		case 0: return L"Read attempt from inaccessible data";
		case 1: return L"Write attempt to inaccessible data";
		case 8: return L"User-mode data execution prevention violation";
		default: return L"Unknown exception operation";
	}
}

static LONG WINAPI msw_exception_filter (struct _EXCEPTION_POINTERS* exinfo)
{
	HMODULE mod;
#if defined(_WIN32_WINNT) && (_WIN32_WINNT >= 0x0501)
	MODULEINFO modinfo;
#endif
	DWORD excode;
	static wchar_t exmsg[256];
	static wchar_t expath[128];

#if defined(_WIN32_WINNT) && (_WIN32_WINNT >= 0x0501)
	GetModuleHandleExW (GET_MODULE_HANDLE_EX_FLAG_FROM_ADDRESS, exinfo->ExceptionRecord->ExceptionAddress, &mod);
	/*GetModuleInformation (GetCurrentProcess(), mod, &modinfo, MOO_SIZEOF(modinfo));*/
	GetModuleFileNameExW (GetCurrentProcess(), mod, expath, MOO_SIZEOF(expath));
#else
	GetModuleFileNameW (MOO_NULL, expath, MOO_SIZEOF(expath));
#endif

	excode = exinfo->ExceptionRecord->ExceptionCode;
	if (excode == EXCEPTION_BREAKPOINT) return EXCEPTION_CONTINUE_SEARCH;

	if (excode == EXCEPTION_ACCESS_VIOLATION || excode == EXCEPTION_IN_PAGE_ERROR)
	{
		_snwprintf (exmsg, MOO_COUNTOF(exmsg), L"Exception %s(%u) at 0x%p - Invalid operation at 0x%p - %s",
			msw_exception_name(excode), (unsigned int)excode,
			exinfo->ExceptionRecord->ExceptionAddress,
			(PVOID)exinfo->ExceptionRecord->ExceptionInformation[1],
			msw_exception_opname(exinfo->ExceptionRecord->ExceptionInformation[0])
		);
	}
	else
	{
		_snwprintf (exmsg, MOO_COUNTOF(exmsg), L"Exception %s(%u) at 0x%p",
			msw_exception_name(excode), (unsigned int)excode,
			exinfo->ExceptionRecord->ExceptionAddress
		);
	}

	/* TODO: use a global output callback like vmprim.assertfail().
	 *       vmprim.assertfail() requires 'moo'. so i need another global level callback for this */
	MessageBoxW (NULL, exmsg, expath, MB_OK | MB_ICONERROR);

	/*return EXCEPTION_CONTINUE_SEARCH;*/
	/*return EXCEPTION_CONTINUE_EXECUTION;*/
	return EXCEPTION_EXECUTE_HANDLER;
}
#endif
/* ========================================================================= */

static struct
{
	const char* name;
	int op;
	moo_bitmask_t mask;
} log_mask_table[] =
{
	{ "app",         0, MOO_LOG_APP },
	{ "compiler",    0, MOO_LOG_COMPILER },
	{ "vm",          0, MOO_LOG_VM },
	{ "mnemonic",    0, MOO_LOG_MNEMONIC },
	{ "gc",          0, MOO_LOG_GC },
	{ "ic",          0, MOO_LOG_IC },
	{ "primitive",   0, MOO_LOG_PRIMITIVE },
	{ "all",         0, MOO_LOG_ALL_TYPES },

	{ "fatal",       0, MOO_LOG_FATAL },
	{ "error",       0, MOO_LOG_ERROR },
	{ "warn",        0, MOO_LOG_WARN },
	{ "info",        0, MOO_LOG_INFO },
	{ "debug",       0, MOO_LOG_DEBUG },

	{ "fatal+",      0, MOO_LOG_FATAL },
	{ "error+",      0, MOO_LOG_FATAL | MOO_LOG_ERROR },
	{ "warn+",       0, MOO_LOG_FATAL | MOO_LOG_ERROR | MOO_LOG_WARN },
	{ "info+",       0, MOO_LOG_FATAL | MOO_LOG_ERROR | MOO_LOG_WARN | MOO_LOG_INFO },
	{ "debug+",      0, MOO_LOG_FATAL | MOO_LOG_ERROR | MOO_LOG_WARN | MOO_LOG_INFO | MOO_LOG_DEBUG },

	{ "fatal-",      0, MOO_LOG_FATAL | MOO_LOG_ERROR | MOO_LOG_WARN | MOO_LOG_INFO | MOO_LOG_DEBUG },
	{ "error-",      0, MOO_LOG_ERROR | MOO_LOG_WARN | MOO_LOG_INFO | MOO_LOG_DEBUG },
	{ "warn-",       0, MOO_LOG_WARN | MOO_LOG_INFO | MOO_LOG_DEBUG },
	{ "info-",       0, MOO_LOG_INFO | MOO_LOG_DEBUG },
	{ "debug-",      0, MOO_LOG_DEBUG },

	{ "-fatal",      1, ~MOO_LOG_FATAL },
	{ "-error",      1, ~MOO_LOG_ERROR },
	{ "-warn",       1, ~MOO_LOG_WARN },
	{ "-info",       1, ~MOO_LOG_INFO },
	{ "-debug",      1, ~MOO_LOG_DEBUG },
};

static struct
{
	const char* name;
	moo_bitmask_t mask;
} dbg_mask_table[] =
{
	{ "bigint", MOO_TRAIT_DEBUG_BIGINT },
	{ "gc",     MOO_TRAIT_DEBUG_GC }
};

static int parse_logoptb (moo_t* moo, const moo_bch_t* str, moo_oow_t* xpathlen, moo_bitmask_t* xlogmask)
{
	/*xtn_t* xtn = GET_XTN(moo);*/
	const moo_bch_t* cm, * flt;
	moo_bitmask_t logmask;
	moo_oow_t i, len, pathlen;

	cm = moo_find_bchar_in_bcstr(str, ',');
	if (cm)
	{
		/* i duplicate this string for open() below as open() doesn't
		 * accept a length-bounded string */
		cm = moo_find_bchar_in_bcstr(str, ',');
		pathlen = cm - str;

		logmask = 0;

		do
		{
			flt = cm + 1;

			cm = moo_find_bchar_in_bcstr(flt, ',');
			len = cm? (cm - flt): moo_count_bcstr(flt);

			for (i = 0; i < MOO_COUNTOF(log_mask_table); i++)
			{
				if (moo_comp_bchars_bcstr(flt, len, log_mask_table[i].name) == 0)
				{
					if (log_mask_table[i].op)
						logmask &= log_mask_table[i].mask;
					else
						logmask |= log_mask_table[i].mask;
					break;
				}
			}

			if (i >= MOO_COUNTOF(log_mask_table))
			{
				moo_seterrbfmt (moo, MOO_EINVAL, "unknown log option value %.*hs", len, flt);
				return -1;
			}
		}
		while (cm);

		if (!(logmask & MOO_LOG_ALL_TYPES)) logmask |= MOO_LOG_ALL_TYPES & ~(moo_bitmask_t)MOO_LOG_GC;  /* no types specified. force to all types except gc */
		if (!(logmask & MOO_LOG_ALL_LEVELS)) logmask |= MOO_LOG_ALL_LEVELS;  /* no levels specified. force to all levels */
	}
	else
	{
		logmask = (MOO_LOG_ALL_LEVELS | MOO_LOG_ALL_TYPES) & ~(moo_bitmask_t)MOO_LOG_GC; /* all types except gc + all levels */
		pathlen = moo_count_bcstr(str);
	}

	/* to have gc logs produced, you should specify 'all' or 'gc' */

	*xlogmask = logmask;
	*xpathlen = pathlen;
	return 0;
}


static int parse_logoptu (moo_t* moo, const moo_uch_t* str, moo_oow_t* xpathlen, moo_bitmask_t* xlogmask)
{
	/*xtn_t* xtn = GET_XTN(moo);*/
	const moo_uch_t* cm, * flt;
	moo_bitmask_t logmask;
	moo_oow_t i, len, pathlen;

	cm = moo_find_uchar_in_ucstr(str, ',');
	if (cm)
	{
		cm = moo_find_uchar_in_ucstr(str, ',');
		pathlen = cm - str;

		logmask = 0;
		do
		{
			flt = cm + 1;

			cm = moo_find_uchar_in_ucstr(flt, ',');
			len = cm? (cm - flt): moo_count_ucstr(flt);

			for (i = 0; i < MOO_COUNTOF(log_mask_table); i++)
			{
				if (moo_comp_uchars_bcstr(flt, len, log_mask_table[i].name) == 0)
				{
					if (log_mask_table[i].op)
						logmask &= log_mask_table[i].mask;
					else
						logmask |= log_mask_table[i].mask;
					break;
				}
			}

			if (i >= MOO_COUNTOF(log_mask_table))
			{
				moo_seterrbfmt (moo, MOO_EINVAL, "unknown log option value %.*ls", len, flt);
				return -1;
			}
		}
		while (cm);

		if (!(logmask & MOO_LOG_ALL_TYPES)) logmask |= MOO_LOG_ALL_TYPES;  /* no types specified. force to all types */
		if (!(logmask & MOO_LOG_ALL_LEVELS)) logmask |= MOO_LOG_ALL_LEVELS;  /* no levels specified. force to all levels */
	}
	else
	{
		logmask = MOO_LOG_ALL_LEVELS | MOO_LOG_ALL_TYPES;
		pathlen = moo_count_ucstr(str);
	}

	*xlogmask = logmask;
	*xpathlen = pathlen;
	return 0;
}

static int handle_logoptb (moo_t* moo, const moo_bch_t* str)
{
	xtn_t* xtn = GET_XTN(moo);
	moo_bitmask_t logmask;
	moo_oow_t pathlen;
	moo_bch_t* xstr = (moo_bch_t*)str;

	if (parse_logoptb(moo, str, &pathlen, &logmask) <= -1) return -1;
	if (str[pathlen] != '\0')
	{
		xstr = moo_dupbchars(moo, str, pathlen);
		if (!xstr) moo_seterrbfmt (moo, moo_geterrnum(moo), "out of memory in duplicating %hs", str);
	}

	xtn->log.fd = open(xstr, O_CREAT | O_WRONLY | O_APPEND , 0644);
	if (xtn->log.fd == -1)
	{
		moo_seterrbfmt (moo, MOO_ESYSERR, "cannot open log file %hs", xstr); /* TODO: use syserrb/u??? */
		if (str != xstr) moo_freemem (moo, xstr);
		return -1;
	}

	xtn->log.fd_flag |= LOGFD_OPENED_HERE;
#if defined(HAVE_ISATTY)
	if (isatty(xtn->log.fd)) xtn->log.fd_flag |= LOGFD_TTY;
#endif

	if (str != xstr) moo_freemem (moo, xstr);
	moo_setoption (moo, MOO_OPTION_LOG_MASK, &logmask);
	return 0;
}

static int handle_logoptu (moo_t* moo, const moo_uch_t* str)
{
	xtn_t* xtn = GET_XTN(moo);
	moo_bitmask_t logmask;
	moo_oow_t pathlen;
	moo_bch_t* xstr;

	if (parse_logoptu(moo, str, &pathlen, &logmask) <= -1) return -1;

	xstr = moo_duputobchars(moo, str, pathlen, MOO_NULL); /* moo_duputobchars() null-terminates xstr */
	if (!xstr) moo_seterrbfmt (moo, moo_geterrnum(moo), "out of memory in duplicating %hs", str);

	xtn->log.fd = open(xstr, O_CREAT | O_WRONLY | O_APPEND , 0644);
	if (xtn->log.fd == -1)
	{
		moo_seterrbfmt (moo, MOO_ESYSERR, "cannot open log file %hs", xstr); /* TODO: use syserrb/u??? */
		moo_freemem (moo, xstr);
		return -1;
	}

	xtn->log.fd_flag |= LOGFD_OPENED_HERE;
#if defined(HAVE_ISATTY)
	if (isatty(xtn->log.fd)) xtn->log.fd_flag |= LOGFD_TTY;
#endif

	moo_freemem (moo, xstr);
	moo_setoption (moo, MOO_OPTION_LOG_MASK, &logmask);
	return 0;
}

static int handle_dbgoptb (moo_t* moo, const moo_bch_t* str)
{
	const moo_bch_t* cm, * flt;
	moo_oow_t len, i;
	moo_bitmask_t trait, dbgopt = 0;

	cm = str - 1;
	do
	{
		flt = cm + 1;

		cm = moo_find_bchar_in_bcstr(flt, ',');
		len = cm? (cm - flt): moo_count_bcstr(flt);

		for (i = 0; i < MOO_COUNTOF(dbg_mask_table); i++)
		{
			if (moo_comp_bchars_bcstr(flt, len, dbg_mask_table[i].name) == 0)
			{
				dbgopt |= dbg_mask_table[i].mask;
				break;
			}
		}
		if (i >= MOO_COUNTOF(log_mask_table))
		{
			moo_seterrbfmt (moo, MOO_EINVAL, "unknown log option value %.*hs", len, flt);
			return -1;
		}
	}
	while (cm);

	moo_getoption (moo, MOO_OPTION_TRAIT, &trait);
	trait |= dbgopt;
	moo_setoption (moo, MOO_OPTION_TRAIT, &trait);

	return 0;
}

static int handle_dbgoptu (moo_t* moo, const moo_uch_t* str)
{
	const moo_uch_t* cm, * flt;
	moo_oow_t len, i;
	moo_bitmask_t trait, dbgopt = 0;

	cm = str - 1;
	do
	{
		flt = cm + 1;

		cm = moo_find_uchar_in_ucstr(flt, ',');
		len = cm? (cm - flt): moo_count_ucstr(flt);

		for (i = 0; i < MOO_COUNTOF(dbg_mask_table); i++)
		{
			if (moo_comp_uchars_bcstr(flt, len, dbg_mask_table[i].name) == 0)
			{
				dbgopt |= dbg_mask_table[i].mask;
				break;
			}
		}
		if (i >= MOO_COUNTOF(log_mask_table))
		{
			moo_seterrbfmt (moo, MOO_EINVAL, "unknown log option value %.*ls", len, flt);
			return -1;
		}
	}
	while (cm);

	moo_getoption (moo, MOO_OPTION_TRAIT, &trait);
	trait |= dbgopt;
	moo_setoption (moo, MOO_OPTION_TRAIT, &trait);

	return 0;
}


static int handle_cfg_options (moo_t* moo, const moo_cfgstd_t* cfg)
{
	switch (cfg->type)
	{
	#if defined(MOO_OOCH_IS_BCH)
		case MOO_CFGSTD_OPT:
			MOO_ASSERT (moo, &cfg->u.optb.log == &cfg->u.opt.log);
			MOO_ASSERT (moo, &cfg->u.optb.dbg == &cfg->u.opt.dbg);
	#endif
		case MOO_CFGSTD_OPTB:
			if ((cfg->u.optb.log && handle_logoptb(moo, cfg->u.optb.log) <= -1) ||
			    (cfg->u.optb.dbg && handle_dbgoptb(moo, cfg->u.optb.dbg) <= -1)) return -1;
			return 0;

	#if defined(MOO_OOCH_IS_UCH)
		case MOO_CFGSTD_OPT:
			MOO_ASSERT (moo, &cfg->u.optu.log == &cfg->u.opt.log);
			MOO_ASSERT (moo, &cfg->u.optu.dbg == &cfg->u.opt.dbg);
	#endif
		case MOO_CFGSTD_OPTU:
			if ((cfg->u.optu.log && handle_logoptu(moo, cfg->u.optu.log) <= -1) ||
			    (cfg->u.optu.dbg && handle_dbgoptu(moo, cfg->u.optu.dbg) <= -1)) return -1;
			return 0;

		default:
			moo_seterrbfmt (moo, MOO_EINVAL, "unsupported configuration option type - %d", (int)cfg->type);
			return -1;
	}
}
/* ========================================================================= */


static MOO_INLINE void reset_log_to_default (xtn_t* xtn)
{
#if defined(ENABLE_LOG_INITIALLY)
	xtn->log.fd = 2;
	xtn->log.fd_flag = 0;
	#if defined(HAVE_ISATTY)
	if (isatty(xtn->log.fd)) xtn->log.fd_flag |= LOGFD_TTY;
	#endif
#else
	xtn->log.fd = -1;
	xtn->log.fd_flag = 0;
#endif
}

static MOO_INLINE void chain (moo_t* moo)
{
	xtn_t* xtn = GET_XTN(moo);

	/* TODO: make this atomic */
	xtn->prev = MOO_NULL;
	xtn->next = g_moo;

	if (g_moo) GET_XTN(g_moo)->prev = moo;
	else g_moo = moo;
	/* TODO: make this atomic */
}

static MOO_INLINE void unchain (moo_t* moo)
{
	xtn_t* xtn = GET_XTN(moo);

	/* TODO: make this atomic */
	if (xtn->prev) GET_XTN(xtn->prev)->next = xtn->next;
	else g_moo = xtn->next;
	if (xtn->next) GET_XTN(xtn->next)->prev = xtn->prev;
	/* TODO: make this atomic */
	xtn->prev = MOO_NULL;
	xtn->prev = MOO_NULL;
}

static void fini_moo (moo_t* moo)
{
	xtn_t* xtn = GET_XTN(moo);
	if ((xtn->log.fd_flag & LOGFD_OPENED_HERE) && xtn->log.fd >= 0) close (xtn->log.fd);
	reset_log_to_default (xtn);
	unchain (moo);
}

static void halting_moo (moo_t* moo)
{
	xtn_t* xtn = GET_XTN(moo);
	xtn->ev.halting = 1; /* once set, vm_sleep() is supposed to return without waiting */
}

moo_t* moo_openstd (moo_oow_t xtnsize, const moo_cfgstd_t* cfg, moo_errinf_t* errinfo)
{
	moo_t* moo;
	moo_vmprim_t vmprim;
	moo_evtcb_t evtcb;
	xtn_t* xtn;

	MOO_MEMSET(&vmprim, 0, MOO_SIZEOF(vmprim));
	if (cfg && cfg->large_pages)
	{
		vmprim.alloc_heap = alloc_heap;
		vmprim.free_heap = free_heap;
	}
	vmprim.log_write = ((cfg && cfg->log_write)? cfg->log_write: log_write);
	vmprim.syserrstrb = syserrstrb;
	vmprim.assertfail = assert_fail;
	vmprim.dl_startup = dl_startup;
	vmprim.dl_cleanup = dl_cleanup;
	vmprim.dl_open = dl_open;
	vmprim.dl_close = dl_close;
	vmprim.dl_getsym = dl_getsym;
	vmprim.vm_startup = vm_startup;
	vmprim.vm_cleanup = vm_cleanup;
	vmprim.vm_gettime = vm_gettime;
	vmprim.vm_muxadd = vm_muxadd;
	vmprim.vm_muxdel = vm_muxdel;
	vmprim.vm_muxmod = vm_muxmod;
	vmprim.vm_muxwait = vm_muxwait;
	vmprim.vm_sleep = vm_sleep;
	vmprim.vm_getsigfd = vm_getsigfd;
	vmprim.vm_getsig = vm_getsig;
	vmprim.vm_setsig = vm_setsig;

	moo = moo_open(&sys_mmgr, MOO_SIZEOF(xtn_t) + xtnsize, ((cfg && cfg->cmgr)? cfg->cmgr: moo_get_utf8_cmgr()), &vmprim, (cfg? cfg->gc_type: MOO_GC_TYPE_DEFAULT), errinfo);
	if (MOO_UNLIKELY(!moo)) return MOO_NULL;

	/* adjust the object size by the sizeof xtn_t so that moo_getxtn() returns the right pointer. */
	moo->_instsize += MOO_SIZEOF(xtn_t);

	xtn = GET_XTN(moo);
	if (cfg) xtn->input_cmgr = cfg->input_cmgr;
	if (!xtn->input_cmgr) xtn->input_cmgr = moo_getcmgr(moo);
	if (cfg) xtn->log_cmgr = cfg->log_cmgr;
	if (!xtn->log_cmgr) xtn->log_cmgr = moo_getcmgr(moo);

	chain (moo); /* call chain() before moo_regevtcb() as fini_moo() calls unchain() */
	reset_log_to_default (xtn);

	MOO_MEMSET (&evtcb, 0, MOO_SIZEOF(evtcb));
	evtcb.fini = fini_moo;
	evtcb.halting = halting_moo;
	moo_regevtcb (moo, &evtcb);

#if defined(EMSCRIPTEN)
{
/* TODO: this is experimental. make it proper */
moo_bitmask_t m = ~(moo_bitmask_t)0;
moo_setoption (moo, MOO_OPTION_LOG_MASK, &m);
}
#endif

	{
		moo_bitmask_t bm = 0;

		moo_getoption (moo, MOO_OPTION_TRAIT, &bm);
		/*bm |= MOO_TRAIT_NOGC;*/
		bm |= MOO_TRAIT_AWAIT_PROCS;
		moo_setoption (moo, MOO_OPTION_TRAIT, &bm);

		/* disable GC logs */
		moo_getoption (moo, MOO_OPTION_LOG_MASK, &bm);
		bm = ~MOO_LOG_GC;
		moo_setoption (moo, MOO_OPTION_LOG_MASK, &bm);
	}

	if (cfg && cfg->proc_stk_size > 0)
	{
		moo_setoption (moo, MOO_OPTION_PROCSTK_SIZE, &cfg->proc_stk_size);
	}

	if (cfg && handle_cfg_options(moo, cfg) <= -1)
	{
		if (errinfo) moo_geterrinf (moo, errinfo);
		moo_close (moo);
		return MOO_NULL;
	}

#if defined(_WIN32)
	SetUnhandledExceptionFilter (msw_exception_filter);
#endif

	return moo;
}

void moo_abortstd (moo_t* moo)
{
#if defined(USE_THREAD)
	xtn_t* xtn = GET_XTN(moo);
	write (xtn->iothr.p[1], "Q", 1);
#endif
	moo_abort (moo);
}

#if defined(MOO_INCLUDE_COMPILER)
int moo_compilestd (moo_t* moo, const moo_iostd_t* in, moo_oow_t count)
{
	xtn_t* xtn = GET_XTN(moo);
	moo_oow_t i;

	for (i = 0; i < count; i++)
	{
		xtn->in = &in[i];
		if (moo_compile(moo, input_handler) <= -1) return -1;
	}

	return 0;
}

int moo_compilefileb (moo_t* moo, const moo_bch_t* path)
{
	moo_iostd_t in;

	MOO_MEMSET (&in, 0, MOO_SIZEOF(in));
	in.type = MOO_IOSTD_FILEB;
	in.u.fileb.path = path;

	return moo_compilestd(moo, &in, 1);
}

int moo_compilefileu (moo_t* moo, const moo_uch_t* path)
{
	moo_iostd_t in;

	MOO_MEMSET (&in, 0, MOO_SIZEOF(in));
	in.type = MOO_IOSTD_FILEU;
	in.u.fileu.path = path;

	return moo_compilestd(moo, &in, 1);
}

#endif

int moo_invokebynameb (moo_t* moo, const moo_bch_t* objname, const moo_bch_t* mthname)
{
#if defined(MOO_OOCH_IS_UCH)
	int n = -1;
	moo_oocs_t o, m;

	o.ptr = moo_dupbtooocstr(moo, objname, &o.len);
	m.ptr = moo_dupbtooocstr(moo, mthname, &m.len);
	if (!o.ptr || !m.ptr) goto done;

	n = moo_invoke(moo, &o, &m);

done:
	if (m.ptr) moo_freemem (moo, m.ptr);
	if (o.ptr) moo_freemem (moo, o.ptr);

	return n;
#else
	moo_oocs_t o, m;

	o.ptr = objname;
	o.len = moo_count_bcstr(objname);

	m.ptr = objname;
	m.len = moo_count_bcstr(objname);

	return moo_invoke(moo, &o, &m);
#endif
}

void moo_rcvtickstd (moo_t* moo, int v)
{
	xtn_t* xtn = GET_XTN(moo);
	xtn->rcv_tick = v;
}

/* ========================================================================= */

static moo_uint32_t ticker_started = 0;

void moo_start_ticker (void)
{
	if (++ticker_started == 1)
	{
		start_ticker ();
	}
}

void moo_stop_ticker (void)
{
	if (ticker_started > 0 && --ticker_started == 0)
	{
		stop_ticker ();
	}
}


/* ========================================================================== */

#if defined(_WIN32)
static BOOL WINAPI handle_term (DWORD ctrl_type)
{
	if (ctrl_type == CTRL_C_EVENT || ctrl_type == CTRL_CLOSE_EVENT)
	{
		abort_all_moos (SIGINT);
		return TRUE;
	}

	return FALSE;
}

void moo_catch_termreq (void)
{
	SetConsoleCtrlHandler (handle_term, TRUE);
}

void moo_uncatch_termreq (void)
{
	SetConsoleCtrlHandler (handle_term, FALSE);
}

#elif defined(__OS2__)

static EXCEPTIONREGISTRATIONRECORD os2_excrr = { 0 };

static ULONG APIENTRY handle_term (
	PEXCEPTIONREPORTRECORD p1,
	PEXCEPTIONREGISTRATIONRECORD p2,
	PCONTEXTRECORD p3,
	PVOID pv)
{
	if (p1->ExceptionNum == XCPT_SIGNAL)
	{
		if (p1->ExceptionInfo[0] == XCPT_SIGNAL_INTR ||
		    p1->ExceptionInfo[0] == XCPT_SIGNAL_KILLPROC ||
		    p1->ExceptionInfo[0] == XCPT_SIGNAL_BREAK)
		{
			abort_all_moos (SIGINT);
			return (DosAcknowledgeSignalException(p1->ExceptionInfo[0]) != NO_ERROR)? 1: XCPT_CONTINUE_EXECUTION;
		}
	}

	return XCPT_CONTINUE_SEARCH; /* exception not resolved */
}

void moo_catch_termreq (void)
{
	os2_excrr.ExceptionHandler = (ERR)handle_term;
	DosSetExceptionHandler (&os2_excrr); /* TODO: check if NO_ERROR is returned */
}

void moo_uncatch_termreq (void)
{
	DosUnsetExceptionHandler (&os2_excrr);
}

#elif defined(__DOS__)

/*#define IRQ_TERM 0x23*/
/*#define IRQ_TERM 0x1B*/
#define IRQ_TERM 0x9

#if defined(_INTELC32_)
static void (*dos_prev_int23_handler) (void);
#pragma interrupt(dos_int23_handler)
static void dos_int23_handler (void)
#else
static void (__interrupt *dos_prev_int23_handler) (void);
static void __interrupt dos_int23_handler (void)
#endif
{
#if (IRQ_TERM == 0x23) && defined(_INTELC32_)
	/* note this code for _INTELC32_ doesn't seem to work properly
	 * unless the program is waiting on getch() or something similar */
	/* prevent the DOS interrupt handler from being called */
	_XSTACK* stk = (_XSTACK*)_get_stk_frame();
	stk->opts |= _STK_NOINT;
	abort_all_moos (SIGINT);
	/* if i call the previous handler, it's likely to kill the application.
	 * so i don't chain-call the previous handler. but another call could
	 * have changed the handler already to something else. then it would be
	 * better to chain-call it. TODO: find a way to chain-call it safely */
	/*_chain_intr (dos_prev_int23_handler);*/
#else


	#if 0
	static int extended = 0;
	static int keyboard[255] = { 0, };
	moo_uint8_t sc, status;
	/* TODO: determine if the key pressed is ctrl-C or ctrl-break ... */

	sc = inp(0x60);
	/*status = inp(0x61);*/
	if (sc == 0xE0)
	{
		/* extended key prefix */
		extended = 1;
	}
	else if (sc == 0xE1)
	{
		/* pause key */
	}
	else
	{
		if (sc & 0x80)
		{
			/* key release */
			sc = sc & 0x7F;
			keyboard[sc] = 0;
			/*printf ("%key released ... %x\n", sc);*/
		}
		else
		{
			keyboard[sc] = 1;
			/*printf ("%key pressed ... %x %c\n", sc, sc);*/
			abort_all_moos (SIGINT);
		}

		extended = 0;
	}

	/*_chain_intr (dos_prev_int23_handler);*/
	outp (0x20, 0x20);
	#else
	abort_all_moos (SIGINT);
	_chain_intr (dos_prev_int23_handler);
	#endif
#endif
}

void moo_catch_termreq (void)
{
	dos_prev_int23_handler = _dos_getvect(IRQ_TERM);
	_dos_setvect (IRQ_TERM, dos_int23_handler);
}

void moo_uncatch_termreq (void)
{
	_dos_setvect (IRQ_TERM, dos_prev_int23_handler);
	dos_prev_int23_handler = MOO_NULL;
}

#else

void moo_catch_termreq (void)
{
	set_signal_handler(SIGTERM, abort_all_moos, 0);
	set_signal_handler(SIGHUP, abort_all_moos, 0);
	set_signal_handler(SIGINT, abort_all_moos, 0);
	set_signal_handler(SIGPIPE, do_nothing, 0);
}

void moo_uncatch_termreq (void)
{
	unset_signal_handler(SIGTERM);
	unset_signal_handler(SIGHUP);
	unset_signal_handler(SIGINT);
	unset_signal_handler(SIGPIPE);
}

#endif