argp-parse.c

贝叶斯学习算法分类文本。基于朴素贝叶斯分类器的文本分类的通用算法

}

/* Lengths of various parser fields which we will allocated.  */
struct parser_sizes
{
  size_t short_len;		/* Getopt short options string.  */
  size_t long_len;		/* Getopt long options vector.  */
  size_t num_groups;		/* Group structures we allocate.  */
  size_t num_child_inputs;	/* Child input slots.  */
};

/* For ARGP, increments the NUM_GROUPS field in SZS by the total number of
 argp structures descended from it, and the SHORT_LEN & LONG_LEN fields by
 the maximum lengths of the resulting merged getopt short options string and
 long-options array, respectively.  */
static void
calc_sizes (const struct argp *argp,  struct parser_sizes *szs)
{
  const struct argp_child *child = argp->children;
  const struct argp_option *opt = argp->options;

  if (opt || argp->parser)
    {
      szs->num_groups++;
      if (opt)
	{
	  int num_opts = 0;
	  while (!__option_is_end (opt++))
	    num_opts++;
	  szs->short_len += num_opts * 3; /* opt + up to 2 `:'s */
	  szs->long_len += num_opts;
	}
    }

  if (child)
    while (child->argp)
      {
	calc_sizes ((child++)->argp, szs);
	szs->num_child_inputs++;
      }
}

/* Initializes PARSER to parse ARGP in a manner described by FLAGS.  */
static error_t
parser_init (struct parser *parser, const struct argp *argp,
	     int argc, char **argv, int flags, void *input)
{
  error_t err = 0;
  struct group *group;
  struct parser_sizes szs;

  szs.short_len = (flags & ARGP_NO_ARGS) ? 0 : 1;
  szs.long_len = 0;
  szs.num_groups = 0;
  szs.num_child_inputs = 0;

  if (argp)
    calc_sizes (argp, &szs);

  /* Lengths of the various bits of storage used by PARSER.  */
#define GLEN (szs.num_groups + 1) * sizeof (struct group)
#define CLEN (szs.num_child_inputs * sizeof (void *))
#define LLEN ((szs.long_len + 1) * sizeof (struct option))
#define SLEN (szs.short_len + 1)

  parser->storage = malloc (GLEN + CLEN + LLEN + SLEN);
  if (! parser->storage)
    return ENOMEM;

  parser->groups = parser->storage;
  parser->child_inputs = parser->storage + GLEN;
  parser->long_opts = parser->storage + GLEN + CLEN;
  parser->short_opts = parser->storage + GLEN + CLEN + LLEN;

  memset (parser->child_inputs, 0, szs.num_child_inputs * sizeof (void *));
  parser_convert (parser, argp, flags);

  memset (&parser->state, 0, sizeof (struct argp_state));
  parser->state.argp = parser->argp;
  parser->state.argc = argc;
  parser->state.argv = argv;
  parser->state.flags = flags;
  parser->state.err_stream = stderr;
  parser->state.out_stream = stdout;
  parser->state.next = 0;	/* Tell getopt to initialize.  */
  parser->state.pstate = parser;

  parser->try_getopt = 1;

  /* Call each parser for the first time, giving it a chance to propagate
     values to child parsers.  */
  if (parser->groups < parser->egroup)
    parser->groups->input = input;
  for (group = parser->groups;
       group < parser->egroup && (!err || err == EBADKEY);
       group++)
    {
      if (group->parent)
	/* If a child parser, get the initial input value from the parent. */
	group->input = group->parent->child_inputs[group->parent_index];
      err = group_parse (group, &parser->state, ARGP_KEY_INIT, 0);
    }
  if (err == EBADKEY)
    err = 0;			/* Some parser didn't understand.  */

  if (err)
    return err;

  /* Getopt is (currently) non-reentrant.  */
  LOCK_GETOPT;

  if (parser->state.flags & ARGP_NO_ERRS)
    {
      opterr = 0;
      if (parser->state.flags & ARGP_PARSE_ARGV0)
	/* getopt always skips ARGV[0], so we have to fake it out.  As long
	   as OPTERR is 0, then it shouldn't actually try to access it.  */
	parser->state.argv--, parser->state.argc++;
    }
  else
    opterr = 1;			/* Print error messages.  */

  if (parser->state.argv == argv && argv[0])
    /* There's an argv[0]; use it for messages.  */
    {
      char *short_name = strrchr (argv[0], '/');
      parser->state.name = short_name ? short_name + 1 : argv[0];
    }
  else
    parser->state.name = program_invocation_short_name;

  return 0;
}

/* Free any storage consumed by PARSER (but not PARSER itself).  */
static error_t
parser_finalize (struct parser *parser,
		 error_t err, int arg_ebadkey, int *end_index)
{
  struct group *group;

  UNLOCK_GETOPT;

  if (err == EBADKEY && arg_ebadkey)
    /* Suppress errors generated by unparsed arguments.  */
    err = 0;

  if (! err)
    if (parser->state.next == parser->state.argc)
      /* We successfully parsed all arguments!  Call all the parsers again,
	 just a few more times... */
      {
	for (group = parser->groups;
	     group < parser->egroup && (!err || err==EBADKEY);
	     group++)
	  if (group->args_processed == 0)
	    err = group_parse (group, &parser->state, ARGP_KEY_NO_ARGS, 0);
	for (group = parser->groups;
	     group < parser->egroup && (!err || err==EBADKEY);
	     group++)
	  err = group_parse (group, &parser->state, ARGP_KEY_END, 0);

	if (err == EBADKEY)
	  err = 0;		/* Some parser didn't understand.  */
      }
    else if (end_index)
      /* Return any remaining arguments to the user.  */
      *end_index = parser->state.next;
    else
      /* No way to return the remaining arguments, they must be bogus. */
      {
	if (!(parser->state.flags & ARGP_NO_ERRS) && parser->state.err_stream)
	  fprintf (parser->state.err_stream,
		   _("%s: Too many arguments\n"), parser->state.name);
	err = EBADKEY;
      }

  /* Okay, we're all done, with either an error or success.  We only call the
     parsers once more, to indicate which one.  */

  if (err)
    {
      /* Maybe print an error message.  */
      if (err == EBADKEY)
	/* An appropriate message describing what the error was should have
	   been printed earlier.  */
	__argp_state_help (&parser->state, parser->state.err_stream,
			   ARGP_HELP_STD_ERR);

      /* Since we didn't exit, give each parser an error indication.  */
      for (group = parser->groups; group < parser->egroup; group++)
	group_parse (group, &parser->state, ARGP_KEY_ERROR, 0);
    }
  else
    /* Do final cleanup, including propagating back values from parsers.  */
    {
      /* We pass over the groups in reverse order so that child groups are
	 given a chance to do there processing before passing back a value to
	 the parent.  */
      for (group = parser->egroup - 1
	   ; group >= parser->groups && (!err || err == EBADKEY)
	   ; group--)
	err = group_parse (group, &parser->state, ARGP_KEY_SUCCESS, 0);
      if (err == EBADKEY)
	err = 0;		/* Some parser didn't understand.  */
    }

  if (err == EBADKEY)
    err = EINVAL;

  free (parser->storage);

  return err;
}

/* Call the user parsers to parse the non-option argument VAL, at the current
   position, returning any error.  */
static error_t
parser_parse_arg (struct parser *parser, char *val)
{
  int index = parser->state.next;
  error_t err = EBADKEY;
  struct group *group;

  for (group = parser->groups
       ; group < parser->egroup && err == EBADKEY
       ; group++)
    err = group_parse (group, &parser->state, ARGP_KEY_ARG, val);

  if (!err)
    if (parser->state.next >= index)
      /* Remember that we successfully processed a non-option
	 argument -- but only if the user hasn't gotten tricky and set
	 the clock back.  */
      (--group)->args_processed++;
    else
      /* The user wants to reparse some args, give getopt another try.  */
      parser->try_getopt = 1;

  return err;
}

/* Call the user parsers to parse the option OPT, with argument VAL, at the
   current position, returning any error.  */
static error_t
parser_parse_opt (struct parser *parser, int opt, char *val)
{
  /* The group key encoded in the high bits; 0 for short opts or
     group_number + 1 for long opts.  */
  int group_key = opt >> USER_BITS;

  if (group_key == 0)
    /* A short option.  By comparing OPT's position in SHORT_OPTS to the
       various starting positions in each group's SHORT_END field, we can
       determine which group OPT came from.  */
    {
      struct group *group;
      char *short_index = strchr (parser->short_opts, opt);

      if (short_index)
	for (group = parser->groups; group < parser->egroup; group++)
	  if (group->short_end > short_index)
	    return group_parse (group, &parser->state, opt, optarg);

      return EBADKEY;		/* until otherwise asserted */
    }
  else
    /* A long option.  We use shifts instead of masking for extracting
       the user value in order to preserve the sign.  */
    return
      group_parse (&parser->groups[group_key - 1], &parser->state,
		   (opt << GROUP_BITS) >> GROUP_BITS, optarg);
}

/* Parse the next argument in PARSER (as indicated by PARSER->state.next).
   Any error from the parsers is returned, and *ARGP_EBADKEY indicates
   whether a value of EBADKEY is due to an unrecognized argument (which is
   generally not fatal).  */
static error_t
parser_parse_next (struct parser *parser, int *arg_ebadkey)
{
  int opt;
  error_t err = 0;

  if (parser->state.quoted && parser->state.next < parser->state.quoted)
    /* The next argument pointer has been moved to before the quoted
       region, so pretend we never saw the quoting `--', and give getopt
       another chance.  If the user hasn't removed it, getopt will just
       process it again.  */
    parser->state.quoted = 0;

  if (parser->try_getopt && !parser->state.quoted)
    /* Give getopt a chance to parse this.  */
    {
      optind = parser->state.next; /* Put it back in OPTIND for getopt.  */
      optopt = KEY_END;		/* Distinguish KEY_ERR from a real option.  */
      if (parser->state.flags & ARGP_LONG_ONLY)
	opt = getopt_long_only (parser->state.argc, parser->state.argv,
				parser->short_opts, parser->long_opts, 0);
      else
	opt = getopt_long (parser->state.argc, parser->state.argv,
			   parser->short_opts, parser->long_opts, 0);
      parser->state.next = optind; /* And see what getopt did.  */

      if (opt == KEY_END)
	/* Getopt says there are no more options, so stop using
	   getopt; we'll continue if necessary on our own.  */
	{
	  parser->try_getopt = 0;
	  if (parser->state.next > 1
	      && strcmp (parser->state.argv[parser->state.next - 1], QUOTE)
	           == 0)
	    /* Not only is this the end of the options, but it's a
	       `quoted' region, which may have args that *look* like
	       options, so we definitely shouldn't try to use getopt past
	       here, whatever happens.  */
	    parser->state.quoted = parser->state.next;
	}
      else if (opt == KEY_ERR && optopt != KEY_END)
	/* KEY_ERR can have the same value as a valid user short
	   option, but in the case of a real error, getopt sets OPTOPT
	   to the offending character, which can never be KEY_END.  */
	{
	  *arg_ebadkey = 0;
	  return EBADKEY;
	}
    }
  else
    opt = KEY_END;

  if (opt == KEY_END)
    /* We're past what getopt considers the options.  */
    if (parser->state.next >= parser->state.argc
	|| (parser->state.flags & ARGP_NO_ARGS))
      /* Indicate that we're done.  */
      {
	*arg_ebadkey = 1;
	return EBADKEY;
      }
    else
      /* A non-option arg.  */
      err =
	parser_parse_arg (parser, parser->state.argv[parser->state.next++]);
  else if (opt == KEY_ARG)
    /* A non-option argument; try each parser in turn.  */
    err = parser_parse_arg (parser, optarg);
  else
    err = parser_parse_opt (parser, opt, optarg);

  if (err == EBADKEY)
    {
      *arg_ebadkey = (opt == KEY_END || opt == KEY_ARG);
      parser->state.next--;	/* Put back the unused argument.  */
    }

  return err;
}

/* Parse the options strings in ARGC & ARGV according to the argp in ARGP.
   FLAGS is one of the ARGP_ flags above.  If END_INDEX is non-NULL, the
   index in ARGV of the first unparsed option is returned in it.  If an
   unknown option is present, EINVAL is returned; if some parser routine
   returned a non-zero value, it is returned; otherwise 0 is returned.  */
error_t
__argp_parse (const struct argp *argp, int argc, char **argv, unsigned flags,
	      int *end_index, void *input)
{
  error_t err;
  struct parser parser;

  /* If true, then err == EBADKEY is a result of a non-option argument failing
     to be parsed (which in some cases isn't actually an error).  */
  int arg_ebadkey = 0;

#if !HAVE_PROGRAM_INVOCATION_NAME
  /* Set the PROGRAM_INVOCATION_NAME and PROGRAM_INVOCATION_SHORT_NAME 
     if this system doesn't automatically do that. */
  {
    char *s;
    program_invocation_name = argv[0];
    s = strrchr (program_invocation_name, '/');
    if (s)
      program_invocation_short_name = s + 1;
    else
      program_invocation_short_name = program_invocation_name;
  }
#endif

  if (! (flags & ARGP_NO_HELP))
    /* Add our own options.  */
    {
      struct argp_child *child = alloca (4 * sizeof (struct argp_child));
      struct argp *top_argp = alloca (sizeof (struct argp));

      /* TOP_ARGP has no options, it just serves to group the user & default
	 argps.  */
      memset (top_argp, 0, sizeof (*top_argp));
      top_argp->children = child;

      memset (child, 0, 4 * sizeof (struct argp_child));

      if (argp)
	(child++)->argp = argp;
      (child++)->argp = &argp_default_argp;
      if (argp_program_version || argp_program_version_hook)
	(child++)->argp = &argp_version_argp;
      child->argp = 0;

      argp = top_argp;
    }

  /* Construct a parser for these arguments.  */
  err = parser_init (&parser, argp, argc, argv, flags, input);

  if (! err)
    /* Parse! */
    {
      while (! err)
	err = parser_parse_next (&parser, &arg_ebadkey);
      err = parser_finalize (&parser, err, arg_ebadkey, end_index);
    }

  return err;
}
#ifdef weak_alias
weak_alias (__argp_parse, argp_parse)
#endif

/* Return the input field for ARGP in the parser corresponding to STATE; used
   by the help routines.  */
void *
__argp_input (const struct argp *argp, const struct argp_state *state)
{
  if (state)
    {
      struct group *group;
      struct parser *parser = state->pstate;

      for (group = parser->groups; group < parser->egroup; group++)
	if (group->argp == argp)
	  return group->input;
    }

  return 0;
}
#ifdef weak_alias
weak_alias (__argp_input, _argp_input)
#endif