/*
** This file is in the public domain, so clarified as of
** 2006-07-17 by Arthur David Olson.
*/

#include "version.h"
#include "private.h"
#include "tzfile.h"

#include <fcntl.h>
#include <locale.h>
#include <stdarg.h>
#include <stddef.h>

#define	ZIC_VERSION_PRE_2013 '2'
#define	ZIC_VERSION	'3'

typedef int_fast64_t	zic_t;
#define ZIC_MIN INT_FAST64_MIN
#define ZIC_MAX INT_FAST64_MAX
#define PRIdZIC PRIdFAST64
#define SCNdZIC SCNdFAST64

#ifndef ZIC_MAX_ABBR_LEN_WO_WARN
#define ZIC_MAX_ABBR_LEN_WO_WARN	6
#endif /* !defined ZIC_MAX_ABBR_LEN_WO_WARN */

#ifdef HAVE_DIRECT_H
# include <direct.h>
# include <io.h>
# undef mkdir
# define mkdir(name, mode) _mkdir(name)
#endif

#if HAVE_SYS_STAT_H
#include <sys/stat.h>
#endif
#ifdef S_IRUSR
#define MKDIR_UMASK (S_IRUSR|S_IWUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
#else
#define MKDIR_UMASK 0755
#endif

/* The maximum ptrdiff_t value, for pre-C99 platforms.  */
#ifndef PTRDIFF_MAX
static ptrdiff_t const PTRDIFF_MAX = MAXVAL(ptrdiff_t, TYPE_BIT(ptrdiff_t));
#endif

/* The type for line numbers.  Use PRIdMAX to format them; formerly
   there was also "#define PRIdLINENO PRIdMAX" and formats used
   PRIdLINENO, but xgettext cannot grok that.  */
typedef intmax_t lineno;

struct rule {
	const char *	r_filename;
	lineno		r_linenum;
	const char *	r_name;

	zic_t		r_loyear;	/* for example, 1986 */
	zic_t		r_hiyear;	/* for example, 1986 */
	const char *	r_yrtype;
	bool		r_lowasnum;
	bool		r_hiwasnum;

	int		r_month;	/* 0..11 */

	int		r_dycode;	/* see below */
	int		r_dayofmonth;
	int		r_wday;

	zic_t		r_tod;		/* time from midnight */
	bool		r_todisstd;	/* above is standard time if 1 */
					/* or wall clock time if 0 */
	bool		r_todisgmt;	/* above is GMT if 1 */
					/* or local time if 0 */
	zic_t		r_stdoff;	/* offset from standard time */
	const char *	r_abbrvar;	/* variable part of abbreviation */

	bool		r_todo;		/* a rule to do (used in outzone) */
	zic_t		r_temp;		/* used in outzone */
};

/*
**	r_dycode		r_dayofmonth	r_wday
*/

#define DC_DOM		0	/* 1..31 */	/* unused */
#define DC_DOWGEQ	1	/* 1..31 */	/* 0..6 (Sun..Sat) */
#define DC_DOWLEQ	2	/* 1..31 */	/* 0..6 (Sun..Sat) */

struct zone {
	const char *	z_filename;
	lineno		z_linenum;

	const char *	z_name;
	zic_t		z_gmtoff;
	const char *	z_rule;
	const char *	z_format;
	char		z_format_specifier;

	zic_t		z_stdoff;

	struct rule *	z_rules;
	ptrdiff_t	z_nrules;

	struct rule	z_untilrule;
	zic_t		z_untiltime;
};

#if !HAVE_POSIX_DECLS
extern int	getopt(int argc, char * const argv[],
			const char * options);
extern int	link(const char * fromname, const char * toname);
extern char *	optarg;
extern int	optind;
#endif

#if ! HAVE_LINK
# define link(from, to) (errno = ENOTSUP, -1)
#endif
#if ! HAVE_SYMLINK
# define readlink(file, buf, size) (errno = ENOTSUP, -1)
# define symlink(from, to) (errno = ENOTSUP, -1)
# define S_ISLNK(m) 0
#endif
#ifndef AT_SYMLINK_FOLLOW
# define linkat(fromdir, from, todir, to, flag) \
    (itssymlink(from) ? (errno = ENOTSUP, -1) : link(from, to))
#endif

static void	addtt(zic_t starttime, int type);
static int	addtype(zic_t, char const *, bool, bool, bool);
static void	leapadd(zic_t, bool, int, int);
static void	adjleap(void);
static void	associate(void);
static void	dolink(const char *, const char *, bool);
static char **	getfields(char * buf);
static zic_t	gethms(const char * string, const char * errstring,
		       bool);
static void	infile(const char * filename);
static void	inleap(char ** fields, int nfields);
static void	inlink(char ** fields, int nfields);
static void	inrule(char ** fields, int nfields);
static bool	inzcont(char ** fields, int nfields);
static bool	inzone(char ** fields, int nfields);
static bool	inzsub(char **, int, bool);
static bool	itsdir(char const *);
static bool	itssymlink(char const *);
static bool	is_alpha(char a);
static char	lowerit(char);
static void	mkdirs(char const *, bool);
static void	newabbr(const char * abbr);
static zic_t	oadd(zic_t t1, zic_t t2);
static void	outzone(const struct zone * zp, ptrdiff_t ntzones);
static zic_t	rpytime(const struct rule * rp, zic_t wantedy);
static void	rulesub(struct rule * rp,
			const char * loyearp, const char * hiyearp,
			const char * typep, const char * monthp,
			const char * dayp, const char * timep);
static zic_t	tadd(zic_t t1, zic_t t2);
static bool	yearistype(zic_t year, const char * type);

/* Bound on length of what %z can expand to.  */
enum { PERCENT_Z_LEN_BOUND = sizeof "+995959" - 1 };

/* If true, work around a bug in Qt 5.6.1 and earlier, which mishandles
   tz binary files whose POSIX-TZ-style strings contain '<'; see
   QTBUG-53071 <https://bugreports.qt.io/browse/QTBUG-53071>.  This
   workaround will no longer be needed when Qt 5.6.1 and earlier are
   obsolete, say in the year 2021.  */
enum { WORK_AROUND_QTBUG_53071 = true };

static int		charcnt;
static bool		errors;
static bool		warnings;
static const char *	filename;
static int		leapcnt;
static bool		leapseen;
static zic_t		leapminyear;
static zic_t		leapmaxyear;
static lineno		linenum;
static int		max_abbrvar_len = PERCENT_Z_LEN_BOUND;
static int		max_format_len;
static zic_t		max_year;
static zic_t		min_year;
static bool		noise;
static const char *	rfilename;
static lineno		rlinenum;
static const char *	progname;
static ptrdiff_t	timecnt;
static ptrdiff_t	timecnt_alloc;
static int		typecnt;

/*
** Line codes.
*/

#define LC_RULE		0
#define LC_ZONE		1
#define LC_LINK		2
#define LC_LEAP		3

/*
** Which fields are which on a Zone line.
*/

#define ZF_NAME		1
#define ZF_GMTOFF	2
#define ZF_RULE		3
#define ZF_FORMAT	4
#define ZF_TILYEAR	5
#define ZF_TILMONTH	6
#define ZF_TILDAY	7
#define ZF_TILTIME	8
#define ZONE_MINFIELDS	5
#define ZONE_MAXFIELDS	9

/*
** Which fields are which on a Zone continuation line.
*/

#define ZFC_GMTOFF	0
#define ZFC_RULE	1
#define ZFC_FORMAT	2
#define ZFC_TILYEAR	3
#define ZFC_TILMONTH	4
#define ZFC_TILDAY	5
#define ZFC_TILTIME	6
#define ZONEC_MINFIELDS	3
#define ZONEC_MAXFIELDS	7

/*
** Which files are which on a Rule line.
*/

#define RF_NAME		1
#define RF_LOYEAR	2
#define RF_HIYEAR	3
#define RF_COMMAND	4
#define RF_MONTH	5
#define RF_DAY		6
#define RF_TOD		7
#define RF_STDOFF	8
#define RF_ABBRVAR	9
#define RULE_FIELDS	10

/*
** Which fields are which on a Link line.
*/

#define LF_FROM		1
#define LF_TO		2
#define LINK_FIELDS	3

/*
** Which fields are which on a Leap line.
*/

#define LP_YEAR		1
#define LP_MONTH	2
#define LP_DAY		3
#define LP_TIME		4
#define LP_CORR		5
#define LP_ROLL		6
#define LEAP_FIELDS	7

/*
** Year synonyms.
*/

#define YR_MINIMUM	0
#define YR_MAXIMUM	1
#define YR_ONLY		2

static struct rule *	rules;
static ptrdiff_t	nrules;	/* number of rules */
static ptrdiff_t	nrules_alloc;

static struct zone *	zones;
static ptrdiff_t	nzones;	/* number of zones */
static ptrdiff_t	nzones_alloc;

struct link {
	const char *	l_filename;
	lineno		l_linenum;
	const char *	l_from;
	const char *	l_to;
};

static struct link *	links;
static ptrdiff_t	nlinks;
static ptrdiff_t	nlinks_alloc;

struct lookup {
	const char *	l_word;
	const int	l_value;
};

static struct lookup const *	byword(const char * string,
					const struct lookup * lp);

static struct lookup const	line_codes[] = {
	{ "Rule",	LC_RULE },
	{ "Zone",	LC_ZONE },
	{ "Link",	LC_LINK },
	{ "Leap",	LC_LEAP },
	{ NULL,		0}
};

static struct lookup const	mon_names[] = {
	{ "January",	TM_JANUARY },
	{ "February",	TM_FEBRUARY },
	{ "March",	TM_MARCH },
	{ "April",	TM_APRIL },
	{ "May",	TM_MAY },
	{ "June",	TM_JUNE },
	{ "July",	TM_JULY },
	{ "August",	TM_AUGUST },
	{ "September",	TM_SEPTEMBER },
	{ "October",	TM_OCTOBER },
	{ "November",	TM_NOVEMBER },
	{ "December",	TM_DECEMBER },
	{ NULL,		0 }
};

static struct lookup const	wday_names[] = {
	{ "Sunday",	TM_SUNDAY },
	{ "Monday",	TM_MONDAY },
	{ "Tuesday",	TM_TUESDAY },
	{ "Wednesday",	TM_WEDNESDAY },
	{ "Thursday",	TM_THURSDAY },
	{ "Friday",	TM_FRIDAY },
	{ "Saturday",	TM_SATURDAY },
	{ NULL,		0 }
};

static struct lookup const	lasts[] = {
	{ "last-Sunday",	TM_SUNDAY },
	{ "last-Monday",	TM_MONDAY },
	{ "last-Tuesday",	TM_TUESDAY },
	{ "last-Wednesday",	TM_WEDNESDAY },
	{ "last-Thursday",	TM_THURSDAY },
	{ "last-Friday",	TM_FRIDAY },
	{ "last-Saturday",	TM_SATURDAY },
	{ NULL,			0 }
};

static struct lookup const	begin_years[] = {
	{ "minimum",	YR_MINIMUM },
	{ "maximum",	YR_MAXIMUM },
	{ NULL,		0 }
};

static struct lookup const	end_years[] = {
	{ "minimum",	YR_MINIMUM },
	{ "maximum",	YR_MAXIMUM },
	{ "only",	YR_ONLY },
	{ NULL,		0 }
};

static struct lookup const	leap_types[] = {
	{ "Rolling",	true },
	{ "Stationary",	false },
	{ NULL,		0 }
};

static const int	len_months[2][MONSPERYEAR] = {
	{ 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 },
	{ 31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 }
};

static const int	len_years[2] = {
	DAYSPERNYEAR, DAYSPERLYEAR
};

static struct attype {
	zic_t		at;
	bool		dontmerge;
	unsigned char	type;
} *			attypes;
static zic_t		gmtoffs[TZ_MAX_TYPES];
static char		isdsts[TZ_MAX_TYPES];
static unsigned char	abbrinds[TZ_MAX_TYPES];
static bool		ttisstds[TZ_MAX_TYPES];
static bool		ttisgmts[TZ_MAX_TYPES];
static char		chars[TZ_MAX_CHARS];
static zic_t		trans[TZ_MAX_LEAPS];
static zic_t		corr[TZ_MAX_LEAPS];
static char		roll[TZ_MAX_LEAPS];

/*
** Memory allocation.
*/

static _Noreturn void
memory_exhausted(const char *msg)
{
	fprintf(stderr, _("%s: Memory exhausted: %s\n"), progname, msg);
	exit(EXIT_FAILURE);
}

static ATTRIBUTE_PURE size_t
size_product(size_t nitems, size_t itemsize)
{
	if (SIZE_MAX / itemsize < nitems)
		memory_exhausted(_("size overflow"));
	return nitems * itemsize;
}

#if !HAVE_STRDUP
static char *
strdup(char const *str)
{
  char *result = malloc(strlen(str) + 1);
  return result ? strcpy(result, str) : result;
}
#endif

static ATTRIBUTE_PURE void *
memcheck(void *ptr)
{
	if (ptr == NULL)
		memory_exhausted(strerror(errno));
	return ptr;
}

static void *
emalloc(size_t size)
{
  return memcheck(malloc(size));
}

static void *
erealloc(void *ptr, size_t size)
{
  return memcheck(realloc(ptr, size));
}

static char *
ecpyalloc (char const *str)
{
  return memcheck(strdup(str));
}

static void *
growalloc(void *ptr, size_t itemsize, ptrdiff_t nitems, ptrdiff_t *nitems_alloc)
{
	if (nitems < *nitems_alloc)
		return ptr;
	else {
		ptrdiff_t nitems_max = PTRDIFF_MAX - WORK_AROUND_QTBUG_53071;
		ptrdiff_t amax = nitems_max < SIZE_MAX ? nitems_max : SIZE_MAX;
		if ((amax - 1) / 3 * 2 < *nitems_alloc)
			memory_exhausted(_("integer overflow"));
		*nitems_alloc += (*nitems_alloc >> 1) + 1;
		return erealloc(ptr, size_product(*nitems_alloc, itemsize));
	}
}

/*
** Error handling.
*/

static void
eats(char const *name, lineno num, char const *rname, lineno rnum)
{
	filename = name;
	linenum = num;
	rfilename = rname;
	rlinenum = rnum;
}

static void
eat(char const *name, lineno num)
{
	eats(name, num, NULL, -1);
}

static void ATTRIBUTE_FORMAT((printf, 1, 0))
verror(const char *const string, va_list args)
{
	/*
	** Match the format of "cc" to allow sh users to
	**	zic ... 2>&1 | error -t "*" -v
	** on BSD systems.
	*/
	if (filename)
	  fprintf(stderr, _("\"%s\", line %"PRIdMAX": "), filename, linenum);
	vfprintf(stderr, string, args);
	if (rfilename != NULL)
		fprintf(stderr, _(" (rule from \"%s\", line %"PRIdMAX")"),
			rfilename, rlinenum);
	fprintf(stderr, "\n");
}

static void ATTRIBUTE_FORMAT((printf, 1, 2))
error(const char *const string, ...)
{
	va_list args;
	va_start(args, string);
	verror(string, args);
	va_end(args);
	errors = true;
}

static void ATTRIBUTE_FORMAT((printf, 1, 2))
warning(const char *const string, ...)
{
	va_list args;
	fprintf(stderr, _("warning: "));
	va_start(args, string);
	verror(string, args);
	va_end(args);
	warnings = true;
}

static void
close_file(FILE *stream, char const *dir, char const *name)
{
  char const *e = (ferror(stream) ? _("I/O error")
		   : fclose(stream) != 0 ? strerror(errno) : NULL);
  if (e) {
    fprintf(stderr, "%s: %s%s%s%s%s\n", progname,
	    dir ? dir : "", dir ? "/" : "",
	    name ? name : "", name ? ": " : "",
	    e);
    exit(EXIT_FAILURE);
  }
}

static _Noreturn void
usage(FILE *stream, int status)
{
  fprintf(stream,
	  _("%s: usage is %s [ --version ] [ --help ] [ -v ] \\\n"
	    "\t[ -l localtime ] [ -p posixrules ] [ -d directory ] \\\n"
	    "\t[ -L leapseconds ] [ filename ... ]\n\n"
	    "Report bugs to %s.\n"),
	  progname, progname, REPORT_BUGS_TO);
  if (status == EXIT_SUCCESS)
    close_file(stream, NULL, NULL);
  exit(status);
}

/* Change the working directory to DIR, possibly creating DIR and its
   ancestors.  After this is done, all files are accessed with names
   relative to DIR.  */
static void
change_directory (char const *dir)
{
  if (chdir(dir) != 0) {
    int chdir_errno = errno;
    if (chdir_errno == ENOENT) {
      mkdirs(dir, false);
      chdir_errno = chdir(dir) == 0 ? 0 : errno;
    }
    if (chdir_errno != 0) {
      fprintf(stderr, _("%s: Can't chdir to %s: %s\n"),
	      progname, dir, strerror(chdir_errno));
      exit(EXIT_FAILURE);
    }
  }
}

static const char *	psxrules;
static const char *	lcltime;
static const char *	directory;
static const char *	leapsec;
static const char *	yitcommand;

int
main(int argc, char **argv)
{
	register int c, k;
	register ptrdiff_t i, j;

#ifdef S_IWGRP
	umask(umask(S_IWGRP | S_IWOTH) | (S_IWGRP | S_IWOTH));
#endif
#if HAVE_GETTEXT
	setlocale(LC_ALL, "");
#ifdef TZ_DOMAINDIR
	bindtextdomain(TZ_DOMAIN, TZ_DOMAINDIR);
#endif /* defined TEXTDOMAINDIR */
	textdomain(TZ_DOMAIN);
#endif /* HAVE_GETTEXT */
	progname = argv[0];
	if (TYPE_BIT(zic_t) < 64) {
		fprintf(stderr, "%s: %s\n", progname,
			_("wild compilation-time specification of zic_t"));
		return EXIT_FAILURE;
	}
	for (k = 1; k < argc; k++)
		if (strcmp(argv[k], "--version") == 0) {
			printf("zic %s%s\n", PKGVERSION, TZVERSION);
			close_file(stdout, NULL, NULL);
			return EXIT_SUCCESS;
		} else if (strcmp(argv[k], "--help") == 0) {
			usage(stdout, EXIT_SUCCESS);
		}
	while ((c = getopt(argc, argv, "d:l:p:L:vsy:")) != EOF && c != -1)
		switch (c) {
			default:
				usage(stderr, EXIT_FAILURE);
			case 'd':
				if (directory == NULL)
					directory = optarg;
				else {
					fprintf(stderr,
_("%s: More than one -d option specified\n"),
						progname);
					return EXIT_FAILURE;
				}
				break;
			case 'l':
				if (lcltime == NULL)
					lcltime = optarg;
				else {
					fprintf(stderr,
_("%s: More than one -l option specified\n"),
						progname);
					return EXIT_FAILURE;
				}
				break;
			case 'p':
				if (psxrules == NULL)
					psxrules = optarg;
				else {
					fprintf(stderr,
_("%s: More than one -p option specified\n"),
						progname);
					return EXIT_FAILURE;
				}
				break;
			case 'y':
				if (yitcommand == NULL)
					yitcommand = optarg;
				else {
					fprintf(stderr,
_("%s: More than one -y option specified\n"),
						progname);
					return EXIT_FAILURE;
				}
				break;
			case 'L':
				if (leapsec == NULL)
					leapsec = optarg;
				else {
					fprintf(stderr,
_("%s: More than one -L option specified\n"),
						progname);
					return EXIT_FAILURE;
				}
				break;
			case 'v':
				noise = true;
				break;
			case 's':
				warning(_("-s ignored"));
				break;
		}
	if (optind == argc - 1 && strcmp(argv[optind], "=") == 0)
		usage(stderr, EXIT_FAILURE);	/* usage message by request */
	if (directory == NULL)
		directory = TZDIR;
	if (yitcommand == NULL)
		yitcommand = "yearistype";

	if (optind < argc && leapsec != NULL) {
		infile(leapsec);
		adjleap();
	}

	for (k = optind; k < argc; k++)
		infile(argv[k]);
	if (errors)
		return EXIT_FAILURE;
	associate();
	change_directory(directory);
	for (i = 0; i < nzones; i = j) {
		/*
		** Find the next non-continuation zone entry.
		*/
		for (j = i + 1; j < nzones && zones[j].z_name == NULL; ++j)
			continue;
		outzone(&zones[i], j - i);
	}
	/*
	** Make links.
	*/
	for (i = 0; i < nlinks; ++i) {
		eat(links[i].l_filename, links[i].l_linenum);
		dolink(links[i].l_from, links[i].l_to, false);
		if (noise)
			for (j = 0; j < nlinks; ++j)
				if (strcmp(links[i].l_to,
					links[j].l_from) == 0)
						warning(_("link to link"));
	}
	if (lcltime != NULL) {
		eat(_("command line"), 1);
		dolink(lcltime, TZDEFAULT, true);
	}
	if (psxrules != NULL) {
		eat(_("command line"), 1);
		dolink(psxrules, TZDEFRULES, true);
	}
	if (warnings && (ferror(stderr) || fclose(stderr) != 0))
	  return EXIT_FAILURE;
	return errors ? EXIT_FAILURE : EXIT_SUCCESS;
}

static bool
componentcheck(char const *name, char const *component,
	       char const *component_end)
{
	enum { component_len_max = 14 };
	ptrdiff_t component_len = component_end - component;
	if (component_len == 0) {
	  if (!*name)
	    error (_("empty file name"));
	  else
	    error (_(component == name
		     ? "file name '%s' begins with '/'"
		     : *component_end
		     ? "file name '%s' contains '//'"
		     : "file name '%s' ends with '/'"),
		   name);
	  return false;
	}
	if (0 < component_len && component_len <= 2
	    && component[0] == '.' && component_end[-1] == '.') {
	  int len = component_len;
	  error(_("file name '%s' contains '%.*s' component"),
		name, len, component);
	  return false;
	}
	if (noise) {
	  if (0 < component_len && component[0] == '-')
	    warning(_("file name '%s' component contains leading '-'"),
		    name);
	  if (component_len_max < component_len)
	    warning(_("file name '%s' contains overlength component"
		      " '%.*s...'"),
		    name, component_len_max, component);
	}
	return true;
}

static bool
namecheck(const char *name)
{
	register char const *cp;

	/* Benign characters in a portable file name.  */
	static char const benign[] =
	  "-/_"
	  "abcdefghijklmnopqrstuvwxyz"
	  "ABCDEFGHIJKLMNOPQRSTUVWXYZ";

	/* Non-control chars in the POSIX portable character set,
	   excluding the benign characters.  */
	static char const printable_and_not_benign[] =
	  " !\"#$%&'()*+,.0123456789:;<=>?@[\\]^`{|}~";

	register char const *component = name;
	for (cp = name; *cp; cp++) {
		unsigned char c = *cp;
		if (noise && !strchr(benign, c)) {
			warning((strchr(printable_and_not_benign, c)
				 ? _("file name '%s' contains byte '%c'")
				 : _("file name '%s' contains byte '\\%o'")),
				name, c);
		}
		if (c == '/') {
			if (!componentcheck(name, component, cp))
			  return false;
			component = cp + 1;
		}
	}
	return componentcheck(name, component, cp);
}

/* Create symlink contents suitable for symlinking FROM to TO, as a
   freshly allocated string.  FROM should be a relative file name, and
   is relative to the global variable DIRECTORY.  TO can be either
   relative or absolute.  */
static char *
relname(char const *from, char const *to)
{
  size_t i, taillen, dotdotetcsize;
  size_t dir_len = 0, dotdots = 0, linksize = SIZE_MAX;
  char const *f = from;
  char *result = NULL;
  if (*to == '/') {
    /* Make F absolute too.  */
    size_t len = strlen(directory);
    bool needslash = len && directory[len - 1] != '/';
    linksize = len + needslash + strlen(from) + 1;
    f = result = emalloc(linksize);
    strcpy(result, directory);
    result[len] = '/';
    strcpy(result + len + needslash, from);
  }
  for (i = 0; f[i] && f[i] == to[i]; i++)
    if (f[i] == '/')
      dir_len = i + 1;
  for (; to[i]; i++)
    dotdots += to[i] == '/' && to[i - 1] != '/';
  taillen = strlen(f + dir_len);
  dotdotetcsize = 3 * dotdots + taillen + 1;
  if (dotdotetcsize <= linksize) {
    if (!result)
      result = emalloc(dotdotetcsize);
    for (i = 0; i < dotdots; i++)
      memcpy(result + 3 * i, "../", 3);
    memmove(result + 3 * dotdots, f + dir_len, taillen + 1);
  }
  return result;
}

/* Hard link FROM to TO, following any symbolic links.
   Return 0 if successful, an error number otherwise.  */
static int
hardlinkerr(char const *from, char const *to)
{
  int r = linkat(AT_FDCWD, from, AT_FDCWD, to, AT_SYMLINK_FOLLOW);
  return r == 0 ? 0 : errno;
}

static void
dolink(char const *fromfield, char const *tofield, bool staysymlink)
{
	bool todirs_made = false;
	int link_errno;

	/*
	** We get to be careful here since
	** there's a fair chance of root running us.
	*/
	if (itsdir(fromfield)) {
		fprintf(stderr, _("%s: link from %s/%s failed: %s\n"),
			progname, directory, fromfield, strerror(EPERM));
		exit(EXIT_FAILURE);
	}
	if (staysymlink)
	  staysymlink = itssymlink(tofield);
	if (remove(tofield) == 0)
	  todirs_made = true;
	else if (errno != ENOENT) {
	  char const *e = strerror(errno);
	  fprintf(stderr, _("%s: Can't remove %s/%s: %s\n"),
		  progname, directory, tofield, e);
	  exit(EXIT_FAILURE);
	}
	link_errno = staysymlink ? ENOTSUP : hardlinkerr(fromfield, tofield);
	if (link_errno == ENOENT && !todirs_made) {
	  mkdirs(tofield, true);
	  todirs_made = true;
	  link_errno = hardlinkerr(fromfield, tofield);
	}
	if (link_errno != 0) {
	  bool absolute = *fromfield == '/';
	  char *linkalloc = absolute ? NULL : relname(fromfield, tofield);
	  char const *contents = absolute ? fromfield : linkalloc;
	  int symlink_errno = symlink(contents, tofield) == 0 ? 0 : errno;
	  if (symlink_errno == ENOENT && !todirs_made) {
	    mkdirs(tofield, true);
	    symlink_errno = symlink(contents, tofield) == 0 ? 0 : errno;
	  }
	  free(linkalloc);
	  if (symlink_errno == 0) {
	    if (link_errno != ENOTSUP)
	      warning(_("symbolic link used because hard link failed: %s"),
		      strerror(link_errno));
	  } else {
	    FILE *fp, *tp;
	    int c;
	    fp = fopen(fromfield, "rb");
	    if (!fp) {
	      char const *e = strerror(errno);
	      fprintf(stderr, _("%s: Can't read %s/%s: %s\n"),
		      progname, directory, fromfield, e);
	      exit(EXIT_FAILURE);
	    }
	    tp = fopen(tofield, "wb");
	    if (!tp) {
	      char const *e = strerror(errno);
	      fprintf(stderr, _("%s: Can't create %s/%s: %s\n"),
		      progname, directory, tofield, e);
	      exit(EXIT_FAILURE);
	    }
	    while ((c = getc(fp)) != EOF)
	      putc(c, tp);
	    close_file(fp, directory, fromfield);
	    close_file(tp, directory, tofield);
	    if (link_errno != ENOTSUP)
	      warning(_("copy used because hard link failed: %s"),
		      strerror(link_errno));
	    else if (symlink_errno != ENOTSUP)
	      warning(_("copy used because symbolic link failed: %s"),
		      strerror(symlink_errno));
	  }
	}
}

#define TIME_T_BITS_IN_FILE	64

static zic_t const min_time = MINVAL(zic_t, TIME_T_BITS_IN_FILE);
static zic_t const max_time = MAXVAL(zic_t, TIME_T_BITS_IN_FILE);

/* Estimated time of the Big Bang, in seconds since the POSIX epoch.
   rounded downward to the negation of a power of two that is
   comfortably outside the error bounds.

   For the time of the Big Bang, see:

   Ade PAR, Aghanim N, Armitage-Caplan C et al.  Planck 2013 results.
   I. Overview of products and scientific results.
   arXiv:1303.5062 2013-03-20 20:10:01 UTC
   <http://arxiv.org/pdf/1303.5062v1> [PDF]

   Page 36, Table 9, row Age/Gyr, column Planck+WP+highL+BAO 68% limits
   gives the value 13.798 plus-or-minus 0.037 billion years.
   Multiplying this by 1000000000 and then by 31557600 (the number of
   seconds in an astronomical year) gives a value that is comfortably
   less than 2**59, so BIG_BANG is - 2**59.

   BIG_BANG is approximate, and may change in future versions.
   Please do not rely on its exact value.  */

#ifndef BIG_BANG
#define BIG_BANG (- (1LL << 59))
#endif

/* If true, work around GNOME bug 730332
   <https://bugzilla.gnome.org/show_bug.cgi?id=730332>
   by refusing to output time stamps before BIG_BANG.
   Such time stamps are physically suspect anyway.

   The GNOME bug is scheduled to be fixed in GNOME 3.22, and if so
   this workaround will no longer be needed when GNOME 3.21 and
   earlier are obsolete, say in the year 2021.  */
enum { WORK_AROUND_GNOME_BUG_730332 = true };

static const zic_t early_time = (WORK_AROUND_GNOME_BUG_730332
				 ? BIG_BANG
				 : MINVAL(zic_t, TIME_T_BITS_IN_FILE));

/* Return true if NAME is a directory.  */
static bool
itsdir(char const *name)
{
	struct stat st;
	int res = stat(name, &st);
#ifdef S_ISDIR
	if (res == 0)
		return S_ISDIR(st.st_mode) != 0;
#endif
	if (res == 0 || errno == EOVERFLOW) {
		size_t n = strlen(name);
		char *nameslashdot = emalloc(n + 3);
		bool dir;
		memcpy(nameslashdot, name, n);
		strcpy(&nameslashdot[n], &"/."[! (n && name[n - 1] != '/')]);
		dir = stat(nameslashdot, &st) == 0 || errno == EOVERFLOW;
		free(nameslashdot);
		return dir;
	}
	return false;
}

/* Return true if NAME is a symbolic link.  */
static bool
itssymlink(char const *name)
{
  char c;
  return 0 <= readlink(name, &c, 1);
}

/*
** Associate sets of rules with zones.
*/

/*
** Sort by rule name.
*/

static int
rcomp(const void *cp1, const void *cp2)
{
	return strcmp(((const struct rule *) cp1)->r_name,
		((const struct rule *) cp2)->r_name);
}

static void
associate(void)
{
	register struct zone *	zp;
	register struct rule *	rp;
	register ptrdiff_t i, j, base, out;

	if (nrules != 0) {
		qsort(rules, nrules, sizeof *rules, rcomp);
		for (i = 0; i < nrules - 1; ++i) {
			if (strcmp(rules[i].r_name,
				rules[i + 1].r_name) != 0)
					continue;
			if (strcmp(rules[i].r_filename,
				rules[i + 1].r_filename) == 0)
					continue;
			eat(rules[i].r_filename, rules[i].r_linenum);
			warning(_("same rule name in multiple files"));
			eat(rules[i + 1].r_filename, rules[i + 1].r_linenum);
			warning(_("same rule name in multiple files"));
			for (j = i + 2; j < nrules; ++j) {
				if (strcmp(rules[i].r_name,
					rules[j].r_name) != 0)
						break;
				if (strcmp(rules[i].r_filename,
					rules[j].r_filename) == 0)
						continue;
				if (strcmp(rules[i + 1].r_filename,
					rules[j].r_filename) == 0)
						continue;
				break;
			}
			i = j - 1;
		}
	}
	for (i = 0; i < nzones; ++i) {
		zp = &zones[i];
		zp->z_rules = NULL;
		zp->z_nrules = 0;
	}
	for (base = 0; base < nrules; base = out) {
		rp = &rules[base];
		for (out = base + 1; out < nrules; ++out)
			if (strcmp(rp->r_name, rules[out].r_name) != 0)
				break;
		for (i = 0; i < nzones; ++i) {
			zp = &zones[i];
			if (strcmp(zp->z_rule, rp->r_name) != 0)
				continue;
			zp->z_rules = rp;
			zp->z_nrules = out - base;
		}
	}
	for (i = 0; i < nzones; ++i) {
		zp = &zones[i];
		if (zp->z_nrules == 0) {
			/*
			** Maybe we have a local standard time offset.
			*/
			eat(zp->z_filename, zp->z_linenum);
			zp->z_stdoff = gethms(zp->z_rule, _("unruly zone"),
				true);
			/*
			** Note, though, that if there's no rule,
			** a '%s' in the format is a bad thing.
			*/
			if (zp->z_format_specifier == 's')
				error("%s", _("%s in ruleless zone"));
		}
	}
	if (errors)
		exit(EXIT_FAILURE);
}

static void
infile(const char *name)
{
	register FILE *			fp;
	register char **		fields;
	register char *			cp;
	register const struct lookup *	lp;
	register int			nfields;
	register bool			wantcont;
	register lineno			num;
	char				buf[BUFSIZ];

	if (strcmp(name, "-") == 0) {
		name = _("standard input");
		fp = stdin;
	} else if ((fp = fopen(name, "r")) == NULL) {
		const char *e = strerror(errno);

		fprintf(stderr, _("%s: Can't open %s: %s\n"),
			progname, name, e);
		exit(EXIT_FAILURE);
	}
	wantcont = false;
	for (num = 1; ; ++num) {
		eat(name, num);
		if (fgets(buf, sizeof buf, fp) != buf)
			break;
		cp = strchr(buf, '\n');
		if (cp == NULL) {
			error(_("line too long"));
			exit(EXIT_FAILURE);
		}
		*cp = '\0';
		fields = getfields(buf);
		nfields = 0;
		while (fields[nfields] != NULL) {
			static char	nada;

			if (strcmp(fields[nfields], "-") == 0)
				fields[nfields] = &nada;
			++nfields;
		}
		if (nfields == 0) {
			/* nothing to do */
		} else if (wantcont) {
			wantcont = inzcont(fields, nfields);
		} else {
			lp = byword(fields[0], line_codes);
			if (lp == NULL)
				error(_("input line of unknown type"));
			else switch (lp->l_value) {
				case LC_RULE:
					inrule(fields, nfields);
					wantcont = false;
					break;
				case LC_ZONE:
					wantcont = inzone(fields, nfields);
					break;
				case LC_LINK:
					inlink(fields, nfields);
					wantcont = false;
					break;
				case LC_LEAP:
					if (name != leapsec)
					  warning(_("%s: Leap line in non leap"
						    " seconds file %s"),
						  progname, name);
					else	inleap(fields, nfields);
					wantcont = false;
					break;
				default:	/* "cannot happen" */
					fprintf(stderr,
_("%s: panic: Invalid l_value %d\n"),
						progname, lp->l_value);
					exit(EXIT_FAILURE);
			}
		}
		free(fields);
	}
	close_file(fp, NULL, filename);
	if (wantcont)
		error(_("expected continuation line not found"));
}

/*
** Convert a string of one of the forms
**	h	-h	hh:mm	-hh:mm	hh:mm:ss	-hh:mm:ss
** into a number of seconds.
** A null string maps to zero.
** Call error with errstring and return zero on errors.
*/

static zic_t
gethms(char const *string, char const *errstring, bool signable)
{
	zic_t	hh;
	int	mm, ss, sign;
	char xs;

	if (string == NULL || *string == '\0')
		return 0;
	if (!signable)
		sign = 1;
	else if (*string == '-') {
		sign = -1;
		++string;
	} else	sign = 1;
	if (sscanf(string, "%"SCNdZIC"%c", &hh, &xs) == 1)
		mm = ss = 0;
	else if (sscanf(string, "%"SCNdZIC":%d%c", &hh, &mm, &xs) == 2)
		ss = 0;
	else if (sscanf(string, "%"SCNdZIC":%d:%d%c", &hh, &mm, &ss, &xs)
		 != 3) {
			error("%s", errstring);
			return 0;
	}
	if (hh < 0 ||
		mm < 0 || mm >= MINSPERHOUR ||
		ss < 0 || ss > SECSPERMIN) {
			error("%s", errstring);
			return 0;
	}
	if (ZIC_MAX / SECSPERHOUR < hh) {
		error(_("time overflow"));
		return 0;
	}
	if (noise && (hh > HOURSPERDAY ||
		(hh == HOURSPERDAY && (mm != 0 || ss != 0))))
warning(_("values over 24 hours not handled by pre-2007 versions of zic"));
	return oadd(sign * hh * SECSPERHOUR,
		    sign * (mm * SECSPERMIN + ss));
}

static void
inrule(char **fields, int nfields)
{
	static struct rule	r;

	if (nfields != RULE_FIELDS) {
		error(_("wrong number of fields on Rule line"));
		return;
	}
	if (*fields[RF_NAME] == '\0') {
		error(_("nameless rule"));
		return;
	}
	r.r_filename = filename;
	r.r_linenum = linenum;
	r.r_stdoff = gethms(fields[RF_STDOFF], _("invalid saved time"), true);
	rulesub(&r, fields[RF_LOYEAR], fields[RF_HIYEAR], fields[RF_COMMAND],
		fields[RF_MONTH], fields[RF_DAY], fields[RF_TOD]);
	r.r_name = ecpyalloc(fields[RF_NAME]);
	r.r_abbrvar = ecpyalloc(fields[RF_ABBRVAR]);
	if (max_abbrvar_len < strlen(r.r_abbrvar))
		max_abbrvar_len = strlen(r.r_abbrvar);
	rules = growalloc(rules, sizeof *rules, nrules, &nrules_alloc);
	rules[nrules++] = r;
}

static bool
inzone(char **fields, int nfields)
{
	register ptrdiff_t i;

	if (nfields < ZONE_MINFIELDS || nfields > ZONE_MAXFIELDS) {
		error(_("wrong number of fields on Zone line"));
		return false;
	}
	if (strcmp(fields[ZF_NAME], TZDEFAULT) == 0 && lcltime != NULL) {
		error(
_("\"Zone %s\" line and -l option are mutually exclusive"),
			TZDEFAULT);
		return false;
	}
	if (strcmp(fields[ZF_NAME], TZDEFRULES) == 0 && psxrules != NULL) {
		error(
_("\"Zone %s\" line and -p option are mutually exclusive"),
			TZDEFRULES);
		return false;
	}
	for (i = 0; i < nzones; ++i)
		if (zones[i].z_name != NULL &&
			strcmp(zones[i].z_name, fields[ZF_NAME]) == 0) {
				error(_("duplicate zone name %s"
					" (file \"%s\", line %"PRIdMAX")"),
					fields[ZF_NAME],
					zones[i].z_filename,
					zones[i].z_linenum);
				return false;
		}
	return inzsub(fields, nfields, false);
}

static bool
inzcont(char **fields, int nfields)
{
	if (nfields < ZONEC_MINFIELDS || nfields > ZONEC_MAXFIELDS) {
		error(_("wrong number of fields on Zone continuation line"));
		return false;
	}
	return inzsub(fields, nfields, true);
}

static bool
inzsub(char **fields, int nfields, bool iscont)
{
	register char *		cp;
	char *			cp1;
	static struct zone	z;
	register int		i_gmtoff, i_rule, i_format;
	register int		i_untilyear, i_untilmonth;
	register int		i_untilday, i_untiltime;
	register bool		hasuntil;

	if (iscont) {
		i_gmtoff = ZFC_GMTOFF;
		i_rule = ZFC_RULE;
		i_format = ZFC_FORMAT;
		i_untilyear = ZFC_TILYEAR;
		i_untilmonth = ZFC_TILMONTH;
		i_untilday = ZFC_TILDAY;
		i_untiltime = ZFC_TILTIME;
		z.z_name = NULL;
	} else if (!namecheck(fields[ZF_NAME]))
		return false;
	else {
		i_gmtoff = ZF_GMTOFF;
		i_rule = ZF_RULE;
		i_format = ZF_FORMAT;
		i_untilyear = ZF_TILYEAR;
		i_untilmonth = ZF_TILMONTH;
		i_untilday = ZF_TILDAY;
		i_untiltime = ZF_TILTIME;
		z.z_name = ecpyalloc(fields[ZF_NAME]);
	}
	z.z_filename = filename;
	z.z_linenum = linenum;
	z.z_gmtoff = gethms(fields[i_gmtoff], _("invalid UT offset"), true);
	if ((cp = strchr(fields[i_format], '%')) != 0) {
		if ((*++cp != 's' && *cp != 'z') || strchr(cp, '%')
		    || strchr(fields[i_format], '/')) {
			error(_("invalid abbreviation format"));
			return false;
		}
	}
	z.z_rule = ecpyalloc(fields[i_rule]);
	z.z_format = cp1 = ecpyalloc(fields[i_format]);
	z.z_format_specifier = cp ? *cp : '\0';
	if (z.z_format_specifier == 'z') {
	  if (noise)
	    warning(_("format '%s' not handled by pre-2015 versions of zic"),
		    z.z_format);
	  cp1[cp - fields[i_format]] = 's';
	}
	if (max_format_len < strlen(z.z_format))
		max_format_len = strlen(z.z_format);
	hasuntil = nfields > i_untilyear;
	if (hasuntil) {
		z.z_untilrule.r_filename = filename;
		z.z_untilrule.r_linenum = linenum;
		rulesub(&z.z_untilrule,
			fields[i_untilyear],
			"only",
			"",
			(nfields > i_untilmonth) ?
			fields[i_untilmonth] : "Jan",
			(nfields > i_untilday) ? fields[i_untilday] : "1",
			(nfields > i_untiltime) ? fields[i_untiltime] : "0");
		z.z_untiltime = rpytime(&z.z_untilrule,
			z.z_untilrule.r_loyear);
		if (iscont && nzones > 0 &&
			z.z_untiltime > min_time &&
			z.z_untiltime < max_time &&
			zones[nzones - 1].z_untiltime > min_time &&
			zones[nzones - 1].z_untiltime < max_time &&
			zones[nzones - 1].z_untiltime >= z.z_untiltime) {
				error(_(
"Zone continuation line end time is not after end time of previous line"
					));
				return false;
		}
	}
	zones = growalloc(zones, sizeof *zones, nzones, &nzones_alloc);
	zones[nzones++] = z;
	/*
	** If there was an UNTIL field on this line,
	** there's more information about the zone on the next line.
	*/
	return hasuntil;
}

static void
inleap(char **fields, int nfields)
{
	register const char *		cp;
	register const struct lookup *	lp;
	register zic_t			i, j;
	zic_t				year;
	int				month, day;
	zic_t				dayoff, tod;
	zic_t				t;
	char xs;

	if (nfields != LEAP_FIELDS) {
		error(_("wrong number of fields on Leap line"));
		return;
	}
	dayoff = 0;
	cp = fields[LP_YEAR];
	if (sscanf(cp, "%"SCNdZIC"%c", &year, &xs) != 1) {
		/*
		** Leapin' Lizards!
		*/
		error(_("invalid leaping year"));
		return;
	}
	if (!leapseen || leapmaxyear < year)
		leapmaxyear = year;
	if (!leapseen || leapminyear > year)
		leapminyear = year;
	leapseen = true;
	j = EPOCH_YEAR;
	while (j != year) {
		if (year > j) {
			i = len_years[isleap(j)];
			++j;
		} else {
			--j;
			i = -len_years[isleap(j)];
		}
		dayoff = oadd(dayoff, i);
	}
	if ((lp = byword(fields[LP_MONTH], mon_names)) == NULL) {
		error(_("invalid month name"));
		return;
	}
	month = lp->l_value;
	j = TM_JANUARY;
	while (j != month) {
		i = len_months[isleap(year)][j];
		dayoff = oadd(dayoff, i);
		++j;
	}
	cp = fields[LP_DAY];
	if (sscanf(cp, "%d%c", &day, &xs) != 1 ||
		day <= 0 || day > len_months[isleap(year)][month]) {
			error(_("invalid day of month"));
			return;
	}
	dayoff = oadd(dayoff, day - 1);
	if (dayoff < min_time / SECSPERDAY) {
		error(_("time too small"));
		return;
	}
	if (dayoff > max_time / SECSPERDAY) {
		error(_("time too large"));
		return;
	}
	t = dayoff * SECSPERDAY;
	tod = gethms(fields[LP_TIME], _("invalid time of day"), false);
	cp = fields[LP_CORR];
	{
		register bool	positive;
		int		count;

		if (strcmp(cp, "") == 0) { /* infile() turns "-" into "" */
			positive = false;
			count = 1;
		} else if (strcmp(cp, "--") == 0) {
			positive = false;
			count = 2;
		} else if (strcmp(cp, "+") == 0) {
			positive = true;
			count = 1;
		} else if (strcmp(cp, "++") == 0) {
			positive = true;
			count = 2;
		} else {
			error(_("illegal CORRECTION field on Leap line"));
			return;
		}
		if ((lp = byword(fields[LP_ROLL], leap_types)) == NULL) {
			error(_(
				"illegal Rolling/Stationary field on Leap line"
				));
			return;
		}
		t = tadd(t, tod);
		if (t < early_time) {
			error(_("leap second precedes Big Bang"));
			return;
		}
		leapadd(t, positive, lp->l_value, count);
	}
}

static void
inlink(char **fields, int nfields)
{
	struct link	l;

	if (nfields != LINK_FIELDS) {
		error(_("wrong number of fields on Link line"));
		return;
	}
	if (*fields[LF_FROM] == '\0') {
		error(_("blank FROM field on Link line"));
		return;
	}
	if (! namecheck(fields[LF_TO]))
	  return;
	l.l_filename = filename;
	l.l_linenum = linenum;
	l.l_from = ecpyalloc(fields[LF_FROM]);
	l.l_to = ecpyalloc(fields[LF_TO]);
	links = growalloc(links, sizeof *links, nlinks, &nlinks_alloc);
	links[nlinks++] = l;
}

static void
rulesub(struct rule *rp, const char *loyearp, const char *hiyearp,
	const char *typep, const char *monthp, const char *dayp,
	const char *timep)
{
	register const struct lookup *	lp;
	register const char *		cp;
	register char *			dp;
	register char *			ep;
	char xs;

	if ((lp = byword(monthp, mon_names)) == NULL) {
		error(_("invalid month name"));
		return;
	}
	rp->r_month = lp->l_value;
	rp->r_todisstd = false;
	rp->r_todisgmt = false;
	dp = ecpyalloc(timep);
	if (*dp != '\0') {
		ep = dp + strlen(dp) - 1;
		switch (lowerit(*ep)) {
			case 's':	/* Standard */
				rp->r_todisstd = true;
				rp->r_todisgmt = false;
				*ep = '\0';
				break;
			case 'w':	/* Wall */
				rp->r_todisstd = false;
				rp->r_todisgmt = false;
				*ep = '\0';
				break;
			case 'g':	/* Greenwich */
			case 'u':	/* Universal */
			case 'z':	/* Zulu */
				rp->r_todisstd = true;
				rp->r_todisgmt = true;
				*ep = '\0';
				break;
		}
	}
	rp->r_tod = gethms(dp, _("invalid time of day"), false);
	free(dp);
	/*
	** Year work.
	*/
	cp = loyearp;
	lp = byword(cp, begin_years);
	rp->r_lowasnum = lp == NULL;
	if (!rp->r_lowasnum) switch (lp->l_value) {
		case YR_MINIMUM:
			rp->r_loyear = ZIC_MIN;
			break;
		case YR_MAXIMUM:
			rp->r_loyear = ZIC_MAX;
			break;
		default:	/* "cannot happen" */
			fprintf(stderr,
				_("%s: panic: Invalid l_value %d\n"),
				progname, lp->l_value);
			exit(EXIT_FAILURE);
	} else if (sscanf(cp, "%"SCNdZIC"%c", &rp->r_loyear, &xs) != 1) {
		error(_("invalid starting year"));
		return;
	}
	cp = hiyearp;
	lp = byword(cp, end_years);
	rp->r_hiwasnum = lp == NULL;
	if (!rp->r_hiwasnum) switch (lp->l_value) {
		case YR_MINIMUM:
			rp->r_hiyear = ZIC_MIN;
			break;
		case YR_MAXIMUM:
			rp->r_hiyear = ZIC_MAX;
			break;
		case YR_ONLY:
			rp->r_hiyear = rp->r_loyear;
			break;
		default:	/* "cannot happen" */
			fprintf(stderr,
				_("%s: panic: Invalid l_value %d\n"),
				progname, lp->l_value);
			exit(EXIT_FAILURE);
	} else if (sscanf(cp, "%"SCNdZIC"%c", &rp->r_hiyear, &xs) != 1) {
		error(_("invalid ending year"));
		return;
	}
	if (rp->r_loyear > rp->r_hiyear) {
		error(_("starting year greater than ending year"));
		return;
	}
	if (*typep == '\0')
		rp->r_yrtype = NULL;
	else {
		if (rp->r_loyear == rp->r_hiyear) {
			error(_("typed single year"));
			return;
		}
		rp->r_yrtype = ecpyalloc(typep);
	}
	/*
	** Day work.
	** Accept things such as:
	**	1
	**	last-Sunday
	**	Sun<=20
	**	Sun>=7
	*/
	dp = ecpyalloc(dayp);
	if ((lp = byword(dp, lasts)) != NULL) {
		rp->r_dycode = DC_DOWLEQ;
		rp->r_wday = lp->l_value;
		rp->r_dayofmonth = len_months[1][rp->r_month];
	} else {
		if ((ep = strchr(dp, '<')) != 0)
			rp->r_dycode = DC_DOWLEQ;
		else if ((ep = strchr(dp, '>')) != 0)
			rp->r_dycode = DC_DOWGEQ;
		else {
			ep = dp;
			rp->r_dycode = DC_DOM;
		}
		if (rp->r_dycode != DC_DOM) {
			*ep++ = 0;
			if (*ep++ != '=') {
				error(_("invalid day of month"));
				free(dp);
				return;
			}
			if ((lp = byword(dp, wday_names)) == NULL) {
				error(_("invalid weekday name"));
				free(dp);
				return;
			}
			rp->r_wday = lp->l_value;
		}
		if (sscanf(ep, "%d%c", &rp->r_dayofmonth, &xs) != 1 ||
			rp->r_dayofmonth <= 0 ||
			(rp->r_dayofmonth > len_months[1][rp->r_month])) {
				error(_("invalid day of month"));
				free(dp);
				return;
		}
	}
	free(dp);
}

static void
convert(const int_fast32_t val, char *const buf)
{
	register int	i;
	register int	shift;
	unsigned char *const b = (unsigned char *) buf;

	for (i = 0, shift = 24; i < 4; ++i, shift -= 8)
		b[i] = val >> shift;
}

static void
convert64(const zic_t val, char *const buf)
{
	register int	i;
	register int	shift;
	unsigned char *const b = (unsigned char *) buf;

	for (i = 0, shift = 56; i < 8; ++i, shift -= 8)
		b[i] = val >> shift;
}

static void
puttzcode(const int_fast32_t val, FILE *const fp)
{
	char	buf[4];

	convert(val, buf);
	fwrite(buf, sizeof buf, 1, fp);
}

static void
puttzcode64(const zic_t val, FILE *const fp)
{
	char	buf[8];

	convert64(val, buf);
	fwrite(buf, sizeof buf, 1, fp);
}

static int
atcomp(const void *avp, const void *bvp)
{
	const zic_t	a = ((const struct attype *) avp)->at;
	const zic_t	b = ((const struct attype *) bvp)->at;

	return (a < b) ? -1 : (a > b);
}

static bool
is32(const zic_t x)
{
	return INT32_MIN <= x && x <= INT32_MAX;
}

static void
writezone(const char *const name, const char *const string, char version)
{
	register FILE *			fp;
	register ptrdiff_t		i, j;
	register int			leapcnt32, leapi32;
	register ptrdiff_t		timecnt32, timei32;
	register int			pass;
	static const struct tzhead	tzh0;
	static struct tzhead		tzh;
	bool dir_checked = false;
	zic_t one = 1;
	zic_t y2038_boundary = one << 31;
	ptrdiff_t nats = timecnt + WORK_AROUND_QTBUG_53071;
	zic_t *ats = emalloc(size_product(nats, sizeof *ats + 1));
	void *typesptr = ats + nats;
	unsigned char *types = typesptr;

	/*
	** Sort.
	*/
	if (timecnt > 1)
		qsort(attypes, timecnt, sizeof *attypes, atcomp);
	/*
	** Optimize.
	*/
	{
		ptrdiff_t fromi, toi;

		toi = 0;
		fromi = 0;
		while (fromi < timecnt && attypes[fromi].at < early_time)
			++fromi;
		for ( ; fromi < timecnt; ++fromi) {
			if (toi > 1 && ((attypes[fromi].at +
				gmtoffs[attypes[toi - 1].type]) <=
				(attypes[toi - 1].at +
				gmtoffs[attypes[toi - 2].type]))) {
					attypes[toi - 1].type =
						attypes[fromi].type;
					continue;
			}
			if (toi == 0
			    || attypes[fromi].dontmerge
			    || attypes[toi - 1].type != attypes[fromi].type)
					attypes[toi++] = attypes[fromi];
		}
		timecnt = toi;
	}

	if (noise && timecnt > 1200) {
	  if (timecnt > TZ_MAX_TIMES)
		warning(_("reference clients mishandle"
			  " more than %d transition times"),
			TZ_MAX_TIMES);
	  else
		warning(_("pre-2014 clients may mishandle"
			  " more than 1200 transition times"));
	}
	/*
	** Transfer.
	*/
	for (i = 0; i < timecnt; ++i) {
		ats[i] = attypes[i].at;
		types[i] = attypes[i].type;
	}

	/* Work around QTBUG-53071 for time stamps less than y2038_boundary - 1,
	   by inserting a no-op transition at time y2038_boundary - 1.
	   This works only for timestamps before the boundary, which
	   should be good enough in practice as QTBUG-53071 should be
	   long-dead by 2038.  */
	if (WORK_AROUND_QTBUG_53071 && timecnt != 0
	    && ats[timecnt - 1] < y2038_boundary - 1 && strchr(string, '<')) {
	  ats[timecnt] = y2038_boundary - 1;
	  types[timecnt] = types[timecnt - 1];
	  timecnt++;
	}

	/*
	** Correct for leap seconds.
	*/
	for (i = 0; i < timecnt; ++i) {
		j = leapcnt;
		while (--j >= 0)
			if (ats[i] > trans[j] - corr[j]) {
				ats[i] = tadd(ats[i], corr[j]);
				break;
			}
	}
	/*
	** Figure out 32-bit-limited starts and counts.
	*/
	timecnt32 = timecnt;
	timei32 = 0;
	leapcnt32 = leapcnt;
	leapi32 = 0;
	while (timecnt32 > 0 && !is32(ats[timecnt32 - 1]))
		--timecnt32;
	while (timecnt32 > 0 && !is32(ats[timei32])) {
		--timecnt32;
		++timei32;
	}
	/*
	** Output an INT32_MIN "transition" if appropriate; see below.
	*/
	if (timei32 > 0 && ats[timei32] > INT32_MIN) {
		--timei32;
		++timecnt32;
	}
	while (leapcnt32 > 0 && !is32(trans[leapcnt32 - 1]))
		--leapcnt32;
	while (leapcnt32 > 0 && !is32(trans[leapi32])) {
		--leapcnt32;
		++leapi32;
	}
	/*
	** Remove old file, if any, to snap links.
	*/
	if (remove(name) == 0)
		dir_checked = true;
	else if (errno != ENOENT) {
		const char *e = strerror(errno);

		fprintf(stderr, _("%s: Can't remove %s/%s: %s\n"),
			progname, directory, name, e);
		exit(EXIT_FAILURE);
	}
	fp = fopen(name, "wb");
	if (!fp) {
	  int fopen_errno = errno;
	  if (fopen_errno == ENOENT && !dir_checked) {
	    mkdirs(name, true);
	    fp = fopen(name, "wb");
	    fopen_errno = errno;
	  }
	  if (!fp) {
	    fprintf(stderr, _("%s: Can't create %s/%s: %s\n"),
		    progname, directory, name, strerror(fopen_errno));
	    exit(EXIT_FAILURE);
	  }
	}
	for (pass = 1; pass <= 2; ++pass) {
		register ptrdiff_t thistimei, thistimecnt, thistimelim;
		register int	thisleapi, thisleapcnt, thisleaplim;
		int		writetype[TZ_MAX_TYPES];
		int		typemap[TZ_MAX_TYPES];
		register int	thistypecnt;
		char		thischars[TZ_MAX_CHARS];
		int		thischarcnt;
		bool		toomanytimes;
		int		indmap[TZ_MAX_CHARS];

		if (pass == 1) {
			thistimei = timei32;
			thistimecnt = timecnt32;
			toomanytimes = thistimecnt >> 31 >> 1 != 0;
			thisleapi = leapi32;
			thisleapcnt = leapcnt32;
		} else {
			thistimei = 0;
			thistimecnt = timecnt;
			toomanytimes = thistimecnt >> 31 >> 31 >> 2 != 0;
			thisleapi = 0;
			thisleapcnt = leapcnt;
		}
		if (toomanytimes)
		  error(_("too many transition times"));
		thistimelim = thistimei + thistimecnt;
		thisleaplim = thisleapi + thisleapcnt;
		for (i = 0; i < typecnt; ++i)
			writetype[i] = thistimecnt == timecnt;
		if (thistimecnt == 0) {
			/*
			** No transition times fall in the current
			** (32- or 64-bit) window.
			*/
			if (typecnt != 0)
				writetype[typecnt - 1] = true;
		} else {
			for (i = thistimei - 1; i < thistimelim; ++i)
				if (i >= 0)
					writetype[types[i]] = true;
			/*
			** For America/Godthab and Antarctica/Palmer
			*/
			if (thistimei == 0)
				writetype[0] = true;
		}
#ifndef LEAVE_SOME_PRE_2011_SYSTEMS_IN_THE_LURCH
		/*
		** For some pre-2011 systems: if the last-to-be-written
		** standard (or daylight) type has an offset different from the
		** most recently used offset,
		** append an (unused) copy of the most recently used type
		** (to help get global "altzone" and "timezone" variables
		** set correctly).
		*/
		{
			register int	mrudst, mrustd, hidst, histd, type;

			hidst = histd = mrudst = mrustd = -1;
			for (i = thistimei; i < thistimelim; ++i)
				if (isdsts[types[i]])
					mrudst = types[i];
				else	mrustd = types[i];
			for (i = 0; i < typecnt; ++i)
				if (writetype[i]) {
					if (isdsts[i])
						hidst = i;
					else	histd = i;
				}
			if (hidst >= 0 && mrudst >= 0 && hidst != mrudst &&
				gmtoffs[hidst] != gmtoffs[mrudst]) {
					isdsts[mrudst] = -1;
					type = addtype(gmtoffs[mrudst],
						&chars[abbrinds[mrudst]],
						true,
						ttisstds[mrudst],
						ttisgmts[mrudst]);
					isdsts[mrudst] = 1;
					writetype[type] = true;
			}
			if (histd >= 0 && mrustd >= 0 && histd != mrustd &&
				gmtoffs[histd] != gmtoffs[mrustd]) {
					isdsts[mrustd] = -1;
					type = addtype(gmtoffs[mrustd],
						&chars[abbrinds[mrustd]],
						false,
						ttisstds[mrustd],
						ttisgmts[mrustd]);
					isdsts[mrustd] = 0;
					writetype[type] = true;
			}
		}
#endif /* !defined LEAVE_SOME_PRE_2011_SYSTEMS_IN_THE_LURCH */
		thistypecnt = 0;
		for (i = 0; i < typecnt; ++i)
			typemap[i] = writetype[i] ?  thistypecnt++ : -1;
		for (i = 0; i < sizeof indmap / sizeof indmap[0]; ++i)
			indmap[i] = -1;
		thischarcnt = 0;
		for (i = 0; i < typecnt; ++i) {
			register char *	thisabbr;

			if (!writetype[i])
				continue;
			if (indmap[abbrinds[i]] >= 0)
				continue;
			thisabbr = &chars[abbrinds[i]];
			for (j = 0; j < thischarcnt; ++j)
				if (strcmp(&thischars[j], thisabbr) == 0)
					break;
			if (j == thischarcnt) {
				strcpy(&thischars[thischarcnt], thisabbr);
				thischarcnt += strlen(thisabbr) + 1;
			}
			indmap[abbrinds[i]] = j;
		}
#define DO(field)	fwrite(tzh.field, sizeof tzh.field, 1, fp)
		tzh = tzh0;
		memcpy(tzh.tzh_magic, TZ_MAGIC, sizeof tzh.tzh_magic);
		tzh.tzh_version[0] = version;
		convert(thistypecnt, tzh.tzh_ttisgmtcnt);
		convert(thistypecnt, tzh.tzh_ttisstdcnt);
		convert(thisleapcnt, tzh.tzh_leapcnt);
		convert(thistimecnt, tzh.tzh_timecnt);
		convert(thistypecnt, tzh.tzh_typecnt);
		convert(thischarcnt, tzh.tzh_charcnt);
		DO(tzh_magic);
		DO(tzh_version);
		DO(tzh_reserved);
		DO(tzh_ttisgmtcnt);
		DO(tzh_ttisstdcnt);
		DO(tzh_leapcnt);
		DO(tzh_timecnt);
		DO(tzh_typecnt);
		DO(tzh_charcnt);
#undef DO
		for (i = thistimei; i < thistimelim; ++i)
			if (pass == 1)
				/*
				** Output an INT32_MIN "transition"
				** if appropriate; see above.
				*/
				puttzcode(((ats[i] < INT32_MIN) ?
					INT32_MIN : ats[i]), fp);
			else	puttzcode64(ats[i], fp);
		for (i = thistimei; i < thistimelim; ++i) {
			unsigned char	uc;

			uc = typemap[types[i]];
			fwrite(&uc, sizeof uc, 1, fp);
		}
		for (i = 0; i < typecnt; ++i)
			if (writetype[i]) {
				puttzcode(gmtoffs[i], fp);
				putc(isdsts[i], fp);
				putc((unsigned char) indmap[abbrinds[i]], fp);
			}
		if (thischarcnt != 0)
			fwrite(thischars, sizeof thischars[0],
				      thischarcnt, fp);
		for (i = thisleapi; i < thisleaplim; ++i) {
			register zic_t	todo;

			if (roll[i]) {
				if (timecnt == 0 || trans[i] < ats[0]) {
					j = 0;
					while (isdsts[j])
						if (++j >= typecnt) {
							j = 0;
							break;
						}
				} else {
					j = 1;
					while (j < timecnt &&
						trans[i] >= ats[j])
							++j;
					j = types[j - 1];
				}
				todo = tadd(trans[i], -gmtoffs[j]);
			} else	todo = trans[i];
			if (pass == 1)
				puttzcode(todo, fp);
			else	puttzcode64(todo, fp);
			puttzcode(corr[i], fp);
		}
		for (i = 0; i < typecnt; ++i)
			if (writetype[i])
				putc(ttisstds[i], fp);
		for (i = 0; i < typecnt; ++i)
			if (writetype[i])
				putc(ttisgmts[i], fp);
	}
	fprintf(fp, "\n%s\n", string);
	close_file(fp, directory, name);
	free(ats);
}

static char const *
abbroffset(char *buf, zic_t offset)
{
  char sign = '+';
  int seconds, minutes;

  if (offset < 0) {
    offset = -offset;
    sign = '-';
  }

  seconds = offset % SECSPERMIN;
  offset /= SECSPERMIN;
  minutes = offset % MINSPERHOUR;
  offset /= MINSPERHOUR;
  if (100 <= offset) {
    error(_("%%z UTC offset magnitude exceeds 99:59:59"));
    return "%z";
  } else {
    char *p = buf;
    *p++ = sign;
    *p++ = '0' + offset / 10;
    *p++ = '0' + offset % 10;
    if (minutes | seconds) {
      *p++ = '0' + minutes / 10;
      *p++ = '0' + minutes % 10;
      if (seconds) {
	*p++ = '0' + seconds / 10;
	*p++ = '0' + seconds % 10;
      }
    }
    *p = '\0';
    return buf;
  }
}

static size_t
doabbr(char *abbr, struct zone const *zp, char const *letters,
       zic_t stdoff, bool doquotes)
{
	register char *	cp;
	register char *	slashp;
	register size_t	len;
	char const *format = zp->z_format;

	slashp = strchr(format, '/');
	if (slashp == NULL) {
	  char letterbuf[PERCENT_Z_LEN_BOUND + 1];
	  if (zp->z_format_specifier == 'z')
	    letters = abbroffset(letterbuf, zp->z_gmtoff + stdoff);
	  else if (!letters)
	    letters = "%s";
	  sprintf(abbr, format, letters);
	} else if (stdoff != 0) {
		strcpy(abbr, slashp + 1);
	} else {
		memcpy(abbr, format, slashp - format);
		abbr[slashp - format] = '\0';
	}
	len = strlen(abbr);
	if (!doquotes)
		return len;
	for (cp = abbr; is_alpha(*cp); cp++)
		continue;
	if (len > 0 && *cp == '\0')
		return len;
	abbr[len + 2] = '\0';
	abbr[len + 1] = '>';
	memmove(abbr + 1, abbr, len);
	abbr[0] = '<';
	return len + 2;
}

static void
updateminmax(const zic_t x)
{
	if (min_year > x)
		min_year = x;
	if (max_year < x)
		max_year = x;
}

static int
stringoffset(char *result, zic_t offset)
{
	register int	hours;
	register int	minutes;
	register int	seconds;
	bool negative = offset < 0;
	int len = negative;

	if (negative) {
		offset = -offset;
		result[0] = '-';
	}
	seconds = offset % SECSPERMIN;
	offset /= SECSPERMIN;
	minutes = offset % MINSPERHOUR;
	offset /= MINSPERHOUR;
	hours = offset;
	if (hours >= HOURSPERDAY * DAYSPERWEEK) {
		result[0] = '\0';
		return 0;
	}
	len += sprintf(result + len, "%d", hours);
	if (minutes != 0 || seconds != 0) {
		len += sprintf(result + len, ":%02d", minutes);
		if (seconds != 0)
			len += sprintf(result + len, ":%02d", seconds);
	}
	return len;
}

static int
stringrule(char *result, const struct rule *const rp, const zic_t dstoff,
	   const zic_t gmtoff)
{
	register zic_t	tod = rp->r_tod;
	register int	compat = 0;

	if (rp->r_dycode == DC_DOM) {
		register int	month, total;

		if (rp->r_dayofmonth == 29 && rp->r_month == TM_FEBRUARY)
			return -1;
		total = 0;
		for (month = 0; month < rp->r_month; ++month)
			total += len_months[0][month];
		/* Omit the "J" in Jan and Feb, as that's shorter.  */
		if (rp->r_month <= 1)
		  result += sprintf(result, "%d", total + rp->r_dayofmonth - 1);
		else
		  result += sprintf(result, "J%d", total + rp->r_dayofmonth);
	} else {
		register int	week;
		register int	wday = rp->r_wday;
		register int	wdayoff;

		if (rp->r_dycode == DC_DOWGEQ) {
			wdayoff = (rp->r_dayofmonth - 1) % DAYSPERWEEK;
			if (wdayoff)
				compat = 2013;
			wday -= wdayoff;
			tod += wdayoff * SECSPERDAY;
			week = 1 + (rp->r_dayofmonth - 1) / DAYSPERWEEK;
		} else if (rp->r_dycode == DC_DOWLEQ) {
			if (rp->r_dayofmonth == len_months[1][rp->r_month])
				week = 5;
			else {
				wdayoff = rp->r_dayofmonth % DAYSPERWEEK;
				if (wdayoff)
					compat = 2013;
				wday -= wdayoff;
				tod += wdayoff * SECSPERDAY;
				week = rp->r_dayofmonth / DAYSPERWEEK;
			}
		} else	return -1;	/* "cannot happen" */
		if (wday < 0)
			wday += DAYSPERWEEK;
		result += sprintf(result, "M%d.%d.%d",
				  rp->r_month + 1, week, wday);
	}
	if (rp->r_todisgmt)
		tod += gmtoff;
	if (rp->r_todisstd && rp->r_stdoff == 0)
		tod += dstoff;
	if (tod != 2 * SECSPERMIN * MINSPERHOUR) {
		*result++ = '/';
		if (! stringoffset(result, tod))
			return -1;
		if (tod < 0) {
			if (compat < 2013)
				compat = 2013;
		} else if (SECSPERDAY <= tod) {
			if (compat < 1994)
				compat = 1994;
		}
	}
	return compat;
}

static int
rule_cmp(struct rule const *a, struct rule const *b)
{
	if (!a)
		return -!!b;
	if (!b)
		return 1;
	if (a->r_hiyear != b->r_hiyear)
		return a->r_hiyear < b->r_hiyear ? -1 : 1;
	if (a->r_month - b->r_month != 0)
		return a->r_month - b->r_month;
	return a->r_dayofmonth - b->r_dayofmonth;
}

enum { YEAR_BY_YEAR_ZONE = 1 };

static int
stringzone(char *result, struct zone const *zpfirst, ptrdiff_t zonecount)
{
	register const struct zone *	zp;
	register struct rule *		rp;
	register struct rule *		stdrp;
	register struct rule *		dstrp;
	register ptrdiff_t		i;
	register const char *		abbrvar;
	register int			compat = 0;
	register int			c;
	size_t				len;
	int				offsetlen;
	struct rule			stdr, dstr;

	result[0] = '\0';
	zp = zpfirst + zonecount - 1;
	stdrp = dstrp = NULL;
	for (i = 0; i < zp->z_nrules; ++i) {
		rp = &zp->z_rules[i];
		if (rp->r_hiwasnum || rp->r_hiyear != ZIC_MAX)
			continue;
		if (rp->r_yrtype != NULL)
			continue;
		if (rp->r_stdoff == 0) {
			if (stdrp == NULL)
				stdrp = rp;
			else	return -1;
		} else {
			if (dstrp == NULL)
				dstrp = rp;
			else	return -1;
		}
	}
	if (stdrp == NULL && dstrp == NULL) {
		/*
		** There are no rules running through "max".
		** Find the latest std rule in stdabbrrp
		** and latest rule of any type in stdrp.
		*/
		register struct rule *stdabbrrp = NULL;
		for (i = 0; i < zp->z_nrules; ++i) {
			rp = &zp->z_rules[i];
			if (rp->r_stdoff == 0 && rule_cmp(stdabbrrp, rp) < 0)
				stdabbrrp = rp;
			if (rule_cmp(stdrp, rp) < 0)
				stdrp = rp;
		}
		/*
		** Horrid special case: if year is 2037,
		** presume this is a zone handled on a year-by-year basis;
		** do not try to apply a rule to the zone.
		*/
		if (stdrp != NULL && stdrp->r_hiyear == 2037)
			return YEAR_BY_YEAR_ZONE;

		if (stdrp != NULL && stdrp->r_stdoff != 0) {
			/* Perpetual DST.  */
			dstr.r_month = TM_JANUARY;
			dstr.r_dycode = DC_DOM;
			dstr.r_dayofmonth = 1;
			dstr.r_tod = 0;
			dstr.r_todisstd = dstr.r_todisgmt = false;
			dstr.r_stdoff = stdrp->r_stdoff;
			dstr.r_abbrvar = stdrp->r_abbrvar;
			stdr.r_month = TM_DECEMBER;
			stdr.r_dycode = DC_DOM;
			stdr.r_dayofmonth = 31;
			stdr.r_tod = SECSPERDAY + stdrp->r_stdoff;
			stdr.r_todisstd = stdr.r_todisgmt = false;
			stdr.r_stdoff = 0;
			stdr.r_abbrvar
			  = (stdabbrrp ? stdabbrrp->r_abbrvar : "");
			dstrp = &dstr;
			stdrp = &stdr;
		}
	}
	if (stdrp == NULL && (zp->z_nrules != 0 || zp->z_stdoff != 0))
		return -1;
	abbrvar = (stdrp == NULL) ? "" : stdrp->r_abbrvar;
	len = doabbr(result, zp, abbrvar, 0, true);
	offsetlen = stringoffset(result + len, -zp->z_gmtoff);
	if (! offsetlen) {
		result[0] = '\0';
		return -1;
	}
	len += offsetlen;
	if (dstrp == NULL)
		return compat;
	len += doabbr(result + len, zp, dstrp->r_abbrvar, dstrp->r_stdoff, true);
	if (dstrp->r_stdoff != SECSPERMIN * MINSPERHOUR) {
	  offsetlen = stringoffset(result + len,
				   -(zp->z_gmtoff + dstrp->r_stdoff));
	  if (! offsetlen) {
	    result[0] = '\0';
	    return -1;
	  }
	  len += offsetlen;
	}
	result[len++] = ',';
	c = stringrule(result + len, dstrp, dstrp->r_stdoff, zp->z_gmtoff);
	if (c < 0) {
		result[0] = '\0';
		return -1;
	}
	if (compat < c)
		compat = c;
	len += strlen(result + len);
	result[len++] = ',';
	c = stringrule(result + len, stdrp, dstrp->r_stdoff, zp->z_gmtoff);
	if (c < 0) {
		result[0] = '\0';
		return -1;
	}
	if (compat < c)
		compat = c;
	return compat;
}

static void
outzone(const struct zone *zpfirst, ptrdiff_t zonecount)
{
	register const struct zone *	zp;
	register struct rule *		rp;
	register ptrdiff_t		i, j;
	register bool			usestart, useuntil;
	register zic_t			starttime, untiltime;
	register zic_t			gmtoff;
	register zic_t			stdoff;
	register zic_t			year;
	register zic_t			startoff;
	register bool			startttisstd;
	register bool			startttisgmt;
	register int			type;
	register char *			startbuf;
	register char *			ab;
	register char *			envvar;
	register int			max_abbr_len;
	register int			max_envvar_len;
	register bool			prodstic; /* all rules are min to max */
	register int			compat;
	register bool			do_extend;
	register char			version;
	ptrdiff_t lastatmax = -1;
	zic_t one = 1;
	zic_t y2038_boundary = one << 31;
	zic_t max_year0;

	max_abbr_len = 2 + max_format_len + max_abbrvar_len;
	max_envvar_len = 2 * max_abbr_len + 5 * 9;
	startbuf = emalloc(max_abbr_len + 1);
	ab = emalloc(max_abbr_len + 1);
	envvar = emalloc(max_envvar_len + 1);
	INITIALIZE(untiltime);
	INITIALIZE(starttime);
	/*
	** Now. . .finally. . .generate some useful data!
	*/
	timecnt = 0;
	typecnt = 0;
	charcnt = 0;
	prodstic = zonecount == 1;
	/*
	** Thanks to Earl Chew
	** for noting the need to unconditionally initialize startttisstd.
	*/
	startttisstd = false;
	startttisgmt = false;
	min_year = max_year = EPOCH_YEAR;
	if (leapseen) {
		updateminmax(leapminyear);
		updateminmax(leapmaxyear + (leapmaxyear < ZIC_MAX));
	}
	for (i = 0; i < zonecount; ++i) {
		zp = &zpfirst[i];
		if (i < zonecount - 1)
			updateminmax(zp->z_untilrule.r_loyear);
		for (j = 0; j < zp->z_nrules; ++j) {
			rp = &zp->z_rules[j];
			if (rp->r_lowasnum)
				updateminmax(rp->r_loyear);
			if (rp->r_hiwasnum)
				updateminmax(rp->r_hiyear);
			if (rp->r_lowasnum || rp->r_hiwasnum)
				prodstic = false;
		}
	}
	/*
	** Generate lots of data if a rule can't cover all future times.
	*/
	compat = stringzone(envvar, zpfirst, zonecount);
	version = compat < 2013 ? ZIC_VERSION_PRE_2013 : ZIC_VERSION;
	do_extend = compat < 0 || compat == YEAR_BY_YEAR_ZONE;
	if (noise) {
		if (!*envvar)
			warning("%s %s",
				_("no POSIX environment variable for zone"),
				zpfirst->z_name);
		else if (compat != 0 && compat != YEAR_BY_YEAR_ZONE) {
			/* Circa-COMPAT clients, and earlier clients, might
			   not work for this zone when given dates before
			   1970 or after 2038.  */
			warning(_("%s: pre-%d clients may mishandle"
				  " distant timestamps"),
				zpfirst->z_name, compat);
		}
	}
	if (do_extend) {
		/*
		** Search through a couple of extra years past the obvious
		** 400, to avoid edge cases.  For example, suppose a non-POSIX
		** rule applies from 2012 onwards and has transitions in March
		** and September, plus some one-off transitions in November
		** 2013.  If zic looked only at the last 400 years, it would
		** set max_year=2413, with the intent that the 400 years 2014
		** through 2413 will be repeated.  The last transition listed
		** in the tzfile would be in 2413-09, less than 400 years
		** after the last one-off transition in 2013-11.  Two years
		** might be overkill, but with the kind of edge cases
		** available we're not sure that one year would suffice.
		*/
		enum { years_of_observations = YEARSPERREPEAT + 2 };

		if (min_year >= ZIC_MIN + years_of_observations)
			min_year -= years_of_observations;
		else	min_year = ZIC_MIN;
		if (max_year <= ZIC_MAX - years_of_observations)
			max_year += years_of_observations;
		else	max_year = ZIC_MAX;
		/*
		** Regardless of any of the above,
		** for a "proDSTic" zone which specifies that its rules
		** always have and always will be in effect,
		** we only need one cycle to define the zone.
		*/
		if (prodstic) {
			min_year = 1900;
			max_year = min_year + years_of_observations;
		}
	}
	/*
	** For the benefit of older systems,
	** generate data from 1900 through 2038.
	*/
	if (min_year > 1900)
		min_year = 1900;
	max_year0 = max_year;
	if (max_year < 2038)
		max_year = 2038;
	for (i = 0; i < zonecount; ++i) {
		/*
		** A guess that may well be corrected later.
		*/
		stdoff = 0;
		zp = &zpfirst[i];
		usestart = i > 0 && (zp - 1)->z_untiltime > early_time;
		useuntil = i < (zonecount - 1);
		if (useuntil && zp->z_untiltime <= early_time)
			continue;
		gmtoff = zp->z_gmtoff;
		eat(zp->z_filename, zp->z_linenum);
		*startbuf = '\0';
		startoff = zp->z_gmtoff;
		if (zp->z_nrules == 0) {
			stdoff = zp->z_stdoff;
			doabbr(startbuf, zp, NULL, stdoff, false);
			type = addtype(oadd(zp->z_gmtoff, stdoff),
				startbuf, stdoff != 0, startttisstd,
				startttisgmt);
			if (usestart) {
				addtt(starttime, type);
				usestart = false;
			} else	addtt(early_time, type);
		} else for (year = min_year; year <= max_year; ++year) {
			if (useuntil && year > zp->z_untilrule.r_hiyear)
				break;
			/*
			** Mark which rules to do in the current year.
			** For those to do, calculate rpytime(rp, year);
			*/
			for (j = 0; j < zp->z_nrules; ++j) {
				rp = &zp->z_rules[j];
				eats(zp->z_filename, zp->z_linenum,
					rp->r_filename, rp->r_linenum);
				rp->r_todo = year >= rp->r_loyear &&
						year <= rp->r_hiyear &&
						yearistype(year, rp->r_yrtype);
				if (rp->r_todo) {
					rp->r_temp = rpytime(rp, year);
					rp->r_todo
					  = (rp->r_temp < y2038_boundary
					     || year <= max_year0);
				}
			}
			for ( ; ; ) {
				register ptrdiff_t k;
				register zic_t	jtime, ktime;
				register zic_t	offset;

				INITIALIZE(ktime);
				if (useuntil) {
					/*
					** Turn untiltime into UT
					** assuming the current gmtoff and
					** stdoff values.
					*/
					untiltime = zp->z_untiltime;
					if (!zp->z_untilrule.r_todisgmt)
						untiltime = tadd(untiltime,
							-gmtoff);
					if (!zp->z_untilrule.r_todisstd)
						untiltime = tadd(untiltime,
							-stdoff);
				}
				/*
				** Find the rule (of those to do, if any)
				** that takes effect earliest in the year.
				*/
				k = -1;
				for (j = 0; j < zp->z_nrules; ++j) {
					rp = &zp->z_rules[j];
					if (!rp->r_todo)
						continue;
					eats(zp->z_filename, zp->z_linenum,
						rp->r_filename, rp->r_linenum);
					offset = rp->r_todisgmt ? 0 : gmtoff;
					if (!rp->r_todisstd)
						offset = oadd(offset, stdoff);
					jtime = rp->r_temp;
					if (jtime == min_time ||
						jtime == max_time)
							continue;
					jtime = tadd(jtime, -offset);
					if (k < 0 || jtime < ktime) {
						k = j;
						ktime = jtime;
					} else if (jtime == ktime) {
					  char const *dup_rules_msg =
					    _("two rules for same instant");
					  eats(zp->z_filename, zp->z_linenum,
					       rp->r_filename, rp->r_linenum);
					  warning("%s", dup_rules_msg);
					  rp = &zp->z_rules[k];
					  eats(zp->z_filename, zp->z_linenum,
					       rp->r_filename, rp->r_linenum);
					  error("%s", dup_rules_msg);
					}
				}
				if (k < 0)
					break;	/* go on to next year */
				rp = &zp->z_rules[k];
				rp->r_todo = false;
				if (useuntil && ktime >= untiltime)
					break;
				stdoff = rp->r_stdoff;
				if (usestart && ktime == starttime)
					usestart = false;
				if (usestart) {
					if (ktime < starttime) {
						startoff = oadd(zp->z_gmtoff,
							stdoff);
						doabbr(startbuf, zp,
							rp->r_abbrvar,
							rp->r_stdoff,
							false);
						continue;
					}
					if (*startbuf == '\0' &&
						startoff == oadd(zp->z_gmtoff,
						stdoff)) {
							doabbr(startbuf,
								zp,
								rp->r_abbrvar,
								rp->r_stdoff,
								false);
					}
				}
				eats(zp->z_filename, zp->z_linenum,
					rp->r_filename, rp->r_linenum);
				doabbr(ab, zp, rp->r_abbrvar,
				       rp->r_stdoff, false);
				offset = oadd(zp->z_gmtoff, rp->r_stdoff);
				type = addtype(offset, ab, rp->r_stdoff != 0,
					rp->r_todisstd, rp->r_todisgmt);
				if (rp->r_hiyear == ZIC_MAX
				    && ! (0 <= lastatmax
					  && ktime < attypes[lastatmax].at))
				  lastatmax = timecnt;
				addtt(ktime, type);
			}
		}
		if (usestart) {
			if (*startbuf == '\0' &&
				zp->z_format != NULL &&
				strchr(zp->z_format, '%') == NULL &&
				strchr(zp->z_format, '/') == NULL)
					strcpy(startbuf, zp->z_format);
			eat(zp->z_filename, zp->z_linenum);
			if (*startbuf == '\0')
error(_("can't determine time zone abbreviation to use just after until time"));
			else	addtt(starttime,
					addtype(startoff, startbuf,
						startoff != zp->z_gmtoff,
						startttisstd,
						startttisgmt));
		}
		/*
		** Now we may get to set starttime for the next zone line.
		*/
		if (useuntil) {
			startttisstd = zp->z_untilrule.r_todisstd;
			startttisgmt = zp->z_untilrule.r_todisgmt;
			starttime = zp->z_untiltime;
			if (!startttisstd)
				starttime = tadd(starttime, -stdoff);
			if (!startttisgmt)
				starttime = tadd(starttime, -gmtoff);
		}
	}
	if (0 <= lastatmax)
	  attypes[lastatmax].dontmerge = true;
	if (do_extend) {
		/*
		** If we're extending the explicitly listed observations
		** for 400 years because we can't fill the POSIX-TZ field,
		** check whether we actually ended up explicitly listing
		** observations through that period.  If there aren't any
		** near the end of the 400-year period, add a redundant
		** one at the end of the final year, to make it clear
		** that we are claiming to have definite knowledge of
		** the lack of transitions up to that point.
		*/
		struct rule xr;
		struct attype *lastat;
		xr.r_month = TM_JANUARY;
		xr.r_dycode = DC_DOM;
		xr.r_dayofmonth = 1;
		xr.r_tod = 0;
		for (lastat = &attypes[0], i = 1; i < timecnt; i++)
			if (attypes[i].at > lastat->at)
				lastat = &attypes[i];
		if (lastat->at < rpytime(&xr, max_year - 1)) {
			addtt(rpytime(&xr, max_year + 1), typecnt-1);
			attypes[timecnt - 1].dontmerge = true;
		}
	}
	writezone(zpfirst->z_name, envvar, version);
	free(startbuf);
	free(ab);
	free(envvar);
}

static void
addtt(zic_t starttime, int type)
{
	if (starttime <= early_time
	    || (timecnt == 1 && attypes[0].at < early_time)) {
		gmtoffs[0] = gmtoffs[type];
		isdsts[0] = isdsts[type];
		ttisstds[0] = ttisstds[type];
		ttisgmts[0] = ttisgmts[type];
		if (abbrinds[type] != 0)
			strcpy(chars, &chars[abbrinds[type]]);
		abbrinds[0] = 0;
		charcnt = strlen(chars) + 1;
		typecnt = 1;
		timecnt = 0;
		type = 0;
	}
	attypes = growalloc(attypes, sizeof *attypes, timecnt, &timecnt_alloc);
	attypes[timecnt].at = starttime;
	attypes[timecnt].dontmerge = false;
	attypes[timecnt].type = type;
	++timecnt;
}

static int
addtype(zic_t gmtoff, char const *abbr, bool isdst, bool ttisstd, bool ttisgmt)
{
	register int	i, j;

	/*
	** See if there's already an entry for this zone type.
	** If so, just return its index.
	*/
	for (i = 0; i < typecnt; ++i) {
		if (gmtoff == gmtoffs[i] && isdst == isdsts[i] &&
			strcmp(abbr, &chars[abbrinds[i]]) == 0 &&
			ttisstd == ttisstds[i] &&
			ttisgmt == ttisgmts[i])
				return i;
	}
	/*
	** There isn't one; add a new one, unless there are already too
	** many.
	*/
	if (typecnt >= TZ_MAX_TYPES) {
		error(_("too many local time types"));
		exit(EXIT_FAILURE);
	}
	if (! (-1L - 2147483647L <= gmtoff && gmtoff <= 2147483647L)) {
		error(_("UT offset out of range"));
		exit(EXIT_FAILURE);
	}
	gmtoffs[i] = gmtoff;
	isdsts[i] = isdst;
	ttisstds[i] = ttisstd;
	ttisgmts[i] = ttisgmt;

	for (j = 0; j < charcnt; ++j)
		if (strcmp(&chars[j], abbr) == 0)
			break;
	if (j == charcnt)
		newabbr(abbr);
	abbrinds[i] = j;
	++typecnt;
	return i;
}

static void
leapadd(zic_t t, bool positive, int rolling, int count)
{
	register int	i, j;

	if (leapcnt + (positive ? count : 1) > TZ_MAX_LEAPS) {
		error(_("too many leap seconds"));
		exit(EXIT_FAILURE);
	}
	for (i = 0; i < leapcnt; ++i)
		if (t <= trans[i]) {
			if (t == trans[i]) {
				error(_("repeated leap second moment"));
				exit(EXIT_FAILURE);
			}
			break;
		}
	do {
		for (j = leapcnt; j > i; --j) {
			trans[j] = trans[j - 1];
			corr[j] = corr[j - 1];
			roll[j] = roll[j - 1];
		}
		trans[i] = t;
		corr[i] = positive ? 1 : -count;
		roll[i] = rolling;
		++leapcnt;
	} while (positive && --count != 0);
}

static void
adjleap(void)
{
	register int	i;
	register zic_t	last = 0;

	/*
	** propagate leap seconds forward
	*/
	for (i = 0; i < leapcnt; ++i) {
		trans[i] = tadd(trans[i], last);
		last = corr[i] += last;
	}
}

static char *
shellquote(char *b, char const *s)
{
  *b++ = '\'';
  while (*s) {
    if (*s == '\'')
      *b++ = '\'', *b++ = '\\', *b++ = '\'';
    *b++ = *s++;
  }
  *b++ = '\'';
  return b;
}

static bool
yearistype(zic_t year, const char *type)
{
	char *buf;
	char *b;
	int result;

	if (type == NULL || *type == '\0')
		return true;
	buf = emalloc(1 + 4 * strlen(yitcommand) + 2
		      + INT_STRLEN_MAXIMUM(zic_t) + 2 + 4 * strlen(type) + 2);
	b = shellquote(buf, yitcommand);
	*b++ = ' ';
	b += sprintf(b, "%"PRIdZIC, year);
	*b++ = ' ';
	b = shellquote(b, type);
	*b = '\0';
	result = system(buf);
	if (WIFEXITED(result)) {
	  int status = WEXITSTATUS(result);
	  if (status <= 1) {
	    free(buf);
	    return status == 0;
	  }
	}
	error(_("Wild result from command execution"));
	fprintf(stderr, _("%s: command was '%s', result was %d\n"),
		progname, buf, result);
	exit(EXIT_FAILURE);
}

/* Is A a space character in the C locale?  */
static bool
is_space(char a)
{
	switch (a) {
	  default:
		return false;
	  case ' ': case '\f': case '\n': case '\r': case '\t': case '\v':
		return true;
	}
}

/* Is A an alphabetic character in the C locale?  */
static bool
is_alpha(char a)
{
	switch (a) {
	  default:
		return false;
	  case 'A': case 'B': case 'C': case 'D': case 'E': case 'F': case 'G':
	  case 'H': case 'I': case 'J': case 'K': case 'L': case 'M': case 'N':
	  case 'O': case 'P': case 'Q': case 'R': case 'S': case 'T': case 'U':
	  case 'V': case 'W': case 'X': case 'Y': case 'Z':
	  case 'a': case 'b': case 'c': case 'd': case 'e': case 'f': case 'g':
	  case 'h': case 'i': case 'j': case 'k': case 'l': case 'm': case 'n':
	  case 'o': case 'p': case 'q': case 'r': case 's': case 't': case 'u':
	  case 'v': case 'w': case 'x': case 'y': case 'z':
		return true;
	}
}

/* If A is an uppercase character in the C locale, return its lowercase
   counterpart.  Otherwise, return A.  */
static char
lowerit(char a)
{
	switch (a) {
	  default: return a;
	  case 'A': return 'a'; case 'B': return 'b'; case 'C': return 'c';
	  case 'D': return 'd'; case 'E': return 'e'; case 'F': return 'f';
	  case 'G': return 'g'; case 'H': return 'h'; case 'I': return 'i';
	  case 'J': return 'j'; case 'K': return 'k'; case 'L': return 'l';
	  case 'M': return 'm'; case 'N': return 'n'; case 'O': return 'o';
	  case 'P': return 'p'; case 'Q': return 'q'; case 'R': return 'r';
	  case 'S': return 's'; case 'T': return 't'; case 'U': return 'u';
	  case 'V': return 'v'; case 'W': return 'w'; case 'X': return 'x';
	  case 'Y': return 'y'; case 'Z': return 'z';
	}
}

/* case-insensitive equality */
static ATTRIBUTE_PURE bool
ciequal(register const char *ap, register const char *bp)
{
	while (lowerit(*ap) == lowerit(*bp++))
		if (*ap++ == '\0')
			return true;
	return false;
}

static ATTRIBUTE_PURE bool
itsabbr(register const char *abbr, register const char *word)
{
	if (lowerit(*abbr) != lowerit(*word))
		return false;
	++word;
	while (*++abbr != '\0')
		do {
			if (*word == '\0')
				return false;
		} while (lowerit(*word++) != lowerit(*abbr));
	return true;
}

static ATTRIBUTE_PURE const struct lookup *
byword(const char *word, const struct lookup *table)
{
	register const struct lookup *	foundlp;
	register const struct lookup *	lp;

	if (word == NULL || table == NULL)
		return NULL;
	/*
	** Look for exact match.
	*/
	for (lp = table; lp->l_word != NULL; ++lp)
		if (ciequal(word, lp->l_word))
			return lp;
	/*
	** Look for inexact match.
	*/
	foundlp = NULL;
	for (lp = table; lp->l_word != NULL; ++lp)
		if (itsabbr(word, lp->l_word)) {
			if (foundlp == NULL)
				foundlp = lp;
			else	return NULL;	/* multiple inexact matches */
		}
	return foundlp;
}

static char **
getfields(register char *cp)
{
	register char *		dp;
	register char **	array;
	register int		nsubs;

	if (cp == NULL)
		return NULL;
	array = emalloc(size_product(strlen(cp) + 1, sizeof *array));
	nsubs = 0;
	for ( ; ; ) {
		while (is_space(*cp))
				++cp;
		if (*cp == '\0' || *cp == '#')
			break;
		array[nsubs++] = dp = cp;
		do {
			if ((*dp = *cp++) != '"')
				++dp;
			else while ((*dp = *cp++) != '"')
				if (*dp != '\0')
					++dp;
				else {
				  error(_("Odd number of quotation marks"));
				  exit(EXIT_FAILURE);
				}
		} while (*cp && *cp != '#' && !is_space(*cp));
		if (is_space(*cp))
			++cp;
		*dp = '\0';
	}
	array[nsubs] = NULL;
	return array;
}

static _Noreturn void
time_overflow(void)
{
  error(_("time overflow"));
  exit(EXIT_FAILURE);
}

static ATTRIBUTE_PURE zic_t
oadd(zic_t t1, zic_t t2)
{
	if (t1 < 0 ? t2 < ZIC_MIN - t1 : ZIC_MAX - t1 < t2)
	  time_overflow();
	return t1 + t2;
}

static ATTRIBUTE_PURE zic_t
tadd(zic_t t1, zic_t t2)
{
  if (t1 < 0) {
    if (t2 < min_time - t1) {
      if (t1 != min_time)
	time_overflow();
      return min_time;
    }
  } else {
    if (max_time - t1 < t2) {
      if (t1 != max_time)
	time_overflow();
      return max_time;
    }
  }
  return t1 + t2;
}

/*
** Given a rule, and a year, compute the date (in seconds since January 1,
** 1970, 00:00 LOCAL time) in that year that the rule refers to.
*/

static zic_t
rpytime(const struct rule *rp, zic_t wantedy)
{
	register int	m, i;
	register zic_t	dayoff;			/* with a nod to Margaret O. */
	register zic_t	t, y;

	if (wantedy == ZIC_MIN)
		return min_time;
	if (wantedy == ZIC_MAX)
		return max_time;
	dayoff = 0;
	m = TM_JANUARY;
	y = EPOCH_YEAR;
	while (wantedy != y) {
		if (wantedy > y) {
			i = len_years[isleap(y)];
			++y;
		} else {
			--y;
			i = -len_years[isleap(y)];
		}
		dayoff = oadd(dayoff, i);
	}
	while (m != rp->r_month) {
		i = len_months[isleap(y)][m];
		dayoff = oadd(dayoff, i);
		++m;
	}
	i = rp->r_dayofmonth;
	if (m == TM_FEBRUARY && i == 29 && !isleap(y)) {
		if (rp->r_dycode == DC_DOWLEQ)
			--i;
		else {
			error(_("use of 2/29 in non leap-year"));
			exit(EXIT_FAILURE);
		}
	}
	--i;
	dayoff = oadd(dayoff, i);
	if (rp->r_dycode == DC_DOWGEQ || rp->r_dycode == DC_DOWLEQ) {
		register zic_t	wday;

#define LDAYSPERWEEK	((zic_t) DAYSPERWEEK)
		wday = EPOCH_WDAY;
		/*
		** Don't trust mod of negative numbers.
		*/
		if (dayoff >= 0)
			wday = (wday + dayoff) % LDAYSPERWEEK;
		else {
			wday -= ((-dayoff) % LDAYSPERWEEK);
			if (wday < 0)
				wday += LDAYSPERWEEK;
		}
		while (wday != rp->r_wday)
			if (rp->r_dycode == DC_DOWGEQ) {
				dayoff = oadd(dayoff, 1);
				if (++wday >= LDAYSPERWEEK)
					wday = 0;
				++i;
			} else {
				dayoff = oadd(dayoff, -1);
				if (--wday < 0)
					wday = LDAYSPERWEEK - 1;
				--i;
			}
		if (i < 0 || i >= len_months[isleap(y)][m]) {
			if (noise)
				warning(_("rule goes past start/end of month; \
will not work with pre-2004 versions of zic"));
		}
	}
	if (dayoff < min_time / SECSPERDAY)
		return min_time;
	if (dayoff > max_time / SECSPERDAY)
		return max_time;
	t = (zic_t) dayoff * SECSPERDAY;
	return tadd(t, rp->r_tod);
}

static void
newabbr(const char *string)
{
	register int	i;

	if (strcmp(string, GRANDPARENTED) != 0) {
		register const char *	cp;
		const char *		mp;

		cp = string;
		mp = NULL;
		while (is_alpha(*cp) || ('0' <= *cp && *cp <= '9')
		       || *cp == '-' || *cp == '+')
				++cp;
		if (noise && cp - string < 3)
		  mp = _("time zone abbreviation has fewer than 3 characters");
		if (cp - string > ZIC_MAX_ABBR_LEN_WO_WARN)
		  mp = _("time zone abbreviation has too many characters");
		if (*cp != '\0')
mp = _("time zone abbreviation differs from POSIX standard");
		if (mp != NULL)
			warning("%s (%s)", mp, string);
	}
	i = strlen(string) + 1;
	if (charcnt + i > TZ_MAX_CHARS) {
		error(_("too many, or too long, time zone abbreviations"));
		exit(EXIT_FAILURE);
	}
	strcpy(&chars[charcnt], string);
	charcnt += i;
}

/* Ensure that the directories of ARGNAME exist, by making any missing
   ones.  If ANCESTORS, do this only for ARGNAME's ancestors; otherwise,
   do it for ARGNAME too.  Exit with failure if there is trouble.
   Do not consider an existing non-directory to be trouble.  */
static void
mkdirs(char const *argname, bool ancestors)
{
	register char *	name;
	register char *	cp;

	cp = name = ecpyalloc(argname);

	/* Do not mkdir a root directory, as it must exist.  */
#ifdef HAVE_DOS_FILE_NAMES
	if (is_alpha(name[0]) && name[1] == ':')
	  cp += 2;
#endif
	while (*cp == '/')
	  cp++;

	while (cp && ((cp = strchr(cp, '/')) || !ancestors)) {
		if (cp)
		  *cp = '\0';
		/*
		** Try to create it.  It's OK if creation fails because
		** the directory already exists, perhaps because some
		** other process just created it.  For simplicity do
		** not check first whether it already exists, as that
		** is checked anyway if the mkdir fails.
		*/
		if (mkdir(name, MKDIR_UMASK) != 0) {
			/* For speed, skip itsdir if errno == EEXIST.  Since
			   mkdirs is called only after open fails with ENOENT
			   on a subfile, EEXIST implies itsdir here.  */
			int err = errno;
			if (err != EEXIST && !itsdir(name)) {
				error(_("%s: Can't create directory %s: %s"),
				      progname, name, strerror(err));
				exit(EXIT_FAILURE);
			}
		}
		if (cp)
		  *cp++ = '/';
	}
	free(name);
}