/* Copyright (C) 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
   Free Software Foundation, Inc.
   Contributed by Andy Vaught
   F2003 I/O support contributed by Jerry DeLisle

This file is part of the GNU Fortran runtime library (libgfortran).

Libgfortran is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3, or (at your option)
any later version.

Libgfortran is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

Under Section 7 of GPL version 3, you are granted additional
permissions described in the GCC Runtime Library Exception, version
3.1, as published by the Free Software Foundation.

You should have received a copy of the GNU General Public License and
a copy of the GCC Runtime Library Exception along with this program;
see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
<http://www.gnu.org/licenses/>.  */


/* format.c-- parse a FORMAT string into a binary format suitable for
 * interpretation during I/O statements */

#include "io.h"
#include "format.h"
#include <ctype.h>
#include <string.h>
#include <stdbool.h>
#include <stdlib.h>


static const fnode colon_node = { FMT_COLON, 0, NULL, NULL, {{ 0, 0, 0 }}, 0,
				  NULL };

/* Error messages. */

static const char posint_required[] = "Positive width required in format",
  period_required[] = "Period required in format",
  nonneg_required[] = "Nonnegative width required in format",
  unexpected_element[] = "Unexpected element '%c' in format\n",
  unexpected_end[] = "Unexpected end of format string",
  bad_string[] = "Unterminated character constant in format",
  bad_hollerith[] = "Hollerith constant extends past the end of the format",
  reversion_error[] = "Exhausted data descriptors in format",
  zero_width[] = "Zero width in format descriptor";

/* The following routines support caching format data from parsed format strings
   into a hash table.  This avoids repeatedly parsing duplicate format strings
   or format strings in I/O statements that are repeated in loops.  */


/* Traverse the table and free all data.  */

void
free_format_hash_table (gfc_unit *u)
{
  size_t i;

  /* free_format_data handles any NULL pointers.  */
  for (i = 0; i < FORMAT_HASH_SIZE; i++)
    {
      if (u->format_hash_table[i].hashed_fmt != NULL)
	{
	  free_format_data (u->format_hash_table[i].hashed_fmt);
	  free (u->format_hash_table[i].key);
	}
      u->format_hash_table[i].key = NULL;
      u->format_hash_table[i].key_len = 0;      
      u->format_hash_table[i].hashed_fmt = NULL;
    }
}

/* Traverse the format_data structure and reset the fnode counters.  */

static void
reset_node (fnode *fn)
{
  fnode *f;

  fn->count = 0;
  fn->current = NULL;
  
  if (fn->format != FMT_LPAREN)
    return;

  for (f = fn->u.child; f; f = f->next)
    {
      if (f->format == FMT_RPAREN)
	break;
      reset_node (f);
    }
}

static void
reset_fnode_counters (st_parameter_dt *dtp)
{
  fnode *f;
  format_data *fmt;

  fmt = dtp->u.p.fmt;

  /* Clear this pointer at the head so things start at the right place.  */
  fmt->array.array[0].current = NULL;

  for (f = fmt->array.array[0].u.child; f; f = f->next)
    reset_node (f);
}


/* A simple hashing function to generate an index into the hash table.  */

static inline
uint32_t format_hash (st_parameter_dt *dtp)
{
  char *key;
  gfc_charlen_type key_len;
  uint32_t hash = 0;
  gfc_charlen_type i;

  /* Hash the format string. Super simple, but what the heck!  */
  key = dtp->format;
  key_len = dtp->format_len;
  for (i = 0; i < key_len; i++)
    hash ^= key[i];
  hash &= (FORMAT_HASH_SIZE - 1);
  return hash;
}


static void
save_parsed_format (st_parameter_dt *dtp)
{
  uint32_t hash;
  gfc_unit *u;

  hash = format_hash (dtp);
  u = dtp->u.p.current_unit;

  /* Index into the hash table.  We are simply replacing whatever is there
     relying on probability.  */
  if (u->format_hash_table[hash].hashed_fmt != NULL)
    free_format_data (u->format_hash_table[hash].hashed_fmt);
  u->format_hash_table[hash].hashed_fmt = NULL;

  if (u->format_hash_table[hash].key != NULL)
    free (u->format_hash_table[hash].key);
  u->format_hash_table[hash].key = dtp->format;

  u->format_hash_table[hash].key_len = dtp->format_len;
  u->format_hash_table[hash].hashed_fmt = dtp->u.p.fmt;
}


static format_data *
find_parsed_format (st_parameter_dt *dtp)
{
  uint32_t hash;
  gfc_unit *u;

  hash = format_hash (dtp);
  u = dtp->u.p.current_unit;

  if (u->format_hash_table[hash].key != NULL)
    {
      /* See if it matches.  */
      if (u->format_hash_table[hash].key_len == dtp->format_len)
	{
	  /* So far so good.  */
	  if (strncmp (u->format_hash_table[hash].key,
	      dtp->format, dtp->format_len) == 0)
	    return u->format_hash_table[hash].hashed_fmt;
	}
    }
  return NULL;
}


/* next_char()-- Return the next character in the format string.
 * Returns -1 when the string is done.  If the literal flag is set,
 * spaces are significant, otherwise they are not. */

static int
next_char (format_data *fmt, int literal)
{
  int c;

  do
    {
      if (fmt->format_string_len == 0)
	return -1;

      fmt->format_string_len--;
      c = toupper (*fmt->format_string++);
      fmt->error_element = c;
    }
  while ((c == ' ' || c == '\t') && !literal);

  return c;
}


/* unget_char()-- Back up one character position. */

#define unget_char(fmt) \
  { fmt->format_string--; fmt->format_string_len++; }


/* get_fnode()-- Allocate a new format node, inserting it into the
 * current singly linked list.  These are initially allocated from the
 * static buffer. */

static fnode *
get_fnode (format_data *fmt, fnode **head, fnode **tail, format_token t)
{
  fnode *f;

  if (fmt->avail == &fmt->last->array[FARRAY_SIZE])
    {
      fmt->last->next = get_mem (sizeof (fnode_array));
      fmt->last = fmt->last->next;
      fmt->last->next = NULL;
      fmt->avail = &fmt->last->array[0];
    }
  f = fmt->avail++;
  memset (f, '\0', sizeof (fnode));

  if (*head == NULL)
    *head = *tail = f;
  else
    {
      (*tail)->next = f;
      *tail = f;
    }

  f->format = t;
  f->repeat = -1;
  f->source = fmt->format_string;
  return f;
}


/* free_format_data()-- Free all allocated format data.  */

void
free_format_data (format_data *fmt)
{
  fnode_array *fa, *fa_next;


  if (fmt == NULL)
    return;

  for (fa = fmt->array.next; fa; fa = fa_next)
    {
      fa_next = fa->next;
      free (fa);
    }

  free (fmt);
  fmt = NULL;
}


/* format_lex()-- Simple lexical analyzer for getting the next token
 * in a FORMAT string.  We support a one-level token pushback in the
 * fmt->saved_token variable. */

static format_token
format_lex (format_data *fmt)
{
  format_token token;
  int negative_flag;
  int c;
  char delim;

  if (fmt->saved_token != FMT_NONE)
    {
      token = fmt->saved_token;
      fmt->saved_token = FMT_NONE;
      return token;
    }

  negative_flag = 0;
  c = next_char (fmt, 0);

  switch (c)
    {
    case '*':
       token = FMT_STAR;
       break;

    case '(':
      token = FMT_LPAREN;
      break;

    case ')':
      token = FMT_RPAREN;
      break;

    case '-':
      negative_flag = 1;
      /* Fall Through */

    case '+':
      c = next_char (fmt, 0);
      if (!isdigit (c))
	{
	  token = FMT_UNKNOWN;
	  break;
	}

      fmt->value = c - '0';

      for (;;)
	{
	  c = next_char (fmt, 0);
	  if (!isdigit (c))
	    break;

	  fmt->value = 10 * fmt->value + c - '0';
	}

      unget_char (fmt);

      if (negative_flag)
	fmt->value = -fmt->value;
      token = FMT_SIGNED_INT;
      break;

    case '0':
    case '1':
    case '2':
    case '3':
    case '4':
    case '5':
    case '6':
    case '7':
    case '8':
    case '9':
      fmt->value = c - '0';

      for (;;)
	{
	  c = next_char (fmt, 0);
	  if (!isdigit (c))
	    break;

	  fmt->value = 10 * fmt->value + c - '0';
	}

      unget_char (fmt);
      token = (fmt->value == 0) ? FMT_ZERO : FMT_POSINT;
      break;

    case '.':
      token = FMT_PERIOD;
      break;

    case ',':
      token = FMT_COMMA;
      break;

    case ':':
      token = FMT_COLON;
      break;

    case '/':
      token = FMT_SLASH;
      break;

    case '$':
      token = FMT_DOLLAR;
      break;

    case 'T':
      switch (next_char (fmt, 0))
	{
	case 'L':
	  token = FMT_TL;
	  break;
	case 'R':
	  token = FMT_TR;
	  break;
	default:
	  token = FMT_T;
	  unget_char (fmt);
	  break;
	}

      break;

    case 'X':
      token = FMT_X;
      break;

    case 'S':
      switch (next_char (fmt, 0))
	{
	case 'S':
	  token = FMT_SS;
	  break;
	case 'P':
	  token = FMT_SP;
	  break;
	default:
	  token = FMT_S;
	  unget_char (fmt);
	  break;
	}

      break;

    case 'B':
      switch (next_char (fmt, 0))
	{
	case 'N':
	  token = FMT_BN;
	  break;
	case 'Z':
	  token = FMT_BZ;
	  break;
	default:
	  token = FMT_B;
	  unget_char (fmt);
	  break;
	}

      break;

    case '\'':
    case '"':
      delim = c;

      fmt->string = fmt->format_string;
      fmt->value = 0;		/* This is the length of the string */

      for (;;)
	{
	  c = next_char (fmt, 1);
	  if (c == -1)
	    {
	      token = FMT_BADSTRING;
	      fmt->error = bad_string;
	      break;
	    }

	  if (c == delim)
	    {
	      c = next_char (fmt, 1);

	      if (c == -1)
		{
		  token = FMT_BADSTRING;
		  fmt->error = bad_string;
		  break;
		}

	      if (c != delim)
		{
		  unget_char (fmt);
		  token = FMT_STRING;
		  break;
		}
	    }

	  fmt->value++;
	}

      break;

    case 'P':
      token = FMT_P;
      break;

    case 'I':
      token = FMT_I;
      break;

    case 'O':
      token = FMT_O;
      break;

    case 'Z':
      token = FMT_Z;
      break;

    case 'F':
      token = FMT_F;
      break;

    case 'E':
      switch (next_char (fmt, 0))
	{
	case 'N':
	  token = FMT_EN;
	  break;
	case 'S':
	  token = FMT_ES;
	  break;
	default:
	  token = FMT_E;
	  unget_char (fmt);
	  break;
	}
      break;

    case 'G':
      token = FMT_G;
      break;

    case 'H':
      token = FMT_H;
      break;

    case 'L':
      token = FMT_L;
      break;

    case 'A':
      token = FMT_A;
      break;

    case 'D':
      switch (next_char (fmt, 0))
	{
	case 'P':
	  token = FMT_DP;
	  break;
	case 'C':
	  token = FMT_DC;
	  break;
	default:
	  token = FMT_D;
	  unget_char (fmt);
	  break;
	}
      break;

    case 'R':
      switch (next_char (fmt, 0))
	{
	case 'C':
	  token = FMT_RC;
	  break;
	case 'D':
	  token = FMT_RD;
	  break;
	case 'N':
	  token = FMT_RN;
	  break;
	case 'P':
	  token = FMT_RP;
	  break;
	case 'U':
	  token = FMT_RU;
	  break;
	case 'Z':
	  token = FMT_RZ;
	  break;
	default:
	  unget_char (fmt);
	  token = FMT_UNKNOWN;
	  break;
	}
      break;

    case -1:
      token = FMT_END;
      break;

    default:
      token = FMT_UNKNOWN;
      break;
    }

  return token;
}


/* parse_format_list()-- Parse a format list.  Assumes that a left
 * paren has already been seen.  Returns a list representing the
 * parenthesis node which contains the rest of the list. */

static fnode *
parse_format_list (st_parameter_dt *dtp, bool *seen_dd)
{
  fnode *head, *tail;
  format_token t, u, t2;
  int repeat;
  format_data *fmt = dtp->u.p.fmt;
  bool seen_data_desc = false;

  head = tail = NULL;

  /* Get the next format item */
 format_item:
  t = format_lex (fmt);
 format_item_1:
  switch (t)
    {
    case FMT_STAR:
      t = format_lex (fmt);
      if (t != FMT_LPAREN)
	{
	  fmt->error = "Left parenthesis required after '*'";
	  goto finished;
	}
      get_fnode (fmt, &head, &tail, FMT_LPAREN);
      tail->repeat = -2;  /* Signifies unlimited format.  */
      tail->u.child = parse_format_list (dtp, &seen_data_desc);
      if (fmt->error != NULL)
	goto finished;
      if (!seen_data_desc)
	{
	  fmt->error = "'*' requires at least one associated data descriptor";
	  goto finished;
	}
      goto between_desc;

    case FMT_POSINT:
      repeat = fmt->value;

      t = format_lex (fmt);
      switch (t)
	{
	case FMT_LPAREN:
	  get_fnode (fmt, &head, &tail, FMT_LPAREN);
	  tail->repeat = repeat;
	  tail->u.child = parse_format_list (dtp, &seen_data_desc);
	  *seen_dd = seen_data_desc;
	  if (fmt->error != NULL)
	    goto finished;

	  goto between_desc;

	case FMT_SLASH:
	  get_fnode (fmt, &head, &tail, FMT_SLASH);
	  tail->repeat = repeat;
	  goto optional_comma;

	case FMT_X:
	  get_fnode (fmt, &head, &tail, FMT_X);
	  tail->repeat = 1;
	  tail->u.k = fmt->value;
	  goto between_desc;

	case FMT_P:
	  goto p_descriptor;

	default:
	  goto data_desc;
	}

    case FMT_LPAREN:
      get_fnode (fmt, &head, &tail, FMT_LPAREN);
      tail->repeat = 1;
      tail->u.child = parse_format_list (dtp, &seen_data_desc);
      *seen_dd = seen_data_desc;
      if (fmt->error != NULL)
	goto finished;

      goto between_desc;

    case FMT_SIGNED_INT:	/* Signed integer can only precede a P format.  */
    case FMT_ZERO:		/* Same for zero.  */
      t = format_lex (fmt);
      if (t != FMT_P)
	{
	  fmt->error = "Expected P edit descriptor in format";
	  goto finished;
	}

    p_descriptor:
      get_fnode (fmt, &head, &tail, FMT_P);
      tail->u.k = fmt->value;
      tail->repeat = 1;

      t = format_lex (fmt);
      if (t == FMT_F || t == FMT_EN || t == FMT_ES || t == FMT_D
	  || t == FMT_G || t == FMT_E)
	{
	  repeat = 1;
	  goto data_desc;
	}

      if (t != FMT_COMMA && t != FMT_RPAREN && t != FMT_SLASH
	  && t != FMT_POSINT)
	{
	  fmt->error = "Comma required after P descriptor";
	  goto finished;
	}

      fmt->saved_token = t;
      goto optional_comma;

    case FMT_P:		/* P and X require a prior number */
      fmt->error = "P descriptor requires leading scale factor";
      goto finished;

    case FMT_X:
/*
   EXTENSION!

   If we would be pedantic in the library, we would have to reject
   an X descriptor without an integer prefix:

      fmt->error = "X descriptor requires leading space count";
      goto finished;

   However, this is an extension supported by many Fortran compilers,
   including Cray, HP, AIX, and IRIX.  Therefore, we allow it in the
   runtime library, and make the front end reject it if the compiler
   is in pedantic mode.  The interpretation of 'X' is '1X'.
*/
      get_fnode (fmt, &head, &tail, FMT_X);
      tail->repeat = 1;
      tail->u.k = 1;
      goto between_desc;

    case FMT_STRING:
      get_fnode (fmt, &head, &tail, FMT_STRING);
      tail->u.string.p = fmt->string;
      tail->u.string.length = fmt->value;
      tail->repeat = 1;
      goto optional_comma;
      
    case FMT_RC:
    case FMT_RD:
    case FMT_RN:
    case FMT_RP:
    case FMT_RU:
    case FMT_RZ:
      notify_std (&dtp->common, GFC_STD_F2003, "Fortran 2003: Round "
		  "descriptor not allowed");
      get_fnode (fmt, &head, &tail, t);
      tail->repeat = 1;
      goto between_desc;

    case FMT_DC:
    case FMT_DP:
      notify_std (&dtp->common, GFC_STD_F2003, "Fortran 2003: DC or DP "
		  "descriptor not allowed");
    /* Fall through.  */
    case FMT_S:
    case FMT_SS:
    case FMT_SP:
    case FMT_BN:
    case FMT_BZ:
      get_fnode (fmt, &head, &tail, t);
      tail->repeat = 1;
      goto between_desc;

    case FMT_COLON:
      get_fnode (fmt, &head, &tail, FMT_COLON);
      tail->repeat = 1;
      goto optional_comma;

    case FMT_SLASH:
      get_fnode (fmt, &head, &tail, FMT_SLASH);
      tail->repeat = 1;
      tail->u.r = 1;
      goto optional_comma;

    case FMT_DOLLAR:
      get_fnode (fmt, &head, &tail, FMT_DOLLAR);
      tail->repeat = 1;
      notify_std (&dtp->common, GFC_STD_GNU, "Extension: $ descriptor");
      goto between_desc;

    case FMT_T:
    case FMT_TL:
    case FMT_TR:
      t2 = format_lex (fmt);
      if (t2 != FMT_POSINT)
	{
	  fmt->error = posint_required;
	  goto finished;
	}
      get_fnode (fmt, &head, &tail, t);
      tail->u.n = fmt->value;
      tail->repeat = 1;
      goto between_desc;

    case FMT_I:
    case FMT_B:
    case FMT_O:
    case FMT_Z:
    case FMT_E:
    case FMT_EN:
    case FMT_ES:
    case FMT_D:
    case FMT_L:
    case FMT_A:
    case FMT_F:
    case FMT_G:
      repeat = 1;
      *seen_dd = true;
      goto data_desc;

    case FMT_H:
      get_fnode (fmt, &head, &tail, FMT_STRING);
      if (fmt->format_string_len < 1)
	{
	  fmt->error = bad_hollerith;
	  goto finished;
	}

      tail->u.string.p = fmt->format_string;
      tail->u.string.length = 1;
      tail->repeat = 1;

      fmt->format_string++;
      fmt->format_string_len--;

      goto between_desc;

    case FMT_END:
      fmt->error = unexpected_end;
      goto finished;

    case FMT_BADSTRING:
      goto finished;

    case FMT_RPAREN:
      goto finished;

    default:
      fmt->error = unexpected_element;
      goto finished;
    }

  /* In this state, t must currently be a data descriptor.  Deal with
     things that can/must follow the descriptor */
 data_desc:
  switch (t)
    {
    case FMT_L:
      t = format_lex (fmt);
      if (t != FMT_POSINT)
	{
	  if (notification_std(GFC_STD_GNU) == NOTIFICATION_ERROR)
	    {
	      fmt->error = posint_required;
	      goto finished;
	    }
	  else
	    {
	      fmt->saved_token = t;
	      fmt->value = 1;	/* Default width */
	      notify_std (&dtp->common, GFC_STD_GNU, posint_required);
	    }
	}

      get_fnode (fmt, &head, &tail, FMT_L);
      tail->u.n = fmt->value;
      tail->repeat = repeat;
      break;

    case FMT_A:
      t = format_lex (fmt);
      if (t == FMT_ZERO)
	{
	  fmt->error = zero_width;
	  goto finished;
	}

      if (t != FMT_POSINT)
	{
	  fmt->saved_token = t;
	  fmt->value = -1;		/* Width not present */
	}

      get_fnode (fmt, &head, &tail, FMT_A);
      tail->repeat = repeat;
      tail->u.n = fmt->value;
      break;

    case FMT_D:
    case FMT_E:
    case FMT_F:
    case FMT_G:
    case FMT_EN:
    case FMT_ES:
      get_fnode (fmt, &head, &tail, t);
      tail->repeat = repeat;

      u = format_lex (fmt);
      if (t == FMT_G && u == FMT_ZERO)
	{
	  if (notification_std (GFC_STD_F2008) == NOTIFICATION_ERROR
	      || dtp->u.p.mode == READING)
	    {
	      fmt->error = zero_width;
	      goto finished;
	    }
	  tail->u.real.w = 0;
	  u = format_lex (fmt);
	  if (u != FMT_PERIOD)
	    {
	      fmt->saved_token = u;
	      break;
	    }

	  u = format_lex (fmt);
	  if (u != FMT_POSINT)
	    {
	      fmt->error = posint_required;
	      goto finished;
	    }
	  tail->u.real.d = fmt->value;
	  break;
	}
      if (t == FMT_F && dtp->u.p.mode == WRITING)
	{
	  if (u != FMT_POSINT && u != FMT_ZERO)
	    {
	      fmt->error = nonneg_required;
	      goto finished;
	    }
	}
      else if (u != FMT_POSINT)
	{
	  fmt->error = posint_required;
	  goto finished;
	}

      tail->u.real.w = fmt->value;
      t2 = t;
      t = format_lex (fmt);
      if (t != FMT_PERIOD)
	{
	  /* We treat a missing decimal descriptor as 0.  Note: This is only
	     allowed if -std=legacy, otherwise an error occurs.  */
	  if (compile_options.warn_std != 0)
	    {
	      fmt->error = period_required;
	      goto finished;
	    }
	  fmt->saved_token = t;
	  tail->u.real.d = 0;
	  tail->u.real.e = -1;
	  break;
	}

      t = format_lex (fmt);
      if (t != FMT_ZERO && t != FMT_POSINT)
	{
	  fmt->error = nonneg_required;
	  goto finished;
	}

      tail->u.real.d = fmt->value;
      tail->u.real.e = -1;

      if (t2 == FMT_D || t2 == FMT_F)
	break;


      /* Look for optional exponent */
      t = format_lex (fmt);
      if (t != FMT_E)
	fmt->saved_token = t;
      else
	{
	  t = format_lex (fmt);
	  if (t != FMT_POSINT)
	    {
	      fmt->error = "Positive exponent width required in format";
	      goto finished;
	    }

	  tail->u.real.e = fmt->value;
	}

      break;

    case FMT_H:
      if (repeat > fmt->format_string_len)
	{
	  fmt->error = bad_hollerith;
	  goto finished;
	}

      get_fnode (fmt, &head, &tail, FMT_STRING);
      tail->u.string.p = fmt->format_string;
      tail->u.string.length = repeat;
      tail->repeat = 1;

      fmt->format_string += fmt->value;
      fmt->format_string_len -= repeat;

      break;

    case FMT_I:
    case FMT_B:
    case FMT_O:
    case FMT_Z:
      get_fnode (fmt, &head, &tail, t);
      tail->repeat = repeat;

      t = format_lex (fmt);

      if (dtp->u.p.mode == READING)
	{
	  if (t != FMT_POSINT)
	    {
	      fmt->error = posint_required;
	      goto finished;
	    }
	}
      else
	{
	  if (t != FMT_ZERO && t != FMT_POSINT)
	    {
	      fmt->error = nonneg_required;
	      goto finished;
	    }
	}

      tail->u.integer.w = fmt->value;
      tail->u.integer.m = -1;

      t = format_lex (fmt);
      if (t != FMT_PERIOD)
	{
	  fmt->saved_token = t;
	}
      else
	{
	  t = format_lex (fmt);
	  if (t != FMT_ZERO && t != FMT_POSINT)
	    {
	      fmt->error = nonneg_required;
	      goto finished;
	    }

	  tail->u.integer.m = fmt->value;
	}

      if (tail->u.integer.w != 0 && tail->u.integer.m > tail->u.integer.w)
	{
	  fmt->error = "Minimum digits exceeds field width";
	  goto finished;
	}

      break;

    default:
      fmt->error = unexpected_element;
      goto finished;
    }

  /* Between a descriptor and what comes next */
 between_desc:
  t = format_lex (fmt);
  switch (t)
    {
    case FMT_COMMA:
      goto format_item;

    case FMT_RPAREN:
      goto finished;

    case FMT_SLASH:
    case FMT_COLON:
      get_fnode (fmt, &head, &tail, t);
      tail->repeat = 1;
      goto optional_comma;

    case FMT_END:
      fmt->error = unexpected_end;
      goto finished;

    default:
      /* Assume a missing comma, this is a GNU extension */
      goto format_item_1;
    }

  /* Optional comma is a weird between state where we've just finished
     reading a colon, slash or P descriptor. */
 optional_comma:
  t = format_lex (fmt);
  switch (t)
    {
    case FMT_COMMA:
      break;

    case FMT_RPAREN:
      goto finished;

    default:			/* Assume that we have another format item */
      fmt->saved_token = t;
      break;
    }

  goto format_item;

 finished:

  return head;
}


/* format_error()-- Generate an error message for a format statement.
 * If the node that gives the location of the error is NULL, the error
 * is assumed to happen at parse time, and the current location of the
 * parser is shown.
 *
 * We generate a message showing where the problem is.  We take extra
 * care to print only the relevant part of the format if it is longer
 * than a standard 80 column display. */

void
format_error (st_parameter_dt *dtp, const fnode *f, const char *message)
{
  int width, i, j, offset;
  char *p, buffer[300];
  format_data *fmt = dtp->u.p.fmt;

  if (f != NULL)
    fmt->format_string = f->source;

  if (message == unexpected_element)
    sprintf (buffer, message, fmt->error_element);
  else
    sprintf (buffer, "%s\n", message);

  j = fmt->format_string - dtp->format;

  offset = (j > 60) ? j - 40 : 0;

  j -= offset;
  width = dtp->format_len - offset;

  if (width > 80)
    width = 80;

  /* Show the format */

  p = strchr (buffer, '\0');

  memcpy (p, dtp->format + offset, width);

  p += width;
  *p++ = '\n';

  /* Show where the problem is */

  for (i = 1; i < j; i++)
    *p++ = ' ';

  *p++ = '^';
  *p = '\0';

  generate_error (&dtp->common, LIBERROR_FORMAT, buffer);
}


/* revert()-- Do reversion of the format.  Control reverts to the left
 * parenthesis that matches the rightmost right parenthesis.  From our
 * tree structure, we are looking for the rightmost parenthesis node
 * at the second level, the first level always being a single
 * parenthesis node.  If this node doesn't exit, we use the top
 * level. */

static void
revert (st_parameter_dt *dtp)
{
  fnode *f, *r;
  format_data *fmt = dtp->u.p.fmt;

  dtp->u.p.reversion_flag = 1;

  r = NULL;

  for (f = fmt->array.array[0].u.child; f; f = f->next)
    if (f->format == FMT_LPAREN)
      r = f;

  /* If r is NULL because no node was found, the whole tree will be used */

  fmt->array.array[0].current = r;
  fmt->array.array[0].count = 0;
}

/* parse_format()-- Parse a format string.  */

void
parse_format (st_parameter_dt *dtp)
{
  format_data *fmt;
  bool format_cache_ok, seen_data_desc = false;

  /* Don't cache for internal units and set an arbitrary limit on the size of
     format strings we will cache.  (Avoids memory issues.)  */
  format_cache_ok = !is_internal_unit (dtp);

  /* Lookup format string to see if it has already been parsed.  */
  if (format_cache_ok)
    {
      dtp->u.p.fmt = find_parsed_format (dtp);

      if (dtp->u.p.fmt != NULL)
	{
	  dtp->u.p.fmt->reversion_ok = 0;
	  dtp->u.p.fmt->saved_token = FMT_NONE;
	  dtp->u.p.fmt->saved_format = NULL;
	  reset_fnode_counters (dtp);
	  return;
	}
    }

  /* Not found so proceed as follows.  */

  if (format_cache_ok)
    {
      char *fmt_string = get_mem (dtp->format_len);
      memcpy (fmt_string, dtp->format, dtp->format_len);
      dtp->format = fmt_string;
    }

  dtp->u.p.fmt = fmt = get_mem (sizeof (format_data));
  fmt->format_string = dtp->format;
  fmt->format_string_len = dtp->format_len;

  fmt->string = NULL;
  fmt->saved_token = FMT_NONE;
  fmt->error = NULL;
  fmt->value = 0;

  /* Initialize variables used during traversal of the tree.  */

  fmt->reversion_ok = 0;
  fmt->saved_format = NULL;

  /* Allocate the first format node as the root of the tree.  */

  fmt->last = &fmt->array;
  fmt->last->next = NULL;
  fmt->avail = &fmt->array.array[0];

  memset (fmt->avail, 0, sizeof (*fmt->avail));
  fmt->avail->format = FMT_LPAREN;
  fmt->avail->repeat = 1;
  fmt->avail++;

  if (format_lex (fmt) == FMT_LPAREN)
    fmt->array.array[0].u.child = parse_format_list (dtp, &seen_data_desc);
  else
    fmt->error = "Missing initial left parenthesis in format";

  if (fmt->error)
    {
      format_error (dtp, NULL, fmt->error);
      if (format_cache_ok)
	free (dtp->format);
      free_format_hash_table (dtp->u.p.current_unit);
      return;
    }

  if (format_cache_ok)
    save_parsed_format (dtp);
  else
    dtp->u.p.format_not_saved = 1;
}


/* next_format0()-- Get the next format node without worrying about
 * reversion.  Returns NULL when we hit the end of the list.
 * Parenthesis nodes are incremented after the list has been
 * exhausted, other nodes are incremented before they are returned. */

static const fnode *
next_format0 (fnode * f)
{
  const fnode *r;

  if (f == NULL)
    return NULL;

  if (f->format != FMT_LPAREN)
    {
      f->count++;
      if (f->count <= f->repeat)
	return f;

      f->count = 0;
      return NULL;
    }

  /* Deal with a parenthesis node with unlimited format.  */

  if (f->repeat == -2)  /* -2 signifies unlimited.  */
  for (;;)
    {
      if (f->current == NULL)
	f->current = f->u.child;

      for (; f->current != NULL; f->current = f->current->next)
	{
	  r = next_format0 (f->current);
	  if (r != NULL)
	    return r;
	}
    }

  /* Deal with a parenthesis node with specific repeat count.  */
  for (; f->count < f->repeat; f->count++)
    {
      if (f->current == NULL)
	f->current = f->u.child;

      for (; f->current != NULL; f->current = f->current->next)
	{
	  r = next_format0 (f->current);
	  if (r != NULL)
	    return r;
	}
    }

  f->count = 0;
  return NULL;
}


/* next_format()-- Return the next format node.  If the format list
 * ends up being exhausted, we do reversion.  Reversion is only
 * allowed if we've seen a data descriptor since the
 * initialization or the last reversion.  We return NULL if there
 * are no more data descriptors to return (which is an error
 * condition). */

const fnode *
next_format (st_parameter_dt *dtp)
{
  format_token t;
  const fnode *f;
  format_data *fmt = dtp->u.p.fmt;

  if (fmt->saved_format != NULL)
    {				/* Deal with a pushed-back format node */
      f = fmt->saved_format;
      fmt->saved_format = NULL;
      goto done;
    }

  f = next_format0 (&fmt->array.array[0]);
  if (f == NULL)
    {
      if (!fmt->reversion_ok)
	return NULL;

      fmt->reversion_ok = 0;
      revert (dtp);

      f = next_format0 (&fmt->array.array[0]);
      if (f == NULL)
	{
	  format_error (dtp, NULL, reversion_error);
	  return NULL;
	}

      /* Push the first reverted token and return a colon node in case
       * there are no more data items. */

      fmt->saved_format = f;
      return &colon_node;
    }

  /* If this is a data edit descriptor, then reversion has become OK. */
 done:
  t = f->format;

  if (!fmt->reversion_ok &&
      (t == FMT_I || t == FMT_B || t == FMT_O || t == FMT_Z || t == FMT_F ||
       t == FMT_E || t == FMT_EN || t == FMT_ES || t == FMT_G || t == FMT_L ||
       t == FMT_A || t == FMT_D))
    fmt->reversion_ok = 1;
  return f;
}


/* unget_format()-- Push the given format back so that it will be
 * returned on the next call to next_format() without affecting
 * counts.  This is necessary when we've encountered a data
 * descriptor, but don't know what the data item is yet.  The format
 * node is pushed back, and we return control to the main program,
 * which calls the library back with the data item (or not). */

void
unget_format (st_parameter_dt *dtp, const fnode *f)
{
  dtp->u.p.fmt->saved_format = f;
}