utils.c

wget (command line browser) source code

  /* Approximately, the time elapsed between the true start of the
     measurement and the time represented by START.  */
  double elapsed_pre_start;
};

/* Allocate a timer.  It is not legal to do anything with a freshly
   allocated timer, except call wtimer_reset() or wtimer_delete().  */

struct wget_timer *
wtimer_allocate (void)
{
  struct wget_timer *wt =
    (struct wget_timer *)xmalloc (sizeof (struct wget_timer));
  return wt;
}

/* Allocate a new timer and reset it.  Return the new timer. */

struct wget_timer *
wtimer_new (void)
{
  struct wget_timer *wt = wtimer_allocate ();
  wtimer_reset (wt);
  return wt;
}

/* Free the resources associated with the timer.  Its further use is
   prohibited.  */

void
wtimer_delete (struct wget_timer *wt)
{
  xfree (wt);
}

/* Store system time to WST.  */

static void
wtimer_sys_set (wget_sys_time *wst)
{
#ifdef TIMER_GETTIMEOFDAY
  gettimeofday (wst, NULL);
#endif

#ifdef TIMER_TIME
  time (wst);
#endif

#ifdef TIMER_WINDOWS
  /* We use GetSystemTime to get the elapsed time.  MSDN warns that
     system clock adjustments can skew the output of GetSystemTime
     when used as a timer and gives preference to GetTickCount and
     high-resolution timers.  But GetTickCount can overflow, and hires
     timers are typically used for profiling, not for regular time
     measurement.  Since we handle clock skew anyway, we just use
     GetSystemTime.  */
  FILETIME ft;
  SYSTEMTIME st;
  GetSystemTime (&st);

  /* As recommended by MSDN, we convert SYSTEMTIME to FILETIME, copy
     FILETIME to ULARGE_INTEGER, and use regular 64-bit integer
     arithmetic on that.  */
  SystemTimeToFileTime (&st, &ft);
  wst->HighPart = ft.dwHighDateTime;
  wst->LowPart  = ft.dwLowDateTime;
#endif
}

/* Reset timer WT.  This establishes the starting point from which
   wtimer_elapsed() will return the number of elapsed
   milliseconds.  It is allowed to reset a previously used timer.  */

void
wtimer_reset (struct wget_timer *wt)
{
  /* Set the start time to the current time. */
  wtimer_sys_set (&wt->start);
  wt->elapsed_last = 0;
  wt->elapsed_pre_start = 0;
}

static double
wtimer_sys_diff (wget_sys_time *wst1, wget_sys_time *wst2)
{
#ifdef TIMER_GETTIMEOFDAY
  return ((double)(wst1->tv_sec - wst2->tv_sec) * 1000
	  + (double)(wst1->tv_usec - wst2->tv_usec) / 1000);
#endif

#ifdef TIMER_TIME
  return 1000 * (*wst1 - *wst2);
#endif

#ifdef WINDOWS
  /* VC++ 6 doesn't support direct cast of uint64 to double.  To work
     around this, we subtract, then convert to signed, then finally to
     double.  */
  return (double)(signed __int64)(wst1->QuadPart - wst2->QuadPart) / 10000;
#endif
}

/* Return the number of milliseconds elapsed since the timer was last
   reset.  It is allowed to call this function more than once to get
   increasingly higher elapsed values.  These timers handle clock
   skew.  */

double
wtimer_elapsed (struct wget_timer *wt)
{
  wget_sys_time now;
  double elapsed;

  wtimer_sys_set (&now);
  elapsed = wt->elapsed_pre_start + wtimer_sys_diff (&now, &wt->start);

  /* Ideally we'd just return the difference between NOW and
     wt->start.  However, the system timer can be set back, and we
     could return a value smaller than when we were last called, even
     a negative value.  Both of these would confuse the callers, which
     expect us to return monotonically nondecreasing values.

     Therefore: if ELAPSED is smaller than its previous known value,
     we reset wt->start to the current time and effectively start
     measuring from this point.  But since we don't want the elapsed
     value to start from zero, we set elapsed_pre_start to the last
     elapsed time and increment all future calculations by that
     amount.  */

  if (elapsed < wt->elapsed_last)
    {
      wt->start = now;
      wt->elapsed_pre_start = wt->elapsed_last;
      elapsed = wt->elapsed_last;
    }

  wt->elapsed_last = elapsed;
  return elapsed;
}

/* Return the assessed granularity of the timer implementation, in
   milliseconds.  This is used by code that tries to substitute a
   better value for timers that have returned zero.  */

double
wtimer_granularity (void)
{
#ifdef TIMER_GETTIMEOFDAY
  /* Granularity of gettimeofday varies wildly between architectures.
     However, it appears that on modern machines it tends to be better
     than 1ms.  Assume 100 usecs.  (Perhaps the configure process
     could actually measure this?)  */
  return 0.1;
#endif

#ifdef TIMER_TIME
  return 1000;
#endif

#ifdef TIMER_WINDOWS
  /* According to MSDN, GetSystemTime returns a broken-down time
     structure the smallest member of which are milliseconds.  */
  return 1;
#endif
}

/* This should probably be at a better place, but it doesn't really
   fit into html-parse.c.  */

/* The function returns the pointer to the malloc-ed quoted version of
   string s.  It will recognize and quote numeric and special graphic
   entities, as per RFC1866:

   `&' -> `&amp;'
   `<' -> `&lt;'
   `>' -> `&gt;'
   `"' -> `&quot;'
   SP  -> `&#32;'

   No other entities are recognized or replaced.  */
char *
html_quote_string (const char *s)
{
  const char *b = s;
  char *p, *res;
  int i;

  /* Pass through the string, and count the new size.  */
  for (i = 0; *s; s++, i++)
    {
      if (*s == '&')
	i += 4;			/* `amp;' */
      else if (*s == '<' || *s == '>')
	i += 3;			/* `lt;' and `gt;' */
      else if (*s == '\"')
	i += 5;			/* `quot;' */
      else if (*s == ' ')
	i += 4;			/* #32; */
    }
  res = (char *)xmalloc (i + 1);
  s = b;
  for (p = res; *s; s++)
    {
      switch (*s)
	{
	case '&':
	  *p++ = '&';
	  *p++ = 'a';
	  *p++ = 'm';
	  *p++ = 'p';
	  *p++ = ';';
	  break;
	case '<': case '>':
	  *p++ = '&';
	  *p++ = (*s == '<' ? 'l' : 'g');
	  *p++ = 't';
	  *p++ = ';';
	  break;
	case '\"':
	  *p++ = '&';
	  *p++ = 'q';
	  *p++ = 'u';
	  *p++ = 'o';
	  *p++ = 't';
	  *p++ = ';';
	  break;
	case ' ':
	  *p++ = '&';
	  *p++ = '#';
	  *p++ = '3';
	  *p++ = '2';
	  *p++ = ';';
	  break;
	default:
	  *p++ = *s;
	}
    }
  *p = '\0';
  return res;
}

/* Determine the width of the terminal we're running on.  If that's
   not possible, return 0.  */

int
determine_screen_width (void)
{
  /* If there's a way to get the terminal size using POSIX
     tcgetattr(), somebody please tell me.  */
#ifndef TIOCGWINSZ
  return 0;
#else  /* TIOCGWINSZ */
  int fd;
  struct winsize wsz;

  if (opt.lfilename != NULL)
    return 0;

  fd = fileno (stderr);
  if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
    return 0;			/* most likely ENOTTY */

  return wsz.ws_col;
#endif /* TIOCGWINSZ */
}

/* Return a random number between 0 and MAX-1, inclusive.

   If MAX is greater than the value of RAND_MAX+1 on the system, the
   returned value will be in the range [0, RAND_MAX].  This may be
   fixed in a future release.

   The random number generator is seeded automatically the first time
   it is called.

   This uses rand() for portability.  It has been suggested that
   random() offers better randomness, but this is not required for
   Wget, so I chose to go for simplicity and use rand
   unconditionally.

   DO NOT use this for cryptographic purposes.  It is only meant to be
   used in situations where quality of the random numbers returned
   doesn't really matter.  */

int
random_number (int max)
{
  static int seeded;
  double bounded;
  int rnd;

  if (!seeded)
    {
      srand (time (NULL));
      seeded = 1;
    }
  rnd = rand ();

  /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
     and enforce that assumption by masking other bits.  */
#ifndef RAND_MAX
# define RAND_MAX 32767
  rnd &= RAND_MAX;
#endif

  /* This is equivalent to rand() % max, but uses the high-order bits
     for better randomness on architecture where rand() is implemented
     using a simple congruential generator.  */

  bounded = (double)max * rnd / (RAND_MAX + 1.0);
  return (int)bounded;
}

/* Return a random uniformly distributed floating point number in the
   [0, 1) range.  The precision of returned numbers is 9 digits.

   Modify this to use erand48() where available!  */

double
random_float (void)
{
  /* We can't rely on any specific value of RAND_MAX, but I'm pretty
     sure it's greater than 1000.  */
  int rnd1 = random_number (1000);
  int rnd2 = random_number (1000);
  int rnd3 = random_number (1000);
  return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
}

#if 0
/* A debugging function for checking whether an MD5 library works. */

#include "gen-md5.h"

char *
debug_test_md5 (char *buf)
{
  unsigned char raw[16];
  static char res[33];
  unsigned char *p1;
  char *p2;
  int cnt;
  ALLOCA_MD5_CONTEXT (ctx);

  gen_md5_init (ctx);
  gen_md5_update ((unsigned char *)buf, strlen (buf), ctx);
  gen_md5_finish (ctx, raw);

  p1 = raw;
  p2 = res;
  cnt = 16;
  while (cnt--)
    {
      *p2++ = XNUM_TO_digit (*p1 >> 4);
      *p2++ = XNUM_TO_digit (*p1 & 0xf);
      ++p1;
    }
  *p2 = '\0';

  return res;
}
#endif

/* Implementation of run_with_timeout, a generic timeout-forcing
   routine for systems with Unix-like signal handling.  */

#ifdef USE_SIGNAL_TIMEOUT
# ifdef HAVE_SIGSETJMP
#  define SETJMP(env) sigsetjmp (env, 1)

static sigjmp_buf run_with_timeout_env;

static RETSIGTYPE
abort_run_with_timeout (int sig)
{
  assert (sig == SIGALRM);
  siglongjmp (run_with_timeout_env, -1);
}
# else /* not HAVE_SIGSETJMP */
#  define SETJMP(env) setjmp (env)

static jmp_buf run_with_timeout_env;

static RETSIGTYPE
abort_run_with_timeout (int sig)
{
  assert (sig == SIGALRM);
  /* We don't have siglongjmp to preserve the set of blocked signals;
     if we longjumped out of the handler at this point, SIGALRM would
     remain blocked.  We must unblock it manually. */
  int mask = siggetmask ();
  mask &= ~sigmask (SIGALRM);
  sigsetmask (mask);

  /* Now it's safe to longjump. */
  longjmp (run_with_timeout_env, -1);
}
# endif /* not HAVE_SIGSETJMP */

/* Arrange for SIGALRM to be delivered in TIMEOUT seconds.  This uses
   setitimer where available, alarm otherwise.

   TIMEOUT should be non-zero.  If the timeout value is so small that
   it would be rounded to zero, it is rounded to the least legal value
   instead (1us for setitimer, 1s for alarm).  That ensures that
   SIGALRM will be delivered in all cases.  */

static void
alarm_set (double timeout)
{
#ifdef ITIMER_REAL
  /* Use the modern itimer interface. */
  struct itimerval itv;
  memset (&itv, 0, sizeof (itv));
  itv.it_value.tv_sec = (long) timeout;
  itv.it_value.tv_usec = 1000000L * (timeout - (long)timeout);
  if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
    /* Ensure that we wait for at least the minimum interval.
       Specifying zero would mean "wait forever".  */
    itv.it_value.tv_usec = 1;
  setitimer (ITIMER_REAL, &itv, NULL);
#else  /* not ITIMER_REAL */
  /* Use the old alarm() interface. */
  int secs = (int) timeout;
  if (secs == 0)
    /* Round TIMEOUTs smaller than 1 to 1, not to zero.  This is
       because alarm(0) means "never deliver the alarm", i.e. "wait
       forever", which is not what someone who specifies a 0.5s
       timeout would expect.  */
    secs = 1;
  alarm (secs);
#endif /* not ITIMER_REAL */
}

/* Cancel the alarm set with alarm_set. */

static void
alarm_cancel (void)
{
#ifdef ITIMER_REAL
  struct itimerval disable;
  memset (&disable, 0, sizeof (disable));
  setitimer (ITIMER_REAL, &disable, NULL);
#else  /* not ITIMER_REAL */
  alarm (0);
#endif /* not ITIMER_REAL */
}

/* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
   seconds.  Returns non-zero if the function was interrupted with a
   timeout, zero otherwise.

   This works by setting up SIGALRM to be delivered in TIMEOUT seconds
   using setitimer() or alarm().  The timeout is enforced by
   longjumping out of the SIGALRM handler.  This has several
   advantages compared to the traditional approach of relying on
   signals causing system calls to exit with EINTR:

     * The callback function is *forcibly* interrupted after the
       timeout expires, (almost) regardless of what it was doing and
       whether it was in a syscall.  For example, a calculation that
       takes a long time is interrupted as reliably as an IO
       operation.

     * It works with both SYSV and BSD signals because it doesn't
       depend on the default setting of SA_RESTART.

     * It doesn't special handler setup beyond a simple call to
       signal().  (It does use sigsetjmp/siglongjmp, but they're
       optional.)

   The only downside is that, if FUN allocates internal resources that
   are normally freed prior to exit from the functions, they will be
   lost in case of timeout.  */

int
run_with_timeout (double timeout, void (*fun) (void *), void *arg)
{
  int saved_errno;

  if (timeout == 0)
    {
      fun (arg);
      return 0;
    }

  signal (SIGALRM, abort_run_with_timeout);
  if (SETJMP (run_with_timeout_env) != 0)
    {
      /* Longjumped out of FUN with a timeout. */
      signal (SIGALRM, SIG_DFL);
      return 1;
    }
  alarm_set (timeout);
  fun (arg);

  /* Preserve errno in case alarm() or signal() modifies it. */
  saved_errno = errno;
  alarm_cancel ();
  signal (SIGALRM, SIG_DFL);
  errno = saved_errno;

  return 0;
}

#else  /* not USE_SIGNAL_TIMEOUT */

#ifndef WINDOWS
/* A stub version of run_with_timeout that just calls FUN(ARG).  Don't
   define it under Windows, because Windows has its own version of
   run_with_timeout that uses threads.  */

int
run_with_timeout (double timeout, void (*fun) (void *), void *arg)
{
  fun (arg);
  return 0;
}
#endif /* not WINDOWS */
#endif /* not USE_SIGNAL_TIMEOUT */