@@@ misc wip

[mLib] / test / tvec-types.c

/* -*-c-*-
 *
 * Types for the test-vector framework
 *
 * (c) 2023 Straylight/Edgeware
 */

/*----- Licensing notice --------------------------------------------------*
 *
 * This file is part of the mLib utilities library.
 *
 * mLib is free software: you can redistribute it and/or modify it under
 * the terms of the GNU Library General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or (at
 * your option) any later version.
 *
 * mLib 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 Library General Public
 * License for more details.
 *
 * You should have received a copy of the GNU Library General Public
 * License along with mLib.  If not, write to the Free Software
 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
 * USA.
 */

/*----- Header files ------------------------------------------------------*/

#include <assert.h>
#include <ctype.h>
#include <errno.h>
#include <float.h>
#include <limits.h>
#include <math.h>
#include <stdio.h>
#include <string.h>

#include "buf.h"
#include "codec.h"
#  include "base32.h"
#  include "base64.h"
#  include "hex.h"
#include "dstr.h"
#include "maths.h"
#include "tvec.h"

/*----- Preliminary utilities ---------------------------------------------*/

/* --- @trivial_release@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = a register value
 *              @const struct tvec_regdef@ = the register definition
 *
 * Returns:     ---
 *
 * Use:         Does nothing.  Used for register values which don't retain
 *              resources.
 */

static void trivial_release(union tvec_regval *rv,
                            const struct tvec_regdef *rd)
  { ; }

/*----- Integer utilities -------------------------------------------------*/

/* --- @unsigned_to_buf@, @signed_to_buf@ --- *
 *
 * Arguments:   @buf *b@ = buffer to write on
 *              @unsigned long u@ or @long i@ = integer to write
 *
 * Returns:     Zero on success, @-1@ on failure.
 *
 * Use:         Write @i@ to the buffer, in big-endian (two's-complement, it
 *              signed) format.
 */

static int unsigned_to_buf(buf *b, unsigned long u)
  { kludge64 k; ASSIGN64(k, u); return (buf_putk64l(b, k)); }

static int signed_to_buf(buf *b, long i)
{
  kludge64 k;
  unsigned long u;

  u = i;
  if (i >= 0) ASSIGN64(k, u);
  else { ASSIGN64(k, ~u); CPL64(k, k); }
  return (buf_putk64l(b, k));
}

/* --- @unsigned_from_buf@, @signed_from_buf@ --- *
 *
 * Arguments:   @buf *b@ = buffer to write on
 *              @unsigned long *u_out@ or @long *i_out@ = where to put the
 *                      result
 *
 * Returns:     Zero on success, @-1@ on failure.
 *
 * Use:         Read an integer, in big-endian (two's-complement, if signed)
 *              format, from the buffer.
 */

static int unsigned_from_buf(buf *b, unsigned long *u_out)
{
  kludge64 k, ulmax;

  ASSIGN64(ulmax, ULONG_MAX);
  if (buf_getk64l(b, &k)) return (-1);
  if (CMP64(k, >, ulmax)) { buf_break(b); return (-1); }
  *u_out = GET64(unsigned long, k); return (0);
}

/* --- @hex_width@ --- *
 *
 * Arguments:   @unsigned long u@ = an integer
 *
 * Returns:     A suitable number of digits to use in order to display @u@ in
 *              hex.  Currently, we select a power of two sufficient to show
 *              the value, but at least 2.
 */

static int hex_width(unsigned long u)
{
  int wd;
  unsigned long t;

  for (t = u >> 4, wd = 4; t >>= wd, wd *= 2, t; );
  return (wd/4);
}

/* --- @format_unsigned_hex@, @format_signed_hex@ --- *
 *
 * Arguments:   @const struct gprintf_ops *gops@ = print operations
 *              @void *go@ = print destination
 *              @unsigned long u@ or @long i@ = integer to print
 *
 * Returns:     ---
 *
 * Use:         Print an unsigned or signed integer in hexadecimal.
 */

static void format_unsigned_hex(const struct gprintf_ops *gops, void *go,
                                unsigned long u)
  { gprintf(gops, go, "0x%0*lx", hex_width(u), u); }

static void format_signed_hex(const struct gprintf_ops *gops, void *go,
                              long i)
{
  unsigned long u = i >= 0 ? i : -(unsigned long)i;
  gprintf(gops, go, "%s0x%0*lx", i < 0 ? "-" : "", hex_width(u), u);
}

static int signed_from_buf(buf *b, long *i_out)
{
  kludge64 k, lmax, not_lmin;

  ASSIGN64(lmax, LONG_MAX); ASSIGN64(not_lmin, ~(unsigned long)LONG_MIN);
  if (buf_getk64l(b, &k)) return (-1);
  if (CMP64(k, <=, lmax)) *i_out = (long)GET64(unsigned long, k);
  else {
    CPL64(k, k);
    if (CMP64(k, <=, not_lmin)) *i_out = -(long)GET64(unsigned long, k) - 1;
    else { buf_break(b); return (-1); }
  }
  return (0);
}

/* --- @check_unsigned_range@, @check_signed_range@ --- *
 *
 * Arguments:   @unsigned long u@ or @long i@ = an integer
 *              @const struct tvec_urange *ur@ or
 *                      @const struct tvec_irange *ir@ = range specification,
 *                      or null
 *              @struct tvec_state *tv@ = test vector state
 *              @const char *what@ = description of value
 *
 * Returns:     Zero on success, or @-1@ on error.
 *
 * Use:         Check that the integer is within bounds.  If not, report a
 *              suitable error and return a failure indication.
 */

static int check_signed_range(long i,
                              const struct tvec_irange *ir,
                              struct tvec_state *tv, const char *what)
{
  if (ir && (ir->min > i || i > ir->max)) {
    tvec_error(tv, "%s %ld out of range (must be in [%ld .. %ld])",
               what, i, ir->min, ir->max);
    return (-1);
  }
  return (0);
}

static int check_unsigned_range(unsigned long u,
                                const struct tvec_urange *ur,
                                struct tvec_state *tv, const char *what)
{
  if (ur && (ur->min > u || u > ur->max)) {
    tvec_error(tv, "%s %lu out of range (must be in [%lu .. %lu])",
               what, u, ur->min, ur->max);
    return (-1);
  }
  return (0);
}

/* --- @chtodig@ --- *
 *
 * Arguments:   @int ch@ = a character
 *
 * Returns:     The numeric value of the character as a digit, or @-1@ if
 *              it's not a digit.  Letters count as extended digits starting
 *              with value 10; case is not significant.
 */

static int chtodig(int ch)
{
  if ('0' <= ch && ch <= '9') return (ch - '0');
  else if ('a' <= ch && ch <= 'z') return (ch - 'a' + 10);
  else if ('A' <= ch && ch <= 'Z') return (ch - 'A' + 10);
  else return (-1);
}

/* --- @parse_unsigned_integer@, @parse_signed_integer@ --- *
 *
 * Arguments:   @unsigned long *u_out@, @long *i_out@ = where to put the
 *                      result
 *              @const char **q_out@ = where to put the end position
 *              @const char *p@ = pointer to the string to parse
 *
 * Returns:     Zero on success, @-1@ on error.
 *
 * Use:         Parse an integer from a string in the test-vector format.
 *              This is mostly extension of the traditional C @strtoul@
 *              format: supported inputs include:
 *
 *                * NNN -- a decimal number (even if it starts with `0');
 *                * 0xNNN -- hexadecimal;
 *                * 0oNNN -- octal;
 *                * 0bNNN -- binary;
 *                * NNrNNN -- base NN.
 *
 *              Furthermore, single underscores are permitted internally as
 *              an insignificant digit separator.
 */

static int parse_unsigned_integer(unsigned long *u_out, const char **q_out,
                                  const char *p)
{
  unsigned long u;
  int ch, d, r;
  const char *q;
  unsigned f = 0;
#define f_implicit 1u                   /* implicitly reading base 10 */
#define f_digit 2u                      /* read a real digit */
#define f_uscore 4u                     /* found an underscore */

  /* Initial setup
   *
   * This will deal with the traditional `0[box]...' prefixes.  We'll leave
   * our new `NNr...' syntax for later.
   */
  if (p[0] != '0' || !p[1]) {
    d = chtodig(*p); if (0 > d || d >= 10) return (-1);
    r = 10; u = d; p++; f |= f_implicit | f_digit;
  } else {
    u = 0; d = chtodig(p[2]);
    if (d < 0) { r = 10; f |= f_implicit | f_digit; p++; }
    else if ((p[1] == 'x' || p[1] == 'X') && d < 16) { r = 16; p += 2; }
    else if ((p[1] == 'o' || p[1] == 'O') && d < 8) { r = 8; p += 2; }
    else if ((p[1] == 'b' || p[1] == 'B') && d < 2) { r = 2; p += 2; }
    else { r = 10; f |= f_digit; p++; }
  }

  q = p;
  for (;;) {
    /* Work through the string a character at a time. */

    ch = *p; switch (ch) {

      case '_':
        /* An underscore is OK if we haven't just seen one. */

        if (f&f_uscore) goto done;
        p++; f = (f&~f_implicit) | f_uscore;
        break;

      case 'r': case 'R':
        /* An `r' is OK if the number so far is small enough to be a sensible
         * base, and we're scanning decimal implicitly.
         */

        if (!(f&f_implicit) || !u || u >= 36) goto done;
        d = chtodig(p[1]); if (0 > d || d >= u) goto done;
        r = u; u = d; f = (f&~f_implicit) | f_digit; p += 2; q = p;
        break;

      default:
        /* Otherwise we expect a valid digit and accumulate it. */
        d = chtodig(ch); if (d < 0 || d >= r) goto done;
        if (u > ULONG_MAX/r) return (-1);
        u *= r; if (u > ULONG_MAX - d) return (-1);
        u += d; f = (f&~f_uscore) | f_digit; p++; q = p;
        break;
    }
  }

done:
  if (!(f&f_digit)) return (-1);
  *u_out = u; *q_out = q; return (0);

#undef f_implicit
#undef f_digit
#undef f_uscore
}

static int parse_signed_integer(long *i_out, const char **q_out,
                                const char *p)
{
  unsigned long u;
  unsigned f = 0;
#define f_neg 1u

  /* Read an initial sign. */
  if (*p == '+') p++;
  else if (*p == '-') { f |= f_neg; p++; }

  /* Scan an unsigned number. */
  if (parse_unsigned_integer(&u, q_out, p)) return (-1);

  /* Check for signed overflow and apply the sign. */
  if (!(f&f_neg)) {
    if (u > LONG_MAX) return (-1);
    *i_out = u;
  } else {
    if (u && u - 1 > -(LONG_MIN + 1)) return (-1);
    *i_out = u ? -(long)(u - 1) - 1 : 0;
  }

  return (0);

#undef f_neg
}

/* --- @parse_unsigned@, @parse_signed@ --- *
 *
 * Arguments:   @unsigned long *u_out@ or @long *i_out@ = where to put the
 *                      result
 *              @const char *p@ = string to parse
 *              @const struct tvec_urange *ur@ or
 *                      @const struct tvec_irange *ir@ = range specification,
 *                      or null
 *              @struct tvec_state *tv@ = test vector state
 *
 * Returns:     Zero on success, @-1@ on error.
 *
 * Use:         Parse and range-check an integer.  Unlike @parse_(un)signed_
 *              integer@, these functions check that there's no cruft
 *              following the final digit, and report errors as they find
 *              them rather than leaving that to the caller.
 */

static int parse_unsigned(unsigned long *u_out, const char *p,
                          const struct tvec_urange *ur,
                          struct tvec_state *tv)
{
  unsigned long u;
  const char *q;

  if (parse_unsigned_integer(&u, &q, p))
    return (tvec_error(tv, "invalid unsigned integer `%s'", p));
  if (*q) return (tvec_syntax(tv, *q, "end-of-line"));
  if (check_unsigned_range(u, ur, tv, "integer")) return (-1);
  *u_out = u; return (0);
}

static int parse_signed(long *i_out, const char *p,
                        const struct tvec_irange *ir,
                        struct tvec_state *tv)
{
  long i;
  const char *q;

  if (parse_signed_integer(&i, &q, p))
    return (tvec_error(tv, "invalid signed integer `%s'", p));
  if (*q) return (tvec_syntax(tv, *q, "end-of-line"));
  if (check_signed_range(i, ir, tv, "integer")) return (-1);
  *i_out = i; return (0);
}
static const char size_units[] = "kMGTPEZY";

/* --- @parse_szint@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @unsigned long *u_out@ = where to put the answer
 *              @const char *delims@ = delimiters
 *              @const char *what@ = description of what we're parsing
 *
 * Returns:     Zero on success, %$-1$% on failure.
 *
 * Use:         Parse a memory size.
 */

static int parse_szint(struct tvec_state *tv, unsigned long *u_out,
                       const char *delims, const char *what)
{
  dstr d = DSTR_INIT;
  const char *p, *unit;
  unsigned long u, t;
  int rc;
  unsigned f = 0;
#define f_range 1u

  if (tvec_readword(tv, &d, 0, delims, what)) { rc = -1; goto end; }
  p = d.buf;
  if (parse_unsigned_integer(&u, &p, p)) goto bad;
  if (!*p) tvec_readword(tv, &d, &p, delims, 0);

  for (t = u, unit = size_units; *unit; unit++) {
    if (t > ULONG_MAX/1024) f |= f_range;
    else t *= 1024;
    if (*p == *unit) {
      if (f&f_range) goto rangerr;
      u = t; p++; break;
    }
  }
  if (*p == 'B') p++;
  if (*p) goto bad;

  *u_out = u; rc = 0;
end:
  dstr_destroy(&d);
  return (rc);

bad:
  tvec_error(tv, "invalid %s `%s'", what, d.buf);
  rc = -1; goto end;

rangerr:
  tvec_error(tv, "%s `%s' out of range", what, d.buf);
  rc = -1; goto end;

#undef f_range
}

/* --- @format_size@ --- *
 *
 * Arguments:   @const struct gprintf_ops *gops@ = print operations
 *              @void *go@ = print destination
 *              @unsigned long u@ = a size
 *              @unsigned style@ = style (@TVSF_...@)
 *
 * Returns:     ---
 *
 * Use:         Format @u@ as a size in bytes to the destination, expressing
 *              it with a unit prefix if this is possible exactly.
 */

static void format_size(const struct gprintf_ops *gops, void *go,
                        unsigned long u, unsigned style)
{
  const char *unit;

  if (style&TVSF_RAW)
    gprintf(gops, go, "%lu", u);
  else if (!u || u%1024)
    gprintf(gops, go, "%lu%sB", u, style&TVSF_COMPACT ? "" : " ");
  else {
    for (unit = size_units, u /= 1024;
         !(u%1024) && unit[1];
         u /= 1024, unit++);
    gprintf(gops, go, "%lu%s%cB", u, style&TVSF_COMPACT ? "" : " ", *unit);
  }
}

/*----- Floating-point utilities ------------------------------------------*/

/* --- @eqish_floating_p@ --- *
 *
 * Arguments:   @double x, y@ = two numbers to compare
 *              @const struct tvec_floatinfo *fi@ = floating-point info
 *
 * Returns:     Nonzero if  the comparand @x@ is sufficiently close to the
 *              reference @y@, or zero if it's definitely different.
 */

static int eqish_floating_p(double x, double y,
                            const struct tvec_floatinfo *fi)
{
  double t;

  if (NANP(x)) return (NANP(y)); else if (NANP(y)) return (0);
  if (INFP(x)) return (x == y); else if (INFP(y)) return (0);

  switch (fi ? fi->f&TVFF_EQMASK : TVFF_EXACT) {
    case TVFF_EXACT:
      return (x == y && NEGP(x) == NEGP(y));
    case TVFF_ABSDELTA:
      t = x - y; if (t < 0) t = -t; return (t < fi->delta);
    case TVFF_RELDELTA:
      t = 1.0 - x/y; if (t < 0) t = -t; return (t < fi->delta);
    default:
      abort();
  }
}

/* --- @format_floating@ --- *
 *
 * Arguments:   @const struct gprintf_ops *gops@ = print operations
 *              @void *go@ = print destination
 *              @double x@ = number to print
 *
 * Returns:     ---
 *
 * Use:         Print a floating-point number, accurately.
 */

static void format_floating(const struct gprintf_ops *gops, void *go,
                            double x)
{
  int prec;

  if (NANP(x))
    gprintf(gops, go, "#nan");
  else if (INFP(x))
    gprintf(gops, go, x > 0 ? "#+inf" : "#-inf");
  else {
    /* Ugh.  C doesn't provide any function for just printing a
     * floating-point number /correctly/, i.e., so that you can read the
     * result back and recover the number you first thought of.  There are
     * complicated algorithms published for doing this, but I really don't
     * want to get into that here.  So we have this.
     *
     * The sign doesn't cause significant difficulty so we're going to ignore
     * it for now.  So suppose we're given a number %$x = f b^e$%, in
     * base-%$b$% format, so %$f b^n$% and %$e$% are integers, with
     * %$0 \le f < 1$%.  We're going to convert it into the nearest integer
     * of the form %$X = F B^E$%, with similar conditions, only with the
     * additional requirement that %$X$% is normalized, i.e., that %$X = 0$%
     * or %$F \ge B^{-N}$%.
     *
     * We're rounding to the nearest such %$X$%.  If there is to be ambiguity
     * in the conversion, then some %$x = f b^e$% and the next smallest
     * representable number %$x' = x + b^{e-n}$% must both map to the same
     * %$X$%, which means both %$x$% and %$x'$% must be nearer to %$X$% than
     * any other number representable in the target system.  The nest larger
     * number is %$X' = X + B^{E-N}$%; the next smaller number will normally
     * be %$W = X - B^{E-N}$%, but if %$F = 1/B$ then the next smaller number
     * is actually %$X - B^{E-N-1}$%.  We ignore this latter possibility in
     * the pursuit of a conservative estimate (though actually it doesn't
     * matter).
     *
     * If both %$x$% and %$x'$% map to %$X$% then we must have
     * %$L = X - B^{E-N}/2 \le x$% and %$x + b^{e-n} \le R = X + B^{E-N}/2$%;
     * so firstly %$f b^e = x \ge L = W + B^{E-N}/2 > W = (F - B^{-N}) B^E$%,
     * and secondly %$b^{e-n} \le B^{E-N}$%.  Since these inequalities are in
     * opposite senses, we can divide, giving
     *
     *         %$f b^e/b^{e-n} > (F - B^{-N}) B^E/B^{E-N}$% ,
     *
     * whence
     *
     *         %$f b^n > (F - B^{-N}) B^N = F B^N - 1$% .
     *
     * Now %$f \le 1 - b^{-n}$%, and %$F \ge B^{-1}$%, so, for this to be
     * possible, it must be the case that
     *
     *         %$(1 - b^{-n}) b^n = b^n - 1 > B^{N-1} - 1$% .
     *
     * Then rearrange and take logarithms, obtaining
     *
     *         %$(N - 1) \log B < n \log b$% ,
     *
     * and so
     *
     *         %$N < n \log b/\log B + 1$% .
     *
     * Recall that this is a necessary condition for a collision to occur; we
     * are therefore safe whenever
     *
     *         %$N \ge n \log b/\log B + 1$% ;
     *
     * so, taking ceilings,
     *
     *         %$N \ge \lceil n \log b/\log B \rceil + 1$% .
     *
     * So that's why we have this.
     *
     * I'm going to assume that @n = DBL_MANT_DIG@ is sufficiently small that
     * we can calculate this without ending up on the wrong side of an
     * integer boundary.
     *
     * In C11, we have @DBL_DECIMAL_DIG@, which should be the same value only
     * as a constant.  Except that modern compilers are more than clever
     * enough to work out that this is a constant anyway.
     *
     * This is sometimes an overestimate: we'll print out meaningless digits
     * that don't represent anything we actually know about the number in
     * question.  To fix that, we'd need a complicated algorithm like Steele
     * and White's Dragon4, Gay's @dtoa@, or Burger and Dybvig's algorithm
     * (note that Loitsch's Grisu2 is conservative, and Grisu3 hands off to
     * something else in difficult situations).
     */

    prec = ceil(DBL_MANT_DIG*log(FLT_RADIX)/log(10)) + 1;
    gprintf(gops, go, "%.*g", prec, x);
  }
}

/* --- @parse_floating@ --- *
 *
 * Arguments:   @double *x_out@ = where to put the result
 *              @const char *q_out@ = where to leave end pointer, or null
 *              @const char *p@ = string to parse
 *              @const struct tvec_floatinfo *fi@ = floating-point info
 *              @struct tvec_state *tv@ = test vector state
 *
 * Returns:     Zero on success, @-1@ on error.
 *
 * Use:         Parse a floating-point number from a string.  Reports any
 *              necessary errors.  If @q_out@ is not null then trailing
 *              material is permitted and a pointer to it (or the end of the
 *              string) is left in @*q_out@.
 */

static int parse_floating(double *x_out, const char **q_out, const char *p,
                          const struct tvec_floatinfo *fi,
                          struct tvec_state *tv)
{
  const char *pp; char *q;
  dstr d = DSTR_INIT;
  double x;
  int olderr, rc;

  /* Check for special tokens. */
  if (STRCMP(p, ==, "#nan")) {
#ifdef NAN
    if (q_out) *q_out = p + strlen(p);
    x = NAN; rc = 0;
#else
    tvec_error(tv, "NaN not supported on this system");
    rc = -1; goto end;
#endif
  }

  else if (STRCMP(p, ==, "#inf") ||
           STRCMP(p, ==, "#+inf") || STRCMP(p, ==, "+#inf")) {
#ifdef INFINITY
    if (q_out) *q_out = p + strlen(p);
    x = INFINITY; rc = 0;
#else
    tvec_error(tv, "infinity not supported on this system");
    rc = -1; goto end;
#endif
  }

  else if (STRCMP(p, ==, "#-inf") || STRCMP(p, ==, "-#inf")) {
#ifdef INFINITY
    if (q_out) *q_out = p + strlen(p);
    x = -INFINITY; rc = 0;
#else
    tvec_error(tv, "infinity not supported on this system");
    rc = -1; goto end;
#endif
  }

  /* Check that this looks like a number, so we can exclude `strtod'
   * recognizing its own non-finite number tokens.
   */
  else {
    pp = p;
    if (*pp == '+' || *pp == '-') pp++;
    if (*pp == '.') pp++;
    if (!ISDIGIT(*pp)) {
      tvec_syntax(tv, *p ? *p : fgetc(tv->fp), "floating-point number");
      rc = -1; goto end;
    }

    /* Parse the number using the system parser. */
    olderr = errno; errno = 0;
    x = strtod(p, &q);
    if (q_out) *q_out = q;
    else if (*q) { tvec_syntax(tv, *q, "end-of-line"); rc = -1; goto end; }
    if (errno && (errno != ERANGE || (x > 0 ? -x : x) == HUGE_VAL)) {
      tvec_error(tv, "invalid floating-point number `%.*s': %s",
                 (int)(q - p), p, strerror(errno));
      rc = -1; goto end;
    }
    errno = olderr;
  }

  /* Check that the number is acceptable. */
  if (NANP(x) && fi && !(fi->f&TVFF_NANOK)) {
    tvec_error(tv, "#nan not allowed here");
    rc = -1; goto end;
  }

  if (fi && ((!(fi->f&TVFF_NOMIN) && x < fi->min) ||
             (!(fi->f&TVFF_NOMAX) && x > fi->max))) {
    dstr_puts(&d, "floating-point number ");
    format_floating(&dstr_printops, &d, x);
    dstr_puts(&d, " out of range (must be in ");
    if (fi->f&TVFF_NOMIN)
      dstr_puts(&d, "(#-inf");
    else
      { dstr_putc(&d, '['); format_floating(&dstr_printops, &d, fi->min); }
    dstr_puts(&d, " .. ");
    if (fi->f&TVFF_NOMAX)
      dstr_puts(&d, "#+inf)");
    else
      { format_floating(&dstr_printops, &d, fi->max); dstr_putc(&d, ']'); }
    dstr_putc(&d, ')'); dstr_putz(&d);
    tvec_error(tv, "%s", d.buf); rc = -1; goto end;
  }

  /* All done. */
  *x_out = x; rc = 0;
end:
  dstr_destroy(&d);
  return (rc);
}

/*----- String utilities --------------------------------------------------*/

/* Special character name table. */
static const struct chartab {
  const char *name;                     /* character name */
  int ch;                               /* character value */
  unsigned f;                           /* flags: */
#define CTF_PREFER 1u                   /*   preferred name */
#define CTF_SHORT 2u                    /*   short name (compact style) */
} chartab[] = {
  { "#eof",             EOF,    CTF_PREFER | CTF_SHORT },
  { "#nul",             '\0',   CTF_PREFER },
  { "#bell",            '\a',   CTF_PREFER },
  { "#ding",            '\a',   0 },
  { "#bel",             '\a',   CTF_SHORT },
  { "#backspace",       '\b',   CTF_PREFER },
  { "#bs",              '\b',   CTF_SHORT },
  { "#escape",          '\x1b', CTF_PREFER },
  { "#esc",             '\x1b', CTF_SHORT },
  { "#formfeed",        '\f',   CTF_PREFER },
  { "#ff",              '\f',   CTF_SHORT },
  { "#newline",         '\n',   CTF_PREFER },
  { "#linefeed",        '\n',   0 },
  { "#lf",              '\n',   CTF_SHORT },
  { "#nl",              '\n',   0 },
  { "#return",          '\r',   CTF_PREFER },
  { "#carriage-return", '\r',   0 },
  { "#cr",              '\r',   CTF_SHORT },
  { "#tab",             '\t',   CTF_PREFER | CTF_SHORT },
  { "#horizontal-tab",  '\t',   0 },
  { "#ht",              '\t',   0 },
  { "#vertical-tab",    '\v',   CTF_PREFER },
  { "#vt",              '\v',   CTF_SHORT },
  { "#space",           ' ',    0 },
  { "#spc",             ' ',    CTF_SHORT },
  { "#delete",          '\x7f', CTF_PREFER },
  { "#del",             '\x7f', CTF_SHORT },
  { 0,                  0,      0 }
};

/* --- @find_charname@ --- *
 *
 * Arguments:   @int ch@ = character to match
 *              @unsigned f@ = flags (@CTF_...@) to match
 *
 * Returns:     The name of the character, or null if no match is found.
 *
 * Use:         Looks up a name for a character.  Specifically, it returns
 *              the first entry in the @chartab@ table which matches @ch@ and
 *              which has one of the flags @f@ set.
 */

static const char *find_charname(int ch, unsigned f)
{
  const struct chartab *ct;

  for (ct = chartab; ct->name; ct++)
    if (ct->ch == ch && (ct->f&f)) return (ct->name);
  return (0);
}

/* --- @read_charname@ --- *
 *
 * Arguments:   @int *ch_out@ = where to put the character
 *              @const char *p@ = character name
 *              @unsigned f@ = flags (@TCF_...@)
 *
 * Returns:     Zero if a match was found, @-1@ if not.
 *
 * Use:         Looks up a character by name.  If @RCF_EOFOK@ is set in @f@,
 *              then the @EOF@ marker can be matched; otherwise it can't.
 */

#define RCF_EOFOK 1u
static int read_charname(int *ch_out, const char *p, unsigned f)
{
  const struct chartab *ct;

  for (ct = chartab; ct->name; ct++)
    if (STRCMP(p, ==, ct->name) && ((f&RCF_EOFOK) || ct->ch >= 0))
      { *ch_out = ct->ch; return (0); }
  return (-1);
}

/* --- @format_charesc@ --- *
 *
 * Arguments:   @const struct gprintf_ops *gops@ = print operations
 *              @void *go@ = print destination
 *              @int ch@ = character to format
 *              @unsigned f@ = flags (@FCF_...@)
 *
 * Returns:     ---
 *
 * Use:         Format a character as an escape sequence, possibly as part of
 *              a larger string.  If @FCF_BRACE@ is set in @f@, then put
 *              braces around a `\x...'  code, so that it's suitable for use
 *              in a longer string.
 */

#define FCF_BRACE 1u
static void format_charesc(const struct gprintf_ops *gops, void *go,
                           int ch, unsigned f)
{
  switch (ch) {
    case '\a': gprintf(gops, go, "\\a"); break;
    case '\b': gprintf(gops, go, "\\b"); break;
    case '\x1b': gprintf(gops, go, "\\e"); break;
    case '\f': gprintf(gops, go, "\\f"); break;
    case '\r': gprintf(gops, go, "\\r"); break;
    case '\n': gprintf(gops, go, "\\n"); break;
    case '\t': gprintf(gops, go, "\\t"); break;
    case '\v': gprintf(gops, go, "\\v"); break;
    case '\\': gprintf(gops, go, "\\\\"); break;
    case '\'': gprintf(gops, go, "\\'"); break;
    case '\0':
      if (f&FCF_BRACE) gprintf(gops, go, "\\{0}");
      else gprintf(gops, go, "\\0");
      break;
    default:
      if (f&FCF_BRACE)
        gprintf(gops, go, "\\x{%0*x}", hex_width(UCHAR_MAX), ch);
      else
        gprintf(gops, go, "\\x%0*x", hex_width(UCHAR_MAX), ch);
      break;
  }
}

/* --- @format_char@ --- *
 *
 * Arguments:   @const struct gprintf_ops *gops@ = print operations
 *              @void *go@ = print destination
 *              @int ch@ = character to format
 *
 * Returns:     ---
 *
 * Use:         Format a single character.
 */

static void format_char(const struct gprintf_ops *gops, void *go, int ch)
{
  switch (ch) {
    case '\\': case '\'': escape:
      gprintf(gops, go, "'");
      format_charesc(gops, go, ch, 0);
      gprintf(gops, go, "'");
      break;
    default:
      if (!isprint(ch)) goto escape;
      gprintf(gops, go, "'%c'", ch);
      break;
  }
}

/* --- @maybe_format_unsigned_char@, @maybe_format_signed_char@ --- *
 *
 * Arguments:   @const struct gprintf_ops *gops@ = print operations
 *              @void *go@ = print destination
 *              @unsigned long u@ or @long i@ = an integer
 *
 * Returns:     ---
 *
 * Use:         Format a (signed or unsigned) integer as a character, if it's
 *              in range, printing something like `= 'q''.  It's assumed that
 *              a comment marker has already been output.
 */

static void maybe_format_unsigned_char
  (const struct gprintf_ops *gops, void *go, unsigned long u)
{
  const char *p;

  p = find_charname(u, CTF_PREFER);
  if (p) gprintf(gops, go, " = %s", p);
  if (u < UCHAR_MAX)
    { gprintf(gops, go, " = "); format_char(gops, go, u); }
}

static void maybe_format_signed_char
  (const struct gprintf_ops *gops, void *go, long i)
{
  const char *p;

  p = find_charname(i, CTF_PREFER);
  if (p) gprintf(gops, go, " = %s", p);
  if (0 <= i && i < UCHAR_MAX)
    { gprintf(gops, go, " = "); format_char(gops, go, i); }
}

/* --- @read_charesc@ --- *
 *
 * Arguments:   @int *ch_out@ = where to put the result
 *              @struct tvec_state *tv@ = test vector state
 *
 * Returns:     Zero on success, @-1@ on error.
 *
 * Use:         Parse and convert an escape sequence from @tv@'s input
 *              stream, assuming that the initial `\' has already been read.
 *              Reports errors as appropriate.
 */

static int read_charesc(int *ch_out, struct tvec_state *tv)
{
  int ch, i, esc;
  unsigned f = 0;
#define f_brace 1u

  ch = getc(tv->fp);
  switch (ch) {

    /* Things we shouldn't find. */
    case EOF: case '\n': return (tvec_syntax(tv, ch, "string escape"));

    /* Single-character escapes. */
    case '\'': *ch_out = '\''; break;
    case '\\': *ch_out = '\\'; break;
    case '"': *ch_out = '"'; break;
    case 'a': *ch_out = '\a'; break;
    case 'b': *ch_out = '\b'; break;
    case 'e': *ch_out = '\x1b'; break;
    case 'f': *ch_out = '\f'; break;
    case 'n': *ch_out = '\n'; break;
    case 'r': *ch_out = '\r'; break;
    case 't': *ch_out = '\t'; break;
    case 'v': *ch_out = '\v'; break;

    /* Hex escapes, with and without braces. */
    case 'x':
      ch = getc(tv->fp);
      if (ch == '{') { f |= f_brace; ch = getc(tv->fp); }
      else f &= ~f_brace;
      esc = chtodig(ch);
      if (esc < 0 || esc >= 16) return (tvec_syntax(tv, ch, "hex digit"));
      for (;;) {
        ch = getc(tv->fp); i = chtodig(ch); if (i < 0 || i >= 16) break;
        esc = 16*esc + i;
        if (esc > UCHAR_MAX)
          return (tvec_error(tv,
                             "character code %d out of range", esc));
      }
      if (!(f&f_brace)) ungetc(ch, tv->fp);
      else if (ch != '}') return (tvec_syntax(tv, ch, "`}'"));
      *ch_out = esc;
      break;

    /* Other things, primarily octal escapes. */
    case '{':
      f |= f_brace; ch = getc(tv->fp);
      /* fall through */
    default:
      if ('0' <= ch && ch < '8') {
        i = 1; esc = ch - '0';
        for (;;) {
          ch = getc(tv->fp);
          if ('0' > ch || ch >= '8') { ungetc(ch, tv->fp); break; }
          esc = 8*esc + ch - '0';
          i++; if (i >= 3) break;
        }
        if (f&f_brace) {
          ch = getc(tv->fp);
          if (ch != '}') return (tvec_syntax(tv, ch, "`}'"));
        }
        if (esc > UCHAR_MAX)
          return (tvec_error(tv,
                             "character code %d out of range", esc));
        *ch_out = esc; break;
      } else
        return (tvec_syntax(tv, ch, "string escape"));
  }

  /* Done. */
  return (0);

#undef f_brace
}

/* --- @read_quoted_string@ --- *
 *
 * Arguments:   @dstr *d@ = string to write to
 *              @int quote@ = initial quote, `'' or `"'
 *              @struct tvec_state *tv@ = test vector state
 *
 * Returns:     Zero on success, @-1@ on error.
 *
 * Use:         Read the rest of a quoted string into @d@, reporting errors
 *              as appropriate.
 *
 *              A single-quoted string is entirely literal.  A double-quoted
 *              string may contain C-like escapes.
 */

static int read_quoted_string(dstr *d, int quote, struct tvec_state *tv)
{
  int ch;

  for (;;) {
    ch = getc(tv->fp);
    switch (ch) {
      case EOF: case '\n':
        return (tvec_syntax(tv, ch, "`%c'", quote));
      case '\\':
        if (quote == '\'') goto ordinary;
        ch = getc(tv->fp); if (ch == '\n') { tv->lno++; break; }
        ungetc(ch, tv->fp); if (read_charesc(&ch, tv)) return (-1);
        goto ordinary;
      default:
        if (ch == quote) goto end;
      ordinary:
        DPUTC(d, ch);
        break;
    }
  }

end:
  DPUTZ(d);
  return (0);
}

/* --- @collect_bare@ --- *
 *
 * Arguments:   @dstr *d@ = string to write to
 *              @struct tvec_state *tv@ = test vector state
 *
 * Returns:     Zero on success, @-1@ on error.
 *
 * Use:         Read barewords and the whitespace between them.  Stop when we
 *              encounter something which can't start a bareword.
 */

static int collect_bare(dstr *d, struct tvec_state *tv)
{
  size_t pos = d->len;
  enum { WORD, SPACE, ESCAPE }; unsigned s = WORD;
  int ch, rc;

  for (;;) {
    ch = getc(tv->fp);
    switch (ch) {
      case EOF:
        tvec_syntax(tv, ch, "bareword");
        rc = -1; goto end;
      case '\n':
        if (s == ESCAPE) { tv->lno++; goto addch; }
        if (s == WORD) pos = d->len;
        ungetc(ch, tv->fp); if (tvec_nexttoken(tv)) { rc = -1; goto end; }
        DPUTC(d, ' '); s = SPACE;
        break;
      case '"': case '\'': case '!': case '#': case ')': case '}': case ']':
        if (s == SPACE) { ungetc(ch, tv->fp); goto done; }
        goto addch;
      case '\\':
        s = ESCAPE;
        break;
      default:
        if (s != ESCAPE && isspace(ch)) {
          if (s == WORD) pos = d->len;
          DPUTC(d, ch); s = SPACE;
          break;
        }
      addch:
        DPUTC(d, ch); s = WORD;
    }
  }

done:
  if (s == SPACE) d->len = pos;
  DPUTZ(d); rc = 0;
end:
  return (rc);
}

/* --- @set_up_encoding@ --- *
 *
 * Arguments:   @const codec_class **ccl_out@ = where to put the class
 *              @unsigned *f_out@ = where to put the flags
 *              @unsigned code@ = the coding scheme to use (@TVEC_...@)
 *
 * Returns:     ---
 *
 * Use:         Helper for @read_compound_string@ below.
 *
 *              Return the appropriate codec class and flags for @code@.
 *              Leaves @*ccl_out@ null if the coding scheme doesn't have a
 *              backing codec class (e.g., @TVCODE_BARE@).
 */

enum { TVCODE_BARE, TVCODE_HEX, TVCODE_BASE64, TVCODE_BASE32 };
static void set_up_encoding(const codec_class **ccl_out, unsigned *f_out,
                            unsigned code)
{
  switch (code) {
    case TVCODE_BARE:
      *ccl_out = 0; *f_out = 0;
      break;
    case TVCODE_HEX:
      *ccl_out = &hex_class; *f_out = CDCF_IGNCASE;
      break;
    case TVCODE_BASE32:
      *ccl_out = &base32_class; *f_out = CDCF_IGNCASE | CDCF_IGNEQPAD;
      break;
    case TVCODE_BASE64:
      *ccl_out = &base64_class; *f_out = CDCF_IGNEQPAD;
      break;
    default:
      abort();
  }
}

/* --- @flush_codec@ --- *
 *
 * Arguments:   @codec *cdc@ = a codec, or null
 *              @dstr *d@ = output string
 *              @struct tvec_state *tv@ = test vector state
 *
 * Returns:     Zero on success, @-1@ on error.
 *
 * Use:         Helper for @read_compound_string@ below.
 *
 *              Flush out any final buffered material from @cdc@, and check
 *              that it's in a good state.  Frees the codec on success.  Does
 *              nothing if @cdc@ is null.
 */

static int flush_codec(codec *cdc, dstr *d, struct tvec_state *tv)
{
  int err;

  if (cdc) {
    err = cdc->ops->code(cdc, 0, 0, d);
    if (err)
      return (tvec_error(tv, "invalid %s sequence end: %s",
                         cdc->ops->c->name, codec_strerror(err)));
    cdc->ops->destroy(cdc);
  }
  return (0);
}

/* --- @read_compound_string@ --- *
 *
 * Arguments:   @void **p_inout@ = address of output buffer pointer
 *              @size_t *sz_inout@ = address of buffer size
 *              @unsigned code@ = initial interpretation of barewords
 *              @unsigned f@ = other flags (@RCSF_...@)
 *              @struct tvec_state *tv@ = test vector state
 *
 * Returns:     Zero on success, @-1@ on error.
 *
 * Use:         Parse a compound string, i.e., a sequence of stringish pieces
 *              which might be quoted strings, character names, or barewords
 *              to be decoded accoding to @code@, interspersed with
 *              additional directives.
 *
 *              If the initial buffer pointer is non-null and sufficiently
 *              large, then it will be reused; otherwise, it is freed and a
 *              fresh, sufficiently large buffer is allocated and returned.
 */

#define RCSF_NESTED 1u
static int read_compound_string(void **p_inout, size_t *sz_inout,
                                unsigned code, unsigned f,
                                struct tvec_state *tv)
{
  const codec_class *ccl; unsigned cdf;
  codec *cdc;
  dstr d = DSTR_INIT, w = DSTR_INIT;
  char *p;
  const char *q;
  void *pp = 0; size_t sz;
  unsigned long n;
  int ch, err, rc;

  set_up_encoding(&ccl, &cdf, code); cdc = 0;

  if (tvec_nexttoken(tv)) return (tvec_syntax(tv, fgetc(tv->fp), "string"));
  do {
    ch = getc(tv->fp);
    switch (ch) {

      case ')': case ']': case '}':
        /* Close brackets.  Leave these for recursive caller if there is one,
         * or just complain.
         */

        if (!(f&RCSF_NESTED))
          { rc = tvec_syntax(tv, ch, "string"); goto end; }
        ungetc(ch, tv->fp); goto done;

      case '"': case '\'':
        /* Quotes.  Read a quoted string. */

        if (cdc && flush_codec(cdc, &d, tv)) { rc = -1; goto end; }
        cdc = 0;
        if (read_quoted_string(&d, ch, tv)) { rc = -1; goto end; }
        break;

      case '#':
        /* A named character. */

        ungetc(ch, tv->fp);
        if (cdc && flush_codec(cdc, &d, tv)) { rc = -1; goto end; }
        cdc = 0;
        DRESET(&w); tvec_readword(tv, &w, 0, ";", "character name");
        if (STRCMP(w.buf, ==, "#empty")) break;
        if (read_charname(&ch, w.buf, RCF_EOFOK)) {
          rc = tvec_error(tv, "unknown character name `%s'", d.buf);
          goto end;
        }
        DPUTC(&d, ch); break;

      case '!':
        /* A magic keyword. */

        if (cdc && flush_codec(cdc, &d, tv)) { rc = -1; goto end; }
        cdc = 0;
        ungetc(ch, tv->fp);
        DRESET(&w); tvec_readword(tv, &w, 0, ";", "`!'-keyword");

        /* Change bareword coding system. */
        if (STRCMP(w.buf, ==, "!bare"))
          { code = TVCODE_BARE; set_up_encoding(&ccl, &cdf, code); }
        else if (STRCMP(w.buf, ==, "!hex"))
          { code = TVCODE_HEX; set_up_encoding(&ccl, &cdf, code); }
        else if (STRCMP(w.buf, ==, "!base32"))
          { code = TVCODE_BASE32; set_up_encoding(&ccl, &cdf, code); }
        else if (STRCMP(w.buf, ==, "!base64"))
          { code = TVCODE_BASE64; set_up_encoding(&ccl, &cdf, code); }

        /* Repeated substrings. */
        else if (STRCMP(w.buf, ==, "!repeat")) {
          if (tvec_nexttoken(tv)) {
            rc = tvec_syntax(tv, fgetc(tv->fp), "repeat count");
            goto end;
          }
          DRESET(&w);
          if (tvec_readword(tv, &w, 0, ";{", "repeat count"))
            { rc = -1; goto end;  }
          if (parse_unsigned_integer(&n, &q, w.buf)) {
            rc = tvec_error(tv, "invalid repeat count `%s'", w.buf);
            goto end;
          }
          if (*q) { rc = tvec_syntax(tv, *q, "`{'"); goto end; }
          if (tvec_nexttoken(tv))
            { rc = tvec_syntax(tv, fgetc(tv->fp), "`{'"); goto end; }
          ch = getc(tv->fp); if (ch != '{')
            { rc = tvec_syntax(tv, ch, "`{'"); goto end; }
          sz = 0;
          if (read_compound_string(&pp, &sz, code, f | RCSF_NESTED, tv))
            { rc = -1; goto end; }
          ch = getc(tv->fp); if (ch != '}')
            { rc = tvec_syntax(tv, ch, "`}'"); goto end; }
          if (sz) {
            if (n > (size_t)-1/sz)
              { rc = tvec_error(tv, "repeat size out of range"); goto end; }
            dstr_ensure(&d, n*sz);
            if (sz == 1)
              { memset(d.buf + d.len, *(unsigned char *)pp, n); d.len += n; }
            else
              for (; n--; d.len += sz) memcpy(d.buf + d.len, pp, sz);
          }
          xfree(pp); pp = 0;
        }

        /* Anything else is an error. */
        else {
          tvec_error(tv, "unknown string keyword `%s'", w.buf);
          rc = -1; goto end;
        }
        break;

      default:
        /* A bareword.  Process it according to the current coding system. */

        switch (code) {
          case TVCODE_BARE:
            ungetc(ch, tv->fp);
            if (collect_bare(&d, tv)) goto done;
            break;
          default:
            assert(ccl);
            ungetc(ch, tv->fp); DRESET(&w);
            if (tvec_readword(tv, &w, 0, ";",
                              "%s-encoded fragment", ccl->name))
              { rc = -1; goto end; }
            if (!cdc) cdc = ccl->decoder(cdf);
            err = cdc->ops->code(cdc, w.buf, w.len, &d);
            if (err) {
              tvec_error(tv, "invalid %s fragment `%s': %s",
                         ccl->name, w.buf, codec_strerror(err));
              rc = -1; goto end;
            }
            break;
        }
        break;
    }
  } while (!tvec_nexttoken(tv));

done:
  /* Wrap things up. */
  if (cdc && flush_codec(cdc, &d, tv)) { rc = -1; goto end; }
  cdc = 0;
  if (*sz_inout <= d.len)
    { xfree(*p_inout); *p_inout = xmalloc(d.len + 1); }
  p = *p_inout; memcpy(p, d.buf, d.len); p[d.len] = 0; *sz_inout = d.len;
  rc = 0;

end:
  /* Clean up any debris. */
  if (cdc) cdc->ops->destroy(cdc);
  if (pp) xfree(pp);
  dstr_destroy(&d); dstr_destroy(&w);
  return (rc);
}

/*----- Signed and unsigned integer types ---------------------------------*/

/* --- @init_int@, @init_uint@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     ---
 *
 * Use:         Initialize a register value.
 *
 *              Integer values are initialized to zero.
 */

static void init_int(union tvec_regval *rv, const struct tvec_regdef *rd)
  { rv->i = 0; }

static void init_uint(union tvec_regval *rv, const struct tvec_regdef *rd)
  { rv->u = 0; }

/* --- @eq_int@, @eq_uint@ --- *
 *
 * Arguments:   @const union tvec_regval *rv0, *rv1@ = register values
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Nonzero if the values are equal, zero if unequal
 *
 * Use:         Compare register values for equality.
 */

static int eq_int(const union tvec_regval *rv0, const union tvec_regval *rv1,
                  const struct tvec_regdef *rd)
  { return (rv0->i == rv1->i); }

static int eq_uint(const union tvec_regval *rv0,
                   const union tvec_regval *rv1,
                   const struct tvec_regdef *rd)
  { return (rv0->u == rv1->u); }

/* --- @tobuf_int@, @tobuf_uint@ --- *
 *
 * Arguments:   @buf *b@ = buffer
 *              @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Zero on success, %$-1$% on failure.
 *
 * Use:         Serialize a register value to a buffer.
 *
 *              Integer values are serialized as little-endian 64-bit signed
 *              or unsigned integers.
 */

static int tobuf_int(buf *b, const union tvec_regval *rv,
                     const struct tvec_regdef *rd)
  { return (signed_to_buf(b, rv->i)); }

static int tobuf_uint(buf *b, const union tvec_regval *rv,
                       const struct tvec_regdef *rd)
  { return (unsigned_to_buf(b, rv->u)); }

/* --- @frombuf_int@, @frombuf_uint@ --- *
 *
 * Arguments:   @buf *b@ = buffer
 *              @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Zero on success, %$-1$% on failure.
 *
 * Use:         Deserialize a register value from a buffer.
 *
 *              Integer values are serialized as 64-bit signed or unsigned
 *              integers.
 */

static int frombuf_int(buf *b, union tvec_regval *rv,
                       const struct tvec_regdef *rd)
  { return (signed_from_buf(b, &rv->i)); }

static int frombuf_uint(buf *b, union tvec_regval *rv,
                        const struct tvec_regdef *rd)
  { return (unsigned_from_buf(b, &rv->u)); }

/* --- @parse_int@, @parse_uint@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @struct tvec_state *tv@ = test-vector state
 *
 * Returns:     Zero on success, %$-1$% on error.
 *
 * Use:         Parse a register value from an input file.
 *
 *              Integers may be input in decimal, hex, binary, or octal,
 *              following approximately usual conventions.
 *
 *                * Signed integers may be preceded with a `+' or `-' sign.
 *
 *                * Decimal integers are just a sequence of decimal digits
 *                  `0' ... `9'.
 *
 *                * Octal integers are a sequence of digits `0' ... `7',
 *                  preceded by `0o' or `0O'.
 *
 *                * Hexadecimal integers are a sequence of digits `0'
 *                  ... `9', `a' ... `f', or `A' ... `F', preceded by `0x' or
 *                  `0X'.
 *
 *                * Radix-B integers are a sequence of digits `0' ... `9',
 *                  `a' ... `f', or `A' ... `F', each with value less than B,
 *                  preceded by `Br' or `BR', where 0 < B < 36 is expressed
 *                  in decimal without any leading `0' or internal
 *                  underscores `_'.
 *
 *                * A digit sequence may contain internal underscore `_'
 *                  separators, but not before or after all of the digits;
 *                  and two consecutive `_' characters are not permitted.
 */

static int parse_int(union tvec_regval *rv, const struct tvec_regdef *rd,
                     struct tvec_state *tv)
{
  dstr d = DSTR_INIT;
  int rc;

  if (tvec_readword(tv, &d, 0, ";", "signed integer"))
    { rc = -1; goto end; }
  if (parse_signed(&rv->i, d.buf, rd->arg.p, tv)) { rc = -1; goto end; }
  if (tvec_flushtoeol(tv, 0)) { rc = -1; goto end; }
  rc = 0;
end:
  dstr_destroy(&d);
  return (rc);
}

static int parse_uint(union tvec_regval *rv, const struct tvec_regdef *rd,
                      struct tvec_state *tv)
{
  dstr d = DSTR_INIT;
  int rc;

  if (tvec_readword(tv, &d, 0, ";", "unsigned integer"))
    { rc = -1; goto end; }
  if (parse_unsigned(&rv->u, d.buf, rd->arg.p, tv)) { rc = -1; goto end; }
  if (tvec_flushtoeol(tv, 0)) { rc = -1; goto end; }
  rc = 0;
end:
  dstr_destroy(&d);
  return (rc);
}

/* --- @dump_int@, @dump_uint@ --- *
 *
 * Arguments:   @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @unsigned style@ = output style (@TVSF_...@)
 *              @const struct gprintf_ops *gops@, @void *gp@ = format output
 *
 * Returns:     ---
 *
 * Use:         Dump a register value to the format output.
 *
 *              Integer values are dumped in decimal and, unless compact
 *              output is requested, hex, and maybe a character, as a
 *              comment.
 */

static void dump_int(const union tvec_regval *rv,
                     const struct tvec_regdef *rd,
                     unsigned style,
                     const struct gprintf_ops *gops, void *go)
{
  if (style&TVSF_RAW) gprintf(gops, go, "int:");
  gprintf(gops, go, "%ld", rv->i);
  if (!(style&(TVSF_COMPACT | TVSF_RAW))) {
    gprintf(gops, go, " ; = ");
    format_signed_hex(gops, go, rv->i);
    maybe_format_signed_char(gops, go, rv->i);
  }
}

static void dump_uint(const union tvec_regval *rv,
                      const struct tvec_regdef *rd,
                      unsigned style,
                      const struct gprintf_ops *gops, void *go)
{
  if (style&TVSF_RAW) gprintf(gops, go, "uint:");
  gprintf(gops, go, "%lu", rv->u);
  if (!(style&(TVSF_COMPACT | TVSF_RAW))) {
    gprintf(gops, go, " ; = ");
    format_unsigned_hex(gops, go, rv->u);
    maybe_format_unsigned_char(gops, go, rv->u);
  }
}

/* Integer type definitions. */
const struct tvec_regty tvty_int = {
  init_int, trivial_release, eq_int,
  tobuf_int, frombuf_int,
  parse_int, dump_int
};
const struct tvec_regty tvty_uint = {
  init_uint, trivial_release, eq_uint,
  tobuf_uint, frombuf_uint,
  parse_uint, dump_uint
};

/* Predefined integer ranges. */
const struct tvec_irange
  tvrange_schar = { SCHAR_MIN, SCHAR_MAX },
  tvrange_short = { SHRT_MIN, SHRT_MAX },
  tvrange_int = { INT_MIN, INT_MAX },
  tvrange_long = { LONG_MIN, LONG_MAX },
  tvrange_sbyte = { -128, 127 },
  tvrange_i16 = { -32768, +32767 },
  tvrange_i32 = { -2147483648, 2147483647 };
const struct tvec_urange
  tvrange_uchar = { 0, UCHAR_MAX },
  tvrange_ushort = { 0, USHRT_MAX },
  tvrange_uint = { 0, UINT_MAX },
  tvrange_ulong = { 0, ULONG_MAX },
  tvrange_size = { 0, (size_t)-1 },
  tvrange_byte = { 0, 255 },
  tvrange_u16 = { 0, 65535 },
  tvrange_u32 = { 0, 4294967295 };

/* --- @tvec_claimeq_int@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @long i0, i1@ = two signed integers
 *              @const char *file@, @unsigned @lno@ = calling file and line
 *              @const char *expr@ = the expression to quote on failure
 *
 * Returns:     Nonzero if @i0@ and @i1@ are equal, otherwise zero.
 *
 * Use:         Check that values of @i0@ and @i1@ are equal.  As for
 *              @tvec_claim@ above, a test case is automatically begun and
 *              ended if none is already underway.  If the values are
 *              unequal, then @tvec_fail@ is called, quoting @expr@, and the
 *              mismatched values are dumped: @i0@ is printed as the output
 *              value and @i1@ is printed as the input reference.
 */

int tvec_claimeq_int(struct tvec_state *tv, long i0, long i1,
                     const char *file, unsigned lno, const char *expr)
{
  tv->out[0].v.i = i0; tv->in[0].v.i = i1;
  return (tvec_claimeq(tv, &tvty_int, 0, file, lno, expr));
}

/* --- @tvec_claimeq_uint@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @unsigned long u0, u1@ = two unsigned integers
 *              @const char *file@, @unsigned @lno@ = calling file and line
 *              @const char *expr@ = the expression to quote on failure
 *
 * Returns:     Nonzero if @u0@ and @u1@ are equal, otherwise zero.
 *
 * Use:         Check that values of @u0@ and @u1@ are equal.  As for
 *              @tvec_claim@ above, a test case is automatically begun and
 *              ended if none is already underway.  If the values are
 *              unequal, then @tvec_fail@ is called, quoting @expr@, and the
 *              mismatched values are dumped: @u0@ is printed as the output
 *              value and @u1@ is printed as the input reference.
 */

int tvec_claimeq_uint(struct tvec_state *tv,
                      unsigned long u0, unsigned long u1,
                      const char *file, unsigned lno, const char *expr)
{
  tv->out[0].v.u = u0; tv->in[0].v.u = u1;
  return (tvec_claimeq(tv, &tvty_uint, 0, file, lno, expr));
}

/*----- Size type ---------------------------------------------------------*/

/* --- @parse_size@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @struct tvec_state *tv@ = test-vector state
 *
 * Returns:     Zero on success, %$-1$% on error.
 *
 * Use:         Parse a register value from an input file.
 *
 *              The input format for a size value consists of an unsigned
 *              integer followed by an optional unit specifier consisting of
 *              an SI unit prefix and (optionally) the letter `B'. */

static int parse_size(union tvec_regval *rv, const struct tvec_regdef *rd,
                      struct tvec_state *tv)
{
  unsigned long sz;
  int rc;

  if (parse_szint(tv, &sz, ";", "size")) { rc = -1; goto end; }
  if (check_unsigned_range(sz, rd->arg.p, tv, "size")) { rc = -1; goto end; }
  if (tvec_flushtoeol(tv, 0)) { rc = -1; goto end; }
  rv->u = sz; rc = 0;
end:
  return (rc);
}

/* --- @dump_size@ --- *
 *
 * Arguments:   @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @unsigned style@ = output style (@TVSF_...@)
 *              @const struct gprintf_ops *gops@, @void *gp@ = format output
 *
 * Returns:     ---
 *
 * Use:         Dump a register value to the format output.
 *
 *              Size values are dumped with a unit specifier, with a unit
 *              prefox only if the size is an exact multiple of the relevant
 *              power of two.  Unless compact style is requested, the plain
 *              decimal and hex representations of the value are also
 *              printed.
 */

static void dump_size(const union tvec_regval *rv,
                      const struct tvec_regdef *rd,
                      unsigned style,
                      const struct gprintf_ops *gops, void *go)
{
  if (style&TVSF_RAW) gprintf(gops, go, "size:");
  format_size(gops, go, rv->u, style);
  if (!(style&(TVSF_COMPACT | TVSF_RAW))) {
    gprintf(gops, go, " ; = %lu", (unsigned long)rv->u);
    gprintf(gops, go, " = "); format_unsigned_hex(gops, go, rv->u);
    maybe_format_unsigned_char(gops, go, rv->u);
  }
}

/* Size type definitions. */
const struct tvec_regty tvty_size = {
  init_uint, trivial_release, eq_uint,
  tobuf_uint, frombuf_uint,
  parse_size, dump_size
};

/* --- @tvec_claimeq_size@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @unsigned long sz0, sz1@ = two sizes
 *              @const char *file@, @unsigned @lno@ = calling file and line
 *              @const char *expr@ = the expression to quote on failure
 *
 * Returns:     Nonzero if @sz0@ and @sz1@ are equal, otherwise zero.
 *
 * Use:         Check that values of @u0@ and @u1@ are equal.  As for
 *              @tvec_claim@ above, a test case is automatically begun and
 *              ended if none is already underway.  If the values are
 *              unequal, then @tvec_fail@ is called, quoting @expr@, and the
 *              mismatched values are dumped: @u0@ is printed as the output
 *              value and @u1@ is printed as the input reference.
 */

int tvec_claimeq_size(struct tvec_state *tv,
                      unsigned long sz0, unsigned long sz1,
                      const char *file, unsigned lno, const char *expr)
{
  tv->out[0].v.u = sz0; tv->in[0].v.u = sz1;
  return (tvec_claimeq(tv, &tvty_size, 0, file, lno, expr));
}

/*----- Floating-point type -----------------------------------------------*/

/* --- @int_float@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     ---
 *
 * Use:         Initialize a register value.
 *
 *              Floating-point values are initialized to zero.
 */

static void init_float(union tvec_regval *rv, const struct tvec_regdef *rd)
  { rv->f = 0.0; }

/* --- @eq_float@ --- *
 *
 * Arguments:   @const union tvec_regval *rv0, *rv1@ = register values
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Nonzero if the values are equal, zero if unequal
 *
 * Use:         Compare register values for equality.
 *
 *              Floating-point values may be considered equal if their
 *              absolute or relative difference is sufficiently small, as
 *              described in the register definition.
 */

static int eq_float(const union tvec_regval *rv0,
                    const union tvec_regval *rv1,
                    const struct tvec_regdef *rd)
  { return (eqish_floating_p(rv0->f, rv1->f, rd->arg.p)); }

/* --- @tobuf_float@ --- *
 *
 * Arguments:   @buf *b@ = buffer
 *              @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Zero on success, %$-1$% on failure.
 *
 * Use:         Serialize a register value to a buffer.
 *
 *              Floating-point values are serialized as little-endian
 *              IEEE 754 Binary64.
 */

static int tobuf_float(buf *b, const union tvec_regval *rv,
                     const struct tvec_regdef *rd)
  { return (buf_putf64l(b, rv->f)); }

/* --- @frombuf_float@ --- *
 *
 * Arguments:   @buf *b@ = buffer
 *              @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Zero on success, %$-1$% on failure.
 *
 * Use:         Deserialize a register value from a buffer.
 *
 *              Floating-point values are serialized as little-endian
 *              IEEE 754 Binary64.
 */

static int frombuf_float(buf *b, union tvec_regval *rv,
                       const struct tvec_regdef *rd)
  { return (buf_getf64l(b, &rv->f)); }

/* --- @parse_float@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @struct tvec_state *tv@ = test-vector state
 *
 * Returns:     Zero on success, %$-1$% on error.
 *
 * Use:         Parse a register value from an input file.
 *
 *              Floating-point values are either NaN (%|#nan|%, if supported
 *              by the platform); positive or negative infinity (%|#inf|%,
 *              %|+#inf|%, or %|#+inf|% (preferring the last), and %|-#inf|%
 *              or %|#-inf|% (preferring the latter), if supported by the
 *              platform); or a number in strtod(3) syntax.
 */

static int parse_float(union tvec_regval *rv, const struct tvec_regdef *rd,
                       struct tvec_state *tv)
{
  dstr d = DSTR_INIT;
  int rc;

  if (tvec_readword(tv, &d, 0, ";", "floating-point number"))
    { rc = -1; goto end; }
  if (parse_floating(&rv->f, 0, d.buf, rd->arg.p, tv))
    { rc = -1; goto end; }
  if (tvec_flushtoeol(tv, 0)) { rc = -1; goto end; }
  rc = 0;
end:
  dstr_destroy(&d);
  return (rc);
}

/* --- @dump_float@ --- *
 *
 * Arguments:   @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @unsigned style@ = output style (@TVSF_...@)
 *              @const struct gprintf_ops *gops@, @void *gp@ = format output
 *
 * Returns:     ---
 *
 * Use:         Dump a register value to the format output.
 *
 *              Floating-point values are dumped in decimal or as a special
 *              token beginning with `%|#|%'.  Some effort is taken to ensure
 *              that the output is sufficient to uniquely identify the
 *              original value, but, honestly, C makes this really hard.
 */

static void dump_float(const union tvec_regval *rv,
                       const struct tvec_regdef *rd,
                       unsigned style,
                       const struct gprintf_ops *gops, void *go)
{
  if (style&TVSF_RAW) gprintf(gops, go, "float:");
  format_floating(gops, go, rv->f);
}

/* Floating-point type definition. */
const struct tvec_regty tvty_float = {
  init_float, trivial_release, eq_float,
  tobuf_float, frombuf_float,
  parse_float, dump_float
};

/* Predefined floating-point ranges. */
const struct tvec_floatinfo
  tvflt_finite = { TVFF_EXACT, -DBL_MAX, DBL_MAX, 0.0 },
  tvflt_nonneg = { TVFF_EXACT, 0, DBL_MAX, 0.0 };

/* --- @tvec_claimeqish_float@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @double f0, f1@ = two floating-point numbers
 *              @unsigned f@ = flags (@TVFF_...@)
 *              @double delta@ = maximum tolerable difference
 *              @const char *file@, @unsigned @lno@ = calling file and line
 *              @const char *expr@ = the expression to quote on failure
 *
 * Returns:     Nonzero if @f0@ and @f1@ are sufficiently close, otherwise
 *              zero.
 *
 * Use:         Check that values of @f0@ and @f1@ are sufficiently close.
 *              As for @tvec_claim@ above, a test case is automatically begun
 *              and ended if none is already underway.  If the values are
 *              too far apart, then @tvec_fail@ is called, quoting @expr@,
 *              and the mismatched values are dumped: @f0@ is printed as the
 *              output value and @f1@ is printed as the input reference.
 *
 *              The details for the comparison are as follows.
 *
 *                * A NaN value matches any other NaN, and nothing else.
 *
 *                * An infinity matches another infinity of the same sign,
 *                  and nothing else.
 *
 *                * If @f&TVFF_EQMASK@ is @TVFF_EXACT@, then any
 *                  representable number matches only itself: in particular,
 *                  positive and negative zero are considered distinct.
 *                  (This allows tests to check that they land on the correct
 *                  side of branch cuts, for example.)
 *
 *                * If @f&TVFF_EQMASK@ is @TVFF_ABSDELTA@, then %$x$% matches
 *                  %$y$% when %$|x - y| < \delta$%.
 *
 *                * If @f&TVFF_EQMASK@ is @TVFF_RELDELTA@, then %$x$% matches
 *                  %$y$% when %$|1 - x/y| < \delta$%.  (Note that this
 *                  criterion is asymmetric.  Write %$x \approx_\delta y$%
 *                  if and only if %$|1 - x/y < \delta$%.  Then, for example,
 *                  if %$y/(1 + \delta) < x < y (1 - \delta)$%, then
 *                  %$x \approx_\delta y$%, but %$y \not\approx_\delta x$%.)
 */

int tvec_claimeqish_float(struct tvec_state *tv,
                          double f0, double f1, unsigned f, double delta,
                          const char *file, unsigned lno,
                          const char *expr)
{
  struct tvec_floatinfo fi;
  union tvec_misc arg;

  fi.f = f; fi.min = fi.max = 0.0; fi.delta = delta; arg.p = &fi;
  tv->out[0].v.f = f0; tv->in[0].v.f = f1;
  return (tvec_claimeq(tv, &tvty_float, &arg, file, lno, expr));
}

/* --- @tvec_claimeq_float@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @double f0, f1@ = two floating-point numbers
 *              @const char *file@, @unsigned @lno@ = calling file and line
 *              @const char *expr@ = the expression to quote on failure
 *
 * Returns:     Nonzero if @f0@ and @f1@ are identical, otherwise zero.
 *
 * Use:         Check that values of @f0@ and @f1@ are identical.  The
 *              function is exactly equivalent to @tvec_claimeqish_float@
 *              with @f == TVFF_EXACT@.
 */

int tvec_claimeq_float(struct tvec_state *tv,
                       double f0, double f1,
                       const char *file, unsigned lno,
                       const char *expr)
{
  return (tvec_claimeqish_float(tv, f0, f1, TVFF_EXACT, 0.0,
                                file, lno, expr));
}

/*----- Durations ---------------------------------------------------------*/

/* A duration is a floating-point number of seconds.  Initialization and
 * teardown, equality comparison, and serialization are as for floating-point
 * values.
 */

static const struct duration_unit {
  const char *unit;
  double scale;
  unsigned f;
#define DUF_PREFER 1u
} duration_units[] = {
  { "Ys",       1e+24,          0 },
  { "Zs",       1e+21,          0 },
  { "Es",       1e+18,          0 },
  { "Ps",       1e+15,          0 },
  { "Ts",       1e+12,          0 },
  { "Gs",       1e+9,           0 },
  { "Ms",       1e+6,           0 },
  { "ks",       1e+3,           0 },
  { "hs",       1e+2,           0 },
  { "das",      1e+1,           0 },

  { "yr",       31557600.0,     DUF_PREFER },
  { "y",        31557600.0,     0 },
  { "day",      86400.0,        DUF_PREFER },
  { "dy",       86400.0,        0 },
  { "d",        86400.0,        0 },
  { "hr",       3600.0,         DUF_PREFER },
  { "hour",     3600.0,         0 },
  { "h",        3600.0,         0 },
  { "min",      60.0,           DUF_PREFER },
  { "m",        60.0,           0 },

  { "s",        1.0,            DUF_PREFER },
  { "sec",      1.0,            0 },

  { "ds",       1e-1,           0 },
  { "cs",       1e-2,           0 },
  { "ms",       1e-3,           DUF_PREFER },
  { "µs",      1e-6,           DUF_PREFER },
  { "ns",       1e-9,           DUF_PREFER },
  { "ps",       1e-12,          DUF_PREFER },
  { "fs",       1e-15,          DUF_PREFER },
  { "as",       1e-18,          DUF_PREFER },
  { "zs",       1e-21,          DUF_PREFER },
  { "ys",       1e-24,          DUF_PREFER },

  { 0 }
};

/* --- @tvec_parsedurunit@ --- *
 *
 * Arguments:   @double *scale_out@ = where to leave the scale
 *              @const char **p_inout@ = input unit string, updated
 *
 * Returns:     Zero on success, %$-1$% on error.
 *
 * Use:         If @*p_inout@ begins with a unit string followed by the end
 *              of the string or some non-alphanumeric character, then store
 *              the corresponding scale factor in @*scale_out@, advance
 *              @*p_inout@ past the unit string, and return zero.  Otherwise,
 *              return %$-1$%.
 */

int tvec_parsedurunit(double *scale_out, const char **p_inout)
{
  const char *p = *p_inout, *q;
  const struct duration_unit *u;
  size_t n;

  while (ISSPACE(*p)) p++;
  for (q = p; *q && ISALNUM(*q); q++);
  n = q - p; if (!n) { *scale_out = 1.0; return (0); }

  for (u = duration_units; u->unit; u++)
    if (STRNCMP(p, ==, u->unit, n) && !u->unit[n])
      { *scale_out = u->scale; *p_inout = q; return (0); }
  return (-1);
}

/* --- @parse_duration@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @struct tvec_state *tv@ = test-vector state
 *
 * Returns:     Zero on success, %$-1$% on error.
 *
 * Use:         Parse a register value from an input file.
 *
 *              Duration values are finite nonnegative floating-point
 *              numbers in @strtod@ syntax, optionally followed by a unit .
 */

static int parse_duration(union tvec_regval *rv,
                          const struct tvec_regdef *rd,
                          struct tvec_state *tv)
{
  const struct duration_unit *u;
  const char *q;
  dstr d = DSTR_INIT;
  double t;
  int rc;

  if (tvec_readword(tv, &d, 0, ";", "duration")) { rc = -1; goto end; }
  if (parse_floating(&t, &q, d.buf,
                     rd->arg.p ? rd->arg.p : &tvflt_nonneg, tv))
    { rc = -1; goto end; }

  if (!*q) tvec_readword(tv, &d, &q, ";", 0);
  if (*q) {
    for (u = duration_units; u->unit; u++)
      if (STRCMP(q, ==, u->unit)) { t *= u->scale; goto found_unit; }
    rc = tvec_syntax(tv, *q, "end-of-line"); goto end;
  found_unit:;
  }

  if (tvec_flushtoeol(tv, 0)) { rc = -1; goto end; }
  rv->f = t; rc = 0;
end:
  dstr_destroy(&d);
  return (rc);
}

/* --- @dump_duration@ --- *
 *
 * Arguments:   @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @unsigned style@ = output style (@TVSF_...@)
 *              @const struct gprintf_ops *gops@, @void *gp@ = format output
 *
 * Returns:     ---
 *
 * Use:         Dump a register value to the format output.
 *
 *              Durations are dumped as a human-palatable scaled value with
 *              unit, and, if compact style is not requested, as a raw number
 *              of seconds at full precision as a comment.
 */

static void dump_duration(const union tvec_regval *rv,
                          const struct tvec_regdef *rd,
                          unsigned style,
                          const struct gprintf_ops *gops, void *go)
{
  const struct duration_unit *u;
  double t = rv->f;

  if (style&TVSF_RAW) {
    gprintf(gops, go, "duration:");
    format_floating(gops, go, rv->f);
    gprintf(gops, go, "s");
  } else {
    if (!t) u = 0;
    else {
      for (u = duration_units; u->scale > t && u[1].unit; u++);
      t /= u->scale;
    }
    gprintf(gops, go, "%.4g %s", t, u ? u->unit : "s");

    if (!(style&TVSF_COMPACT)) {
      gprintf(gops, go, "; = ");
      format_floating(gops, go, rv->f);
      gprintf(gops, go, " s");
    }
  }
}

/* Duration type definition. */
const struct tvec_regty tvty_duration = {
  init_float, trivial_release, eq_float,
  tobuf_float, frombuf_float,
  parse_duration, dump_duration
};

/* --- @tvec_claimeqish_duration@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @double to, t1@ = two durations
 *              @unsigned f@ = flags (@TVFF_...@)
 *              @double delta@ = maximum tolerable difference
 *              @const char *file@, @unsigned @lno@ = calling file and line
 *              @const char *expr@ = the expression to quote on failure
 *
 * Returns:     Nonzero if @t0@ and @t1@ are sufficiently close, otherwise
 *              zero.
 *
 * Use:         Check that values of @t0@ and @t1@ are sufficiently close.
 *              This is essentially the same as @tvec_claimeqish_float@, only
 *              it dumps the values as durations on a mismatch.
 */

int tvec_claimeqish_duration(struct tvec_state *tv,
                             double t0, double t1, unsigned f, double delta,
                             const char *file, unsigned lno,
                             const char *expr)
{
  struct tvec_floatinfo fi;
  union tvec_misc arg;

  fi.f = f; fi.min = fi.max = 0.0; fi.delta = delta; arg.p = &fi;
  tv->out[0].v.f = t0; tv->in[0].v.f = t1;
  return (tvec_claimeq(tv, &tvty_duration, &arg, file, lno, expr));
}

/* --- @tvec_claimeq_duration@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @double t0, t1@ = two durations
 *              @const char *file@, @unsigned @lno@ = calling file and line
 *              @const char *expr@ = the expression to quote on failure
 *
 * Returns:     Nonzero if @t0@ and @t1@ are identical, otherwise zero.
 *
 * Use:         Check that values of @t0@ and @t1@ are identical.  The
 *              function is exactly equivalent to @tvec_claimeqish_duration@
 *              with @f == TVFF_EXACT@.
 */

int tvec_claimeq_duration(struct tvec_state *tv,
                          double t0, double t1,
                          const char *file, unsigned lno,
                          const char *expr)
{
  return (tvec_claimeqish_duration(tv, t0, t1, TVFF_EXACT, 0.0,
                                   file, lno, expr));
}

/*----- Enumerations ------------------------------------------------------*/

/* --- @init_tenum@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     ---
 *
 * Use:         Initialize a register value.
 *
 *              Integer and floating-point enumeration values are initialized
 *              as their underlying representations.  Pointer enumerations
 *              are initialized to %|#nil|%.
 */

#define init_ienum init_int
#define init_uenum init_uint
#define init_fenum init_float

static void init_penum(union tvec_regval *rv, const struct tvec_regdef *rd)
  { rv->p = 0; }

/* --- @eq_tenum@ --- *
 *
 * Arguments:   @const union tvec_regval *rv0, *rv1@ = register values
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Nonzero if the values are equal, zero if unequal
 *
 * Use:         Compare register values for equality.
 *
 *              Integer and floating-point enumeration values are compared as
 *              their underlying representations; in particular, floating-
 *              point enumerations may compare equal if their absolute or
 *              relative difference is sufficiently small.  Pointer
 *              enumerations are compared as pointers.
 */

#define eq_ienum eq_int
#define eq_uenum eq_uint

static int eq_fenum(const union tvec_regval *rv0,
                    const union tvec_regval *rv1,
                    const struct tvec_regdef *rd)
{
  const struct tvec_fenuminfo *ei = rd->arg.p;
  return (eqish_floating_p(rv0->f, rv1->f, ei->fi));
}

static int eq_penum(const union tvec_regval *rv0,
                    const union tvec_regval *rv1,
                    const struct tvec_regdef *rd)
  { return (rv0->p == rv1->p); }

/* --- @tobuf_tenum@ --- *
 *
 * Arguments:   @buf *b@ = buffer
 *              @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Zero on success, %$-1$% on failure.
 *
 * Use:         Serialize a register value to a buffer.
 *
 *              Integer and floating-point enumeration values are serialized
 *              as their underlying representations.  Pointer enumerations
 *              are serialized as the signed integer index into the
 *              association table; %|#nil|% serializes as %$-1$%, and
 *              unrecognized pointers cause failure.
 */

#define tobuf_ienum tobuf_int
#define tobuf_uenum tobuf_uint
#define tobuf_fenum tobuf_float

static int tobuf_penum(buf *b, const union tvec_regval *rv,
                       const struct tvec_regdef *rd)
{
  const struct tvec_penuminfo *pei = rd->arg.p;
  const struct tvec_passoc *pa;
  long i;

  for (pa = pei->av, i = 0; pa->tag; pa++, i++)
    if (pa->p == rv->p) goto found;
  if (!rv->p) i = -1;
  else return (-1);
found:
  return (signed_to_buf(b, i));
}

/* --- @frombuf_tenum@ --- *
 *
 * Arguments:   @buf *b@ = buffer
 *              @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Zero on success, %$-1$% on failure.
 *
 * Use:         Deserialize a register value from a buffer.
 *
 *              Integer and floating-point enumeration values are serialized
 *              as their underlying representations.  Pointer enumerations
 *              are serialized as the signed integer index into the
 *              association table; %|#nil|% serializes as %$-1$%; out-of-
 *              range indices cause failure.
 */

#define frombuf_ienum frombuf_int
#define frombuf_uenum frombuf_uint
#define frombuf_fenum frombuf_float
static int frombuf_penum(buf *b, union tvec_regval *rv,
                        const struct tvec_regdef *rd)
{
  const struct tvec_penuminfo *pei = rd->arg.p;
  const struct tvec_passoc *pa;
  long i, n;

  for (pa = pei->av, n = 0; pa->tag; pa++, n++);
  if (signed_from_buf(b, &i)) return (-1);
  if (0 <= i && i < n) rv->p = (/*unconst*/ void *)pei->av[i].p;
  else if (i == -1) rv->p = 0;
  else { buf_break(b); return (-1); }
  return (0);
}

/* --- @parse_tenum@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @struct tvec_state *tv@ = test-vector state
 *
 * Returns:     Zero on success, %$-1$% on error.
 *
 * Use:         Parse a register value from an input file.
 *
 *              An enumerated value may be given by name or as a literal
 *              value.  For enumerations based on numeric types, the literal
 *              values can be written in the same syntax as the underlying
 *              values.  For enumerations based on pointers, the only
 *              permitted literal is %|#nil|%, which denotes a null pointer.
 */

#define DEFPARSE_ENUM(tag_, ty, slot)                                   \
  static int parse_##slot##enum(union tvec_regval *rv,                  \
                                const struct tvec_regdef *rd,           \
                                struct tvec_state *tv)                  \
  {                                                                     \
    const struct tvec_##slot##enuminfo *ei = rd->arg.p;                 \
    const struct tvec_##slot##assoc *a;                                 \
    dstr d = DSTR_INIT;                                                 \
    int rc;                                                             \
                                                                        \
    if (tvec_readword(tv, &d, 0,                                        \
                      ";", "%s tag or " LITSTR_##tag_, ei->name))       \
      { rc = -1; goto end; }                                            \
    for (a = ei->av; a->tag; a++)                                       \
      if (STRCMP(a->tag, ==, d.buf)) { FOUND_##tag_ goto done; }        \
    MISSING_##tag_                                                      \
    done:                                                               \
    if (tvec_flushtoeol(tv, 0)) { rc = -1; goto end; }                  \
    rc = 0;                                                             \
  end:                                                                  \
    dstr_destroy(&d);                                                   \
    return (rc);                                                        \
  }

#define LITSTR_INT      "literal signed integer"
#define FOUND_INT       rv->i = a->i;
#define MISSING_INT     if (parse_signed(&rv->i, d.buf, ei->ir, tv))    \
                          { rc = -1; goto end; }

#define LITSTR_UINT     "literal unsigned integer"
#define FOUND_UINT      rv->u = a->u;
#define MISSING_UINT    if (parse_unsigned(&rv->u, d.buf, ei->ur, tv))  \
                          { rc = -1; goto end; }

#define LITSTR_FLT      "literal floating-point number, "               \
                          "`#-inf', `#+inf', or `#nan'"
#define FOUND_FLT       rv->f = a->f;
#define MISSING_FLT     if (parse_floating(&rv->f, 0, d.buf, ei->fi, tv)) \
                          { rc = -1; goto end; }

#define LITSTR_PTR      "`#nil'"
#define FOUND_PTR       rv->p = (/*unconst*/ void *)a->p;
#define MISSING_PTR     if (STRCMP(d.buf, ==, "#nil"))                  \
                          rv->p = 0;                                    \
                        else {                                          \
                          tvec_error(tv, "unknown `%s' value `%s'",     \
                                     ei->name, d.buf);                  \
                          rc = -1; goto end;                            \
                        }

TVEC_MISCSLOTS(DEFPARSE_ENUM)

#undef LITSTR_INT
#undef FOUND_INT
#undef MISSING_INT

#undef LITSTR_UINT
#undef FOUND_UINT
#undef MISSING_UINT

#undef LITSTR_FLT
#undef FOUND_FLT
#undef MISSING_FLT

#undef LITSTR_PTR
#undef FOUND_PTR
#undef MISSING_PTR

#undef DEFPARSE_ENUM

/* --- @dump_tenum@ --- *
 *
 * Arguments:   @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @unsigned style@ = output style (@TVSF_...@)
 *              @const struct gprintf_ops *gops@, @void *gp@ = format output
 *
 * Returns:     ---
 *
 * Use:         Dump a register value to the format output.
 *
 *              Enumeration values are dumped as their symbolic names, if
 *              possible, with the underlying values provided as a comment
 *              unless compact output is requested, as for the underlying
 *              representation.  A null pointer is printed as %|#nil|%;
 *              non-null pointers are printed as %|#<TYPE PTR>|%, with the
 *              enumeration TYPE and the raw pointer PTR printed with the
 *              system's %|%p|% format specifier.
 */


#define DEFDUMP_ENUM(tag_, ty, slot)                                    \
  static void dump_##slot##enum(const union tvec_regval *rv,            \
                                const struct tvec_regdef *rd,           \
                                unsigned style,                         \
                                const struct gprintf_ops *gops, void *go) \
  {                                                                     \
    const struct tvec_##slot##enuminfo *ei = rd->arg.p;                 \
    const struct tvec_##slot##assoc *a;                                 \
                                                                        \
    if (style&TVSF_RAW) gprintf(gops, go, #slot "enum/%s:", ei->name);  \
    for (a = ei->av; a->tag; a++)                                       \
      if (rv->slot == a->slot) {                                        \
        gprintf(gops, go, "%s", a->tag);                                \
        if (style&TVSF_COMPACT) return;                                 \
        gprintf(gops, go, " ; = "); break;                              \
      }                                                                 \
                                                                        \
    PRINTRAW_##tag_                                                     \
  }

#define MAYBE_PRINT_EXTRA                                               \
        if (style&TVSF_COMPACT) /* nothing to do */;                    \
        else if (!a->tag) { gprintf(gops, go, " ; = "); goto _extra; }  \
        else if (1) { gprintf(gops, go, " = "); goto _extra; }          \
        else _extra:

#define PRINTRAW_INT    gprintf(gops, go, "%ld", rv->i);                \
                        MAYBE_PRINT_EXTRA {                             \
                          format_signed_hex(gops, go, rv->i);           \
                          maybe_format_signed_char(gops, go, rv->i);    \
                        }

#define PRINTRAW_UINT   gprintf(gops, go, "%lu", rv->u);                \
                        MAYBE_PRINT_EXTRA {                             \
                          format_unsigned_hex(gops, go, rv->u);         \
                          maybe_format_unsigned_char(gops, go, rv->u);  \
                        }

#define PRINTRAW_FLT    format_floating(gops, go, rv->f);

#define PRINTRAW_PTR    if (!rv->p) gprintf(gops, go, "#nil");          \
                        else gprintf(gops, go, "#<%s %p>", ei->name, rv->p);

TVEC_MISCSLOTS(DEFDUMP_ENUM)

#undef PRINTRAW_INT
#undef PRINTRAW_UINT
#undef PRINTRAW_FLT
#undef PRINTRAW_PTR

#undef MAYBE_PRINT_EXTRA
#undef DEFDUMP_ENUM

/* Enumeration type definitions. */
#define DEFTY_ENUM(tag, ty, slot)                                       \
  const struct tvec_regty tvty_##slot##enum = {                         \
    init_##slot##enum, trivial_release, eq_##slot##enum,                \
    tobuf_##slot##enum, frombuf_##slot##enum,                           \
    parse_##slot##enum, dump_##slot##enum                               \
  };
TVEC_MISCSLOTS(DEFTY_ENUM)
#undef DEFTY_ENUM

/* Predefined enumeration types. */
static const struct tvec_iassoc bool_assoc[] = {
  { "nil",              0 },
  { "false",            0 },
  { "f",                0 },
  { "no",               0 },
  { "n",                0 },
  { "off",              0 },

  { "t",                1 },
  { "true",             1 },
  { "yes",              1 },
  { "y",                1 },
  { "on",               1 },

  TVEC_ENDENUM
};

const struct tvec_ienuminfo tvenum_bool =
  { "bool", bool_assoc, &tvrange_int };

static const struct tvec_iassoc cmp_assoc[] = {
  { "<",                -1 },
  { "less",             -1 },
  { "lt",               -1 },

  { "=",                 0 },
  { "equal",             0 },
  { "eq",                0 },

  { ">",                +1 },
  { "greater",          +1 },
  { "gt",               +1 },

  TVEC_ENDENUM
};

const struct tvec_ienuminfo tvenum_cmp =
  { "cmp", cmp_assoc, &tvrange_int };

/* --- @tvec_claimeq_tenum@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @const struct tvec_typeenuminfo *ei@ = enumeration type info
 *              @ty t0, t1@ = two values
 *              @const char *file@, @unsigned @lno@ = calling file and line
 *              @const char *expr@ = the expression to quote on failure
 *
 * Returns:     Nonzero if @t0@ and @t1@ are equal, otherwise zero.
 *
 * Use:         Check that values of @t0@ and @t1@ are equal.  As for
 *              @tvec_claim@ above, a test case is automatically begun and
 *              ended if none is already underway.  If the values are
 *              unequal, then @tvec_fail@ is called, quoting @expr@, and the
 *              mismatched values are dumped: @t0@ is printed as the output
 *              value and @t1@ is printed as the input reference.
 */

#define DEFCLAIM(tag, ty, slot)                                         \
        int tvec_claimeq_##slot##enum                                   \
          (struct tvec_state *tv,                                       \
           const struct tvec_##slot##enuminfo *ei, ty e0, ty e1,        \
           const char *file, unsigned lno, const char *expr)            \
        {                                                               \
          union tvec_misc arg;                                          \
                                                                        \
          arg.p = ei;                                                   \
          tv->out[0].v.slot = GET_##tag(e0);                            \
          tv->in[0].v.slot = GET_##tag(e1);                             \
          return (tvec_claimeq(tv, &tvty_##slot##enum, &arg,            \
                               file, lno, expr));                       \
        }
#define GET_INT(e) (e)
#define GET_UINT(e) (e)
#define GET_FLT(e) (e)
#define GET_PTR(e) ((/*unconst*/ void *)(e))
TVEC_MISCSLOTS(DEFCLAIM)
#undef DEFCLAIM
#undef GET_INT
#undef GET_UINT
#undef GET_FLT
#undef GET_PTR

/*----- Flag types --------------------------------------------------------*/

/* Flag types are initialized, compared, and serialized as unsigned
 * integers.
 */

/* --- @parse_flags@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @struct tvec_state *tv@ = test-vector state
 *
 * Returns:     Zero on success, %$-1$% on error.
 *
 * Use:         Parse a register value from an input file.
 *
 *              The input syntax is a sequence of items separated by `|'
 *              signs.  Each item may be the symbolic name of a field value,
 *              or a literal unsigned integer.  The masks associated with the
 *              given symbolic names must be disjoint.  The resulting
 *              numerical value is simply the bitwise OR of the given values.
 */

static int parse_flags(union tvec_regval *rv, const struct tvec_regdef *rd,
                       struct tvec_state *tv)
{
  const struct tvec_flaginfo *fi = rd->arg.p;
  const struct tvec_flag *f;
  unsigned long m = 0, v = 0, t;
  dstr d = DSTR_INIT;
  int ch, rc;

  for (;;) {

    /* Read the next item. */
    DRESET(&d);
    if (tvec_readword(tv, &d, 0, "|;", "%s flag name or integer", fi->name))
      { rc = -1; goto end; }

    /* Try to find a matching entry in the table. */
    for (f = fi->fv; f->tag; f++)
      if (STRCMP(f->tag, ==, d.buf)) {
        if (m&f->m)
          { tvec_error(tv, "colliding flag setting"); rc = -1; goto end; }
        else
          { m |= f->m; v |= f->v; goto next; }
      }

    /* Otherwise, try to parse it as a raw integer. */
    if (parse_unsigned(&t, d.buf, fi->range, tv))
      { rc = -1; goto end; }
    v |= t;

  next:
    /* Advance to the next token.  If it's a separator then consume it, and
     * go round again.  Otherwise we stop here.
     */
    if (tvec_nexttoken(tv)) break;
    ch = getc(tv->fp);
      if (ch != '|') { tvec_syntax(tv, ch, "`|'"); rc = -1; goto end; }
      if (tvec_nexttoken(tv)) {
        tvec_syntax(tv, '\n', "%s flag name or integer", fi->name);
        rc = -1; goto end;
      }
  }

  /* Done. */
  rv->u = v; rc = 0;
end:
  dstr_destroy(&d);
  return (rc);
}

/* --- @dump_flags@ --- *
 *
 * Arguments:   @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @unsigned style@ = output style (@TVSF_...@)
 *              @const struct gprintf_ops *gops@, @void *gp@ = format output
 *
 * Returns:     ---
 *
 * Use:         Dump a register value to the format output.
 *
 *              The table of symbolic names and their associated values and
 *              masks is repeatedly scanned, in order, to find disjoint
 *              matches -- i.e., entries whose value matches the target value
 *              in the bit positions indicated by the mask, and whose mask
 *              doesn't overlap with any previously found matches; the names
 *              are then output, separated by `|'.  Any remaining nonzero
 *              bits not covered by any of the matching masks are output as a
 *              single literal integer, in hex.
 *
 *              Unless compact output is requested, or no symbolic names were
 *              found, the raw numeric value is also printed in hex, as a
 *              comment.
 */

static void dump_flags(const union tvec_regval *rv,
                       const struct tvec_regdef *rd,
                       unsigned style,
                       const struct gprintf_ops *gops, void *go)
{
  const struct tvec_flaginfo *fi = rd->arg.p;
  const struct tvec_flag *f;
  unsigned long m = ~0ul, v = rv->u;
  const char *sep;

  if (style&TVSF_RAW) gprintf(gops, go, "flags/%s:", fi->name);

  for (f = fi->fv, sep = ""; f->tag; f++)
    if ((m&f->m) && (v&f->m) == f->v) {
      gprintf(gops, go, "%s%s", sep, f->tag); m &= ~f->m;
      sep = style&TVSF_COMPACT ? "|" : " | ";
    }

  if (v&m) gprintf(gops, go, "%s0x%0*lx", sep, hex_width(v), v&m);

  if (m != ~0ul && !(style&(TVSF_COMPACT | TVSF_RAW)))
    gprintf(gops, go, " ; = 0x%0*lx", hex_width(rv->u), rv->u);
}

/* Flags type definition. */
const struct tvec_regty tvty_flags = {
  init_uint, trivial_release, eq_uint,
  tobuf_uint, frombuf_uint,
  parse_flags, dump_flags
};

/* --- @tvec_claimeq_flags@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @const struct tvec_flaginfo *fi@ = flags type info
 *              @unsigned long f0, f1@ = two values
 *              @const char *file@, @unsigned @lno@ = calling file and line
 *              @const char *expr@ = the expression to quote on failure
 *
 * Returns:     Nonzero if @f0@ and @f1@ are equal, otherwise zero.
 *
 * Use:         Check that values of @f0@ and @f1@ are equal.  As for
 *              @tvec_claim@ above, a test case is automatically begun and
 *              ended if none is already underway.  If the values are
 *              unequal, then @tvec_fail@ is called, quoting @expr@, and the
 *              mismatched values are dumped: @f0@ is printed as the output
 *              value and @f1@ is printed as the input reference.
 */

int tvec_claimeq_flags(struct tvec_state *tv,
                       const struct tvec_flaginfo *fi,
                       unsigned long f0, unsigned long f1,
                       const char *file, unsigned lno, const char *expr)
{
  union tvec_misc arg;

  arg.p = fi; tv->out[0].v.u = f0; tv->in[0].v.u = f1;
  return (tvec_claimeq(tv, &tvty_flags, &arg, file, lno, expr));
}

/*----- Characters --------------------------------------------------------*/

/* Character values are initialized and compared as signed integers. */

/* --- @tobuf_char@ --- *
 *
 * Arguments:   @buf *b@ = buffer
 *              @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Zero on success, %$-1$% on failure.
 *
 * Use:         Serialize a register value to a buffer.
 *
 *              Character values are serialized as little-endian 32-bit
 *              unsigned integers, with %|EOF|% serialized as all-bits-set.
 */

static int tobuf_char(buf *b, const union tvec_regval *rv,
                      const struct tvec_regdef *rd)
{
  uint32 u;

  if (0 <= rv->i && rv->i <= UCHAR_MAX) u = rv->i;
  else if (rv->i == EOF) u = MASK32;
  else { buf_break(b); return (-1); }
  return (buf_putu32l(b, u));
}

/* --- @frombuf_char@ --- *
 *
 * Arguments:   @buf *b@ = buffer
 *              @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Zero on success, %$-1$% on failure.
 *
 * Use:         Deserialize a register value from a buffer.
 *
 *              Character values are serialized as little-endian 32-bit
 *              unsigned integers, with %|EOF|% serialized as all-bits-set.
 */

static int frombuf_char(buf *b, union tvec_regval *rv,
                        const struct tvec_regdef *rd)
{
  uint32 u;

  if (buf_getu32l(b, &u)) return (-1);
  if (0 <= u && u <= UCHAR_MAX) rv->i = u;
  else if (u == MASK32) rv->i = EOF;
  else { buf_break(b); return (-1); }
  return (0);
}

/* --- @parse_char@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @struct tvec_state *tv@ = test-vector state
 *
 * Returns:     Zero on success, %$-1$% on error.
 *
 * Use:         Parse a register value from an input file.
 *
 *              A character value can be given by symbolic name, with a
 *              leading `%|#|%'; or a character or `%|\|%'-escape sequence,
 *              optionally in single quotes.
 *
 *              The following escape sequences and character names are
 *              recognized.
 *
 *              * `%|#eof|%' is the special end-of-file marker.
 *
 *              * `%|#nul|%' is the NUL character, sometimes used to
 *                terminate strings.
 *
 *              * `%|bell|%', `%|bel|%', `%|ding|%', or `%|\a|%' is the BEL
 *                character used to ring the terminal bell (or do some other
 *                thing to attract the user's attention).
 *
 *              * %|#backspace|%, %|#bs|%, or %|\b|% is the backspace
 *                character, used to move the cursor backwords by one cell.
 *
 *              * %|#escape|% %|#esc|%, or%|\e|% is the escape character,
 *                used to introduce special terminal commands.
 *
 *              * %|#formfeed|%, %|#ff|%, or %|\f|% is the formfeed
 *                character, used to separate pages of text.
 *
 *              * %|#newline|%, %|#linefeed|%, %|#lf|%, %|#nl|%, or %|\n|% is
 *                the newline character, used to terminate lines of text or
 *                advance the cursor to the next line (perhaps without
 *                returning it to the start of the line).
 *
 *              * %|#return|%, %|#carriage-return|%, %|#cr|%, or %|\r|% is
 *                the carriage-return character, used to return the cursor to
 *                the start of the line.
 *
 *              * %|#tab|%, %|#horizontal-tab|%, %|#ht|%, or %|\t|% is the
 *                tab character, used to advance the cursor to the next tab
 *                stop on the current line.
 *
 *              * %|#vertical-tab|%, %|#vt|%, %|\v|% is the vertical tab
 *                character.
 *
 *              * %|#space|%, %|#spc|% is the space character.
 *
 *              * %|#delete|%, %|#del|% is the delete character, used to
 *                erase the most recent character.
 *
 *              * %|\'|% is the single-quote character.
 *
 *              * %|\\|% is the backslash character.
 *
 *              * %|\"|% is the double-quote character.
 *
 *              * %|\NNN|% or %|\{NNN}|% is the character with code NNN in
 *                octal.  The NNN may be up to three digits long.
 *
 *              * %|\xNN|% or %|\x{NN}|% is the character with code NNN in
 *                hexadecimal.
 */

static int parse_char(union tvec_regval *rv, const struct tvec_regdef *rd,
                      struct tvec_state *tv)
{
  dstr d = DSTR_INIT;
  int ch, rc;
  unsigned f = 0;
#define f_quote 1u

  /* Inspect the character to see what we're up against. */
  ch = getc(tv->fp);

  if (ch == '#') {
    /* It looks like a special token.  Push the `%|#|%' back and fetch the
     * whole word.  If there's just the `%|#|%' after all, then treat it as
     * literal.
     */

    ungetc(ch, tv->fp);
    if (tvec_readword(tv, &d, 0, ";", "character name"))
      { rc = -1; goto end; }
    if (STRCMP(d.buf, !=, "#")) {
      if (read_charname(&ch, d.buf, RCF_EOFOK)) {
        rc = tvec_error(tv, "unknown character name `%s'", d.buf);
        goto end;
      }
      if (tvec_flushtoeol(tv, 0)) { rc = -1; goto end; }
      rv->i = ch; rc = 0; goto end;
    }
  }

  /* If this is a single quote then we expect to see a matching one later,
   * and we should process backslash escapes.  Get the next character and see
   * what happens.
   */
  if (ch == '\'') { f |= f_quote; ch = getc(tv->fp); }

  /* Main character dispatch. */
  switch (ch) {

    case ';':
      /* Unquoted, semicolon begins a comment. */
      if (!(f&f_quote)) { rc = tvec_syntax(tv, ch, "character"); goto end; }
      else goto plain;

    case '\n':
      /* A newline.  If we saw a single quote, then treat that as literal.
       * Otherwise this is an error.
       */
      if (!(f&f_quote)) goto nochar;
      else { f &= ~f_quote; ungetc(ch, tv->fp); ch = '\''; goto plain; }

    case EOF:
      /* End-of-file.  Similar to newline, but with slightly different
       * effects on the parse state.
       */
      if (!(f&f_quote)) goto nochar;
      else { f &= ~f_quote; ch = '\''; goto plain; }

    case '\'': nochar:
      /* A single quote.  This must be the second of a pair, and there should
       * have been a character or escape sequence between them.
       */
      rc = tvec_syntax(tv, ch, "character"); goto end;

    case '\\':
      /* A backslash.  Read a character escape. */
      if (read_charesc(&ch, tv)) return (-1);

    default: plain:
      /* Anything else.  Treat as literal. */
      rv->i = ch; break;
  }

  /* If we saw an opening quote, then expect the closing quote. */
  if (f&f_quote) {
    ch = getc(tv->fp);
    if (ch != '\'') { rc = tvec_syntax(tv, ch, "`''"); goto end; }
  }

  /* Done. */
  if (tvec_flushtoeol(tv, 0)) { rc = -1; goto end; }
  rc = 0;
end:
  dstr_destroy(&d);
  return (rc);

#undef f_quote
}

/* --- @dump_char@ --- *
 *
 * Arguments:   @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @unsigned style@ = output style (@TVSF_...@)
 *              @const struct gprintf_ops *gops@, @void *gp@ = format output
 *
 * Returns:     ---
 *
 * Use:         Dump a register value to the format output.
 *
 *              Character values are dumped as their symbolic names, if any,
 *              or as a character or escape sequence within single quotes
 *              (which may be omitted in compact style).  If compact output
 *              is not requested, then the single-quoted representation (for
 *              characters dumped as symbolic names) and integer code in
 *              decimal and hex are printed as a comment.
 */

static void dump_char(const union tvec_regval *rv,
                      const struct tvec_regdef *rd,
                      unsigned style,
                      const struct gprintf_ops *gops, void *go)
{
  const char *p;
  unsigned f = 0;
#define f_semi 1u

  if (style&TVSF_RAW) {
    /* Print the raw character unconditionally in single quotes. */

    gprintf(gops, go, "char:'");
    format_char(gops, go, rv->i);
    gprintf(gops, go, "'");
  } else {
    /* Print ina pleasant human-readable way. */

    /* Print a character name if we can find one. */
    p = find_charname(rv->i, (style&TVSF_COMPACT) ? CTF_SHORT : CTF_PREFER);
    if (p) {
      gprintf(gops, go, "%s", p);
      if (style&TVSF_COMPACT) return;
      else { gprintf(gops, go, " ;"); f |= f_semi; }
    }

    /* If the character isn't @EOF@ then print it as a single-quoted thing.
     * In compact style, see if we can omit the quotes.
     */
    if (rv->i >= 0) {
      if (f&f_semi) gprintf(gops, go, " = ");
      switch (rv->i) {
        case ' ': case '\\': case '\'': quote:
          format_char(gops, go, rv->i);
          break;
        default:
          if (!(style&TVSF_COMPACT) || !isprint(rv->i)) goto quote;
          gprintf(gops, go, "%c", (int)rv->i);
          return;
      }
    }

    /* And the character code as an integer. */
    if (!(style&TVSF_COMPACT)) {
      if (!(f&f_semi)) gprintf(gops, go, " ;");
      gprintf(gops, go, " = %ld = ", rv->i);
      format_signed_hex(gops, go, rv->i);
    }
  }

#undef f_semi
}

/* Character type definition. */
const struct tvec_regty tvty_char = {
  init_int, trivial_release, eq_int,
  tobuf_char, frombuf_char,
  parse_char, dump_char
};

/* --- @tvec_claimeq_char@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @int ch0, ch1@ = two character codes
 *              @const char *file@, @unsigned @lno@ = calling file and line
 *              @const char *expr@ = the expression to quote on failure
 *
 * Returns:     Nonzero if @ch0@ and @ch1@ are equal, otherwise zero.
 *
 * Use:         Check that values of @ch0@ and @ch1@ are equal.  As for
 *              @tvec_claim@ above, a test case is automatically begun and
 *              ended if none is already underway.  If the values are
 *              unequal, then @tvec_fail@ is called, quoting @expr@, and the
 *              mismatched values are dumped: @ch0@ is printed as the output
 *              value and @ch1@ is printed as the input reference.
 */

int tvec_claimeq_char(struct tvec_state *tv, int c0, int c1,
                      const char *file, unsigned lno, const char *expr)
{
  tv->out[0].v.i = c0; tv->in[0].v.i = c1;
  return (tvec_claimeq(tv, &tvty_char, 0, file, lno, expr));
}

/*----- Text and byte strings ---------------------------------------------*/

/* --- @init_text@, @init_bytes@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     ---
 *
 * Use:         Initialize a register value.
 *
 *              Text and binary string values are initialized with a null
 *              pointer and zero length.
 */

static void init_text(union tvec_regval *rv, const struct tvec_regdef *rd)
  { rv->text.p = 0; rv->text.sz = 0; }

static void init_bytes(union tvec_regval *rv, const struct tvec_regdef *rd)
  { rv->bytes.p = 0; rv->bytes.sz = 0; }

/* --- @release_string@, @release_bytes@ --- *
 *
 * Arguments:   @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     ---
 *
 * Use:         Release resources held by a register value.
 *
 *              Text and binary string buffers are freed.
 */

static void release_text(union tvec_regval *rv,
                         const struct tvec_regdef *rd)
  { xfree(rv->text.p); }

static void release_bytes(union tvec_regval *rv,
                          const struct tvec_regdef *rd)
  { xfree(rv->bytes.p); }

/* --- @eq_text@, @eq_bytes@ --- *
 *
 * Arguments:   @const union tvec_regval *rv0, *rv1@ = register values
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Nonzero if the values are equal, zero if unequal
 *
 * Use:         Compare register values for equality.
 */

static int eq_text(const union tvec_regval *rv0,
                   const union tvec_regval *rv1,
                   const struct tvec_regdef *rd)
{
  return (rv0->text.sz == rv1->text.sz &&
          (!rv0->text.sz ||
           MEMCMP(rv0->text.p, ==, rv1->text.p, rv1->text.sz)));
}

static int eq_bytes(const union tvec_regval *rv0,
                    const union tvec_regval *rv1,
                    const struct tvec_regdef *rd)
{
  return (rv0->bytes.sz == rv1->bytes.sz &&
          (!rv0->bytes.sz ||
           MEMCMP(rv0->bytes.p, ==, rv1->bytes.p, rv1->bytes.sz)));
}

/* --- @tobuf_text@, @tobuf_bytes@ --- *
 *
 * Arguments:   @buf *b@ = buffer
 *              @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Zero on success, %$-1$% on failure.
 *
 * Use:         Serialize a register value to a buffer.
 *
 *              Text and binary string values are serialized as a little-
 *              endian 64-bit length %$n$% in bytes followed by %$n$% bytes
 *              of string data.
 */

static int tobuf_text(buf *b, const union tvec_regval *rv,
                      const struct tvec_regdef *rd)
  { return (buf_putmem64l(b, rv->text.p, rv->text.sz)); }

static int tobuf_bytes(buf *b, const union tvec_regval *rv,
                       const struct tvec_regdef *rd)
  { return (buf_putmem64l(b, rv->bytes.p, rv->bytes.sz)); }

/* --- @frombuf_text@, @frombuf_bytes@ --- *
 *
 * Arguments:   @buf *b@ = buffer
 *              @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Zero on success, %$-1$% on failure.
 *
 * Use:         Deserialize a register value from a buffer.
 *
 *              Text and binary string values are serialized as a little-
 *              endian 64-bit length %$n$% in bytes followed by %$n$% bytes
 *              of string data.
 */

static int frombuf_text(buf *b, union tvec_regval *rv,
                        const struct tvec_regdef *rd)
{
  const void *p;
  size_t sz;

  p = buf_getmem64l(b, &sz); if (!p) return (-1);
  tvec_alloctext(rv, sz); memcpy(rv->text.p, p, sz); rv->text.p[sz] = 0;
  return (0);
}

static int frombuf_bytes(buf *b, union tvec_regval *rv,
                         const struct tvec_regdef *rd)
{
  const void *p;
  size_t sz;

  p = buf_getmem64l(b, &sz); if (!p) return (-1);
  tvec_allocbytes(rv, sz); memcpy(rv->bytes.p, p, sz);
  return (0);
}

/* --- @check_string_length@ --- *
 *
 * Arguments:   @size_t sz@ = found string length
 *              @const struct tvec_urange *ur@ = acceptable range
 *              @struct tvec_state *tv@ = test-vector state
 *
 * Returns:     Zero on success, %$-1$% on error.
 *
 * Use:         Checks that @sz@ is within the bounds described by @ur@,
 *              reporting an error if not.
 */

static int check_string_length(size_t sz, const struct tvec_urange *ur,
                               struct tvec_state *tv)
{
  if (ur && (ur->min > sz || sz > ur->max))
    return (tvec_error(tv,
                       "invalid string length %lu; must be in [%lu .. %lu]",
                       (unsigned long)sz, ur->min, ur->max));
  return (0);
}

/* --- @parse_text@, @parse_bytes@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @struct tvec_state *tv@ = test-vector state
 *
 * Returns:     Zero on success, %$-1$% on error.
 *
 * Use:         Parse a register value from an input file.
 *
 *              The input format for both kinds of strings is basically the
 *              same: a `compound string', consisting of
 *
 *                * single-quoted strings, which are interpreted entirely
 *                  literally, but can't contain single quotes or newlines;
 *
 *                * double-quoted strings, in which `%|\|%'-escapes are
 *                  interpreted as for characters;
 *
 *                * character names, marked by an initial `%|#|%' sign;
 *
 *                * special tokens marked by an initial `%|!|%' sign; or
 *
 *                * barewords interpreted according to the current coding
 *                  scheme.
 *
 *              The special tokens are
 *
 *                * `%|!bare|%', which causes subsequent sequences of
 *                  barewords to be treated as plain text;
 *
 *                * `%|!hex|%', `%|!base32|%', `%|!base64|%', which cause
 *                  subsequent barewords to be decoded in the requested
 *                  manner.
 *
 *                * `%|!repeat|% %$n$% %|{|% %%\textit{string}%% %|}|%',
 *                  which includes %$n$% copies of the (compound) string.
 *
 *              The only difference between text and binary strings is that
 *              the initial coding scheme is %|bare|% for text strings and
 *              %|hex|% for binary strings.
 */

static int parse_text(union tvec_regval *rv, const struct tvec_regdef *rd,
                      struct tvec_state *tv)
{
  void *p = rv->text.p;

  if (read_compound_string(&p, &rv->text.sz, TVCODE_BARE, 0, tv))
    return (-1);
  rv->text.p = p;
  if (check_string_length(rv->text.sz, rd->arg.p, tv)) return (-1);
  return (0);
}

static int parse_bytes(union tvec_regval *rv, const struct tvec_regdef *rd,
                       struct tvec_state *tv)
{
  void *p = rv->bytes.p;

  if (read_compound_string(&p, &rv->bytes.sz, TVCODE_HEX, 0, tv))
    return (-1);
  rv->bytes.p = p;
  if (check_string_length(rv->bytes.sz, rd->arg.p, tv)) return (-1);
  return (0);
}

/* --- @dump_text@, @dump_bytes@ --- *
 *
 * Arguments:   @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @unsigned style@ = output style (@TVSF_...@)
 *              @const struct gprintf_ops *gops@, @void *gp@ = format output
 *
 * Returns:     ---
 *
 * Use:         Dump a register value to the format output.
 *
 *              Text string values are dumped as plain text, in double quotes
 *              if necessary, and using backslash escape sequences for
 *              nonprintable characters.  Unless compact output is requested,
 *              strings consisting of multiple lines are dumped with each
 *              line of the string on a separate output line.
 *
 *              Binary string values are dumped in hexadecimal.  In compact
 *              style, the output simply consists of a single block of hex
 *              digits.  Otherwise, the dump is a display consisting of
 *              groups of hex digits, with comments showing the offset (if
 *              the string is long enough) and the corresponding plain text.
 *
 *              Empty strings are dumped as %|#empty|%.
 */

static void dump_empty(const char *ty, unsigned style,
                       const struct gprintf_ops *gops, void *go)
{
  if (style&TVSF_RAW) gprintf(gops, go, "%s:", ty);
  if (!(style&TVSF_COMPACT)) gprintf(gops, go, "#empty");
  if (!(style&(TVSF_COMPACT | TVSF_RAW))) gprintf(gops, go, " ; = ");
  if (!(style&TVSF_RAW)) gprintf(gops, go, "\"\"");
}


static void dump_text(const union tvec_regval *rv,
                      const struct tvec_regdef *rd,
                      unsigned style,
                      const struct gprintf_ops *gops, void *go)
{
  const unsigned char *p, *q, *l;
  unsigned f = 0;
#define f_nonword 1u
#define f_newline 2u

  if (!rv->text.sz) { dump_empty("text", style, gops, go); return; }

  p = (const unsigned char *)rv->text.p; l = p + rv->text.sz;
  if (style&TVSF_RAW) { gprintf(gops, go, "text:"); goto quote; }
  else if (style&TVSF_COMPACT) goto quote;

  switch (*p) {
    case '!': case '#': case ';': case '"': case '\'':
    case '(': case '{': case '[': case ']': case '}': case ')':
      f |= f_nonword; break;
  }
  for (q = p; q < l; q++)
    if (*q == '\n' && q != l - 1) f |= f_newline;
    else if (!*q || !ISGRAPH(*q) || *q == '\\') f |= f_nonword;
  if (f&f_newline) { gprintf(gops, go, "\n\t"); goto quote; }
  else if (f&f_nonword) goto quote;

  gops->putm(go, (const char *)p, rv->text.sz);
  return;

quote:
  gprintf(gops, go, "\"");
  for (q = p; q < l; q++)
    if (!ISPRINT(*q) || *q == '"') {
      if (p < q) gops->putm(go, (const char *)p, q - p);
      if (*q != '\n' || (style&TVSF_COMPACT))
        format_charesc(gops, go, *q, FCF_BRACE);
      else {
        if (q + 1 == l) { gprintf(gops, go, "\\n\""); return; }
        else gprintf(gops, go, "\\n\"\n\t\"");
      }
      p = q + 1;
    }
  if (p < q) gops->putm(go, (const char *)p, q - p);
  gprintf(gops, go, "\"");

#undef f_nonword
#undef f_newline
}

static void dump_bytes(const union tvec_regval *rv,
                       const struct tvec_regdef *rd,
                       unsigned style,
                       const struct gprintf_ops *gops, void *go)
{
  const unsigned char *p = rv->bytes.p, *l = p + rv->bytes.sz;
  size_t off, sz = rv->bytes.sz;
  unsigned i, n;
  int wd;

  if (!rv->text.sz) { dump_empty("bytes", style, gops, go); return; }

  if (style&(TVSF_COMPACT | TVSF_RAW)) {
    while (p < l) gprintf(gops, go, "%02x", *p++);
    return;
  }

  if (sz > 16) gprintf(gops, go, "\n\t");

  off = 0; wd = hex_width(sz);
  while (p < l) {
    if (l - p < 16) n = l - p;
    else n = 16;

    for (i = 0; i < n; i++) {
      if (i < n) gprintf(gops, go, "%02x", p[i]);
      else gprintf(gops, go, "  ");
      if (i < n - 1 && i%4 == 3) gprintf(gops, go, " ");
    }
    gprintf(gops, go, " ; ");
    if (sz > 16) gprintf(gops, go, "[%0*lx] ", wd, (unsigned long)off);
    for (i = 0; i < n; i++)
      gprintf(gops, go, "%c", isprint(p[i]) ? p[i] : '.');
    p += n; off += n;
    if (p < l) gprintf(gops, go, "\n\t");
  }
}

/* Text and byte string type definitions. */
const struct tvec_regty tvty_text = {
  init_text, release_text, eq_text,
  tobuf_text, frombuf_text,
  parse_text, dump_text
};
const struct tvec_regty tvty_bytes = {
  init_bytes, release_bytes, eq_bytes,
  tobuf_bytes, frombuf_bytes,
  parse_bytes, dump_bytes
};

/* --- @tvec_claimeq_text@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @const char *p0@, @size_t sz0@ = first string with length
 *              @const char *p1@, @size_t sz1@ = second string with length
 *              @const char *file@, @unsigned @lno@ = calling file and line
 *              @const char *expr@ = the expression to quote on failure
 *
 * Returns:     Nonzero if the strings at @p0@ and @p1@ are equal, otherwise
 *              zero.
 *
 * Use:         Check that strings at @p0@ and @p1@ are equal.  As for
 *              @tvec_claim@ above, a test case is automatically begun and
 *              ended if none is already underway.  If the values are
 *              unequal, then @tvec_fail@ is called, quoting @expr@, and the
 *              mismatched values are dumped: @p0@ is printed as the output
 *              value and @p1@ is printed as the input reference.
 */

int tvec_claimeq_text(struct tvec_state *tv,
                      const char *p0, size_t sz0,
                      const char *p1, size_t sz1,
                      const char *file, unsigned lno, const char *expr)
{
  tv->out[0].v.text.p = (/*unconst*/ char *)p0; tv->out[0].v.text.sz = sz0;
  tv->in[0].v.text.p =(/*unconst*/ char *) p1; tv->in[0].v.text.sz = sz1;
  return (tvec_claimeq(tv, &tvty_text, 0, file, lno, expr));
}

/* --- @tvec_claimeq_textz@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @const char *p0, *p1@ = two strings to compare
 *              @const char *file@, @unsigned @lno@ = calling file and line
 *              @const char *expr@ = the expression to quote on failure
 *
 * Returns:     Nonzero if the strings at @p0@ and @p1@ are equal, otherwise
 *              zero.
 *
 * Use:         Check that strings at @p0@ and @p1@ are equal, as for
 *              @tvec_claimeq_string@, except that the strings are assumed
 *              null-terminated, so their lengths don't need to be supplied
 *              explicitly.
 */

int tvec_claimeq_textz(struct tvec_state *tv,
                       const char *p0, const char *p1,
                       const char *file, unsigned lno, const char *expr)
{
  tv->out[0].v.text.p = (/*unconst*/ char *)p0;
    tv->out[0].v.text.sz = strlen(p0);
  tv->in[0].v.text.p = (/*unconst*/ char *)p1;
    tv->in[0].v.text.sz = strlen(p1);
  return (tvec_claimeq(tv, &tvty_text, 0, file, lno, expr));
}

/* --- @tvec_claimeq_bytes@ --- *
 *
 * Arguments:   @struct tvec_state *tv@ = test-vector state
 *              @const void *p0@, @size_t sz0@ = first string with length
 *              @const void *p1@, @size_t sz1@ = second string with length
 *              @const char *file@, @unsigned @lno@ = calling file and line
 *              @const char *expr@ = the expression to quote on failure
 *
 * Returns:     Nonzero if the strings at @p0@ and @p1@ are equal, otherwise
 *              zero.
 *
 * Use:         Check that binary strings at @p0@ and @p1@ are equal.  As for
 *              @tvec_claim@ above, a test case is automatically begun and
 *              ended if none is already underway.  If the values are
 *              unequal, then @tvec_fail@ is called, quoting @expr@, and the
 *              mismatched values are dumped: @p0@ is printed as the output
 *              value and @p1@ is printed as the input reference.
 */

int tvec_claimeq_bytes(struct tvec_state *tv,
                       const void *p0, size_t sz0,
                       const void *p1, size_t sz1,
                       const char *file, unsigned lno, const char *expr)
{
  tv->out[0].v.bytes.p = (/*unconst*/ void *)p0;
    tv->out[0].v.bytes.sz = sz0;
  tv->in[0].v.bytes.p = (/*unconst*/ void *)p1;
    tv->in[0].v.bytes.sz = sz1;
  return (tvec_claimeq(tv, &tvty_bytes, 0, file, lno, expr));
}

/* --- @tvec_alloctext@, @tvec_allocbytes@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @size_t sz@ = required size
 *
 * Returns:     ---
 *
 * Use:         Allocated space in a text or binary string register.  If the
 *              current register size is sufficient, its buffer is left
 *              alone; otherwise, the old buffer, if any, is freed and a
 *              fresh buffer allocated.  These functions are not intended to
 *              be used to adjust a buffer repeatedly, e.g., while building
 *              output incrementally: (a) they will perform badly, and (b)
 *              the old buffer contents are simply discarded if reallocation
 *              is necessary.  Instead, use a @dbuf@ or @dstr@.
 *
 *              The @tvec_alloctext@ function sneakily allocates an extra
 *              byte for a terminating zero.  The @tvec_allocbytes@ function
 *              doesn't do this.
 */

void tvec_alloctext(union tvec_regval *rv, size_t sz)
{
  if (rv->text.sz <= sz) { xfree(rv->text.p); rv->text.p = xmalloc(sz + 1); }
  rv->text.sz = sz;
}

void tvec_allocbytes(union tvec_regval *rv, size_t sz)
{
  if (rv->bytes.sz < sz) { xfree(rv->bytes.p); rv->bytes.p = xmalloc(sz); }
  rv->bytes.sz = sz;
}

/*----- Buffer type -------------------------------------------------------*/

/* --- @init_buffer@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     ---
 *
 * Use:         Initialize a register value.
 *
 *              Buffer values values are initialized with a null pointer,
 *              zero length, and zero residue, modulus, and offset.
 */

static void init_buffer(union tvec_regval *rv, const struct tvec_regdef *rd)
  { rv->buf.p = 0; rv->buf.sz = rv->buf.a = rv->buf.m = rv->buf.off = 0; }

/* --- @release_buffer@, @release_bytes@ --- *
 *
 * Arguments:   @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     ---
 *
 * Use:         Release resources held by a register value.
 *
 *              Buffers are freed.
 */

static void release_buffer(union tvec_regval *rv,
                           const struct tvec_regdef *rd)
  { if (rv->buf.p) xfree(rv->buf.p - rv->buf.off); }

/* --- @eq_buffer@ --- *
 *
 * Arguments:   @const union tvec_regval *rv0, *rv1@ = register values
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Nonzero if the values are equal, zero if unequal
 *
 * Use:         Compare register values for equality.
 *
 *              Buffer values are equal if and only if their sizes and
 *              alignment parameters are equal; their contents are
 *              %%\emph{not}%% compared.
 */

static int eq_buffer(const union tvec_regval *rv0,
                     const union tvec_regval *rv1,
                     const struct tvec_regdef *rd)
{
  return (rv0->buf.sz == rv1->buf.sz &&
          rv0->buf.a == rv1->buf.a &&
          rv0->buf.m == rv1->buf.m);
}

/* --- @tobuf_buffer@ --- *
 *
 * Arguments:   @buf *b@ = buffer
 *              @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Zero on success, %$-1$% on failure.
 *
 * Use:         Serialize a register value to a buffer.
 *
 *              Buffer values are serialized as their lengths, residues, and
 *              moduli, as unsigned integers.
 */

static int tobuf_buffer(buf *b, const union tvec_regval *rv,
                         const struct tvec_regdef *rd)
{
  return (unsigned_to_buf(b, rv->buf.sz) ||
          unsigned_to_buf(b, rv->buf.a) ||
          unsigned_to_buf(b, rv->buf.m));
}

/* --- @frombuf_buffer@ --- *
 *
 * Arguments:   @buf *b@ = buffer
 *              @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *
 * Returns:     Zero on success, %$-1$% on failure.
 *
 * Use:         Deserialize a register value from a buffer.
 *
 *              Buffer values are serialized as just their lengths, as
 *              unsigned integers.  The buffer is allocated on
 *              deserialization and filled with a distinctive pattern.
 */

static int frombuf_buffer(buf *b, union tvec_regval *rv,
                          const struct tvec_regdef *rd)
{
  unsigned long sz, a, m;

  if (unsigned_from_buf(b, &sz)) return (-1);
  if (unsigned_from_buf(b, &a)) return (-1);
  if (unsigned_from_buf(b, &m)) return (-1);
  if (sz > (size_t)-1 || a > (size_t)-1 || m > (size_t)-1)
    { buf_break(b); return (-1); }
  rv->buf.sz = sz; rv->buf.a = a; rv->buf.m = m;
  return (0);
}

/* --- @parse_buffer@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @struct tvec_state *tv@ = test-vector state
 *
 * Returns:     Zero on success, %$-1$% on error.
 *
 * Use:         Parse a register value from an input file.
 *
 *              The input format for a buffer value is a size, followed by an
 *              optional `%|@$%' and an alignment quantum and a further
 *              optional `%|+|%' and an alignment offset.  The size, quantum,
 *              and offset are syntactically sizes.
 *
 *              The buffer is not allocated.
 */

static int parse_buffer(union tvec_regval *rv,
                        const struct tvec_regdef *rd,
                        struct tvec_state *tv)
{
  unsigned long sz, a = 0, m = 0;
  int ch, rc;

  if (parse_szint(tv, &sz, "@;", "buffer length")) { rc = -1; goto end; }
  if (check_unsigned_range(sz, &tvrange_size, tv, "buffer length"))
    { rc = -1; goto end; }
  if (check_string_length(sz, rd->arg.p, tv)) { rc = -1; goto end; }

  tvec_skipspc(tv);
  ch = getc(tv->fp);
  if (ch == ';' || ch == '\n') { ungetc(ch, tv->fp); goto done; }
  else if (ch != '@') { rc = tvec_syntax(tv, ch, "`@'"); goto end; }

  if (parse_szint(tv, &m, "+;", "alignment quantum")) { rc = -1; goto end; }
  if (check_unsigned_range(a, &tvrange_size, tv, "alignment quantum"))
    { rc = -1; goto end; }
  if (m == 1) m = 0;

  tvec_skipspc(tv);
  ch = getc(tv->fp);
  if (ch == ';' || ch == '\n') { ungetc(ch, tv->fp); goto done; }
  else if (ch != '+') { rc = tvec_syntax(tv, ch, "`+'"); goto end; }

  if (parse_szint(tv, &a, ";", "alignment offset")) { rc = -1; goto end; }
  if (check_unsigned_range(m, &tvrange_size, tv, "alignment offset"))
    { rc = -1; goto end; }
  if (a >= m) {
    rc = tvec_error(tv, "alignment offset %lu >= quantum %lu",
                    (unsigned long)a, (unsigned long)m);
    goto end;
  }

done:
  if (tvec_flushtoeol(tv, 0)) { rc = -1; goto end; }
  rv->buf.sz = sz; rv->buf.a = a; rv->buf.m = m;
  rc = 0;
end:
  return (rc);
}

/* --- @dump_buffer@ --- *
 *
 * Arguments:   @const union tvec_regval *rv@ = register value
 *              @const struct tvec_regdef *rd@ = register definition
 *              @unsigned style@ = output style (@TVSF_...@)
 *              @const struct gprintf_ops *gops@, @void *gp@ = format output
 *
 * Returns:     ---
 *
 * Use:         Dump a register value to the format output.
 *
 *              Buffer values are dumped as their size, with the alignment
 *              quantum and alignment offset if these are non-default.
 */

static void dump_buffer(const union tvec_regval *rv,
                        const struct tvec_regdef *rd,
                        unsigned style,
                        const struct gprintf_ops *gops, void *go)
{
  format_size(gops, go, rv->buf.sz, style);
  if (rv->buf.m) {
    gprintf(gops, go, style&(TVSF_COMPACT | TVSF_RAW) ? "@" : " @ ");
    format_size(gops, go, rv->buf.m, style);
    if (rv->buf.a) {
      gprintf(gops, go, style&(TVSF_COMPACT | TVSF_RAW) ? "+" : " + ");
      format_size(gops, go, rv->buf.a, style);
    }
  }
  if (!(style&TVSF_COMPACT)) {
    gprintf(gops, go, " ; = %lu", (unsigned long)rv->buf.sz);
    if (rv->buf.m) {
      gprintf(gops, go, " @ %lu", (unsigned long)rv->buf.m);
      if (rv->buf.a) gprintf(gops, go, " + %lu", (unsigned long)rv->buf.a);
    }
    gprintf(gops, go, " = "); format_unsigned_hex(gops, go, rv->buf.sz);
    if (rv->buf.m) {
      gprintf(gops, go, " @ "); format_unsigned_hex(gops, go, rv->buf.m);
      if (rv->buf.a) {
        gprintf(gops, go, " + ");
        format_unsigned_hex(gops, go, rv->buf.a);
      }
    }
  }
}

/* Buffer type definition. */
const struct tvec_regty tvty_buffer = {
  init_buffer, release_buffer, eq_buffer,
  tobuf_buffer, frombuf_buffer,
  parse_buffer, dump_buffer
};

/* --- @tvec_initbuffer@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *              @const union tvec_regval *ref@ = source buffer
 *              @size_t sz@ = size to allocate
 *
 * Returns:     ---
 *
 * Use:         Initialize the alignment parameters in @rv@ to match @ref@,
 *              and the size to @sz@.
 */

void tvec_initbuffer(union tvec_regval *rv,
                     const union tvec_regval *ref, size_t sz)
  { rv->buf.sz = sz; rv->buf.a = ref->buf.a; rv->buf.m = ref->buf.m; }

/* --- @tvec_allocbuffer@ --- *
 *
 * Arguments:   @union tvec_regval *rv@ = register value
 *
 * Returns:     ---
 *
 * Use:         Allocate @sz@ bytes to the buffer and fill the space with a
 *              distinctive pattern.
 */

void tvec_allocbuffer(union tvec_regval *rv)
{
  unsigned char *p; size_t n;

  if (rv->buf.p) xfree(rv->buf.p - rv->buf.off);

  if (rv->buf.m < 2) {
    rv->buf.p = xmalloc(rv->buf.sz); rv->buf.off = 0;
  } else {
    p = xmalloc(rv->buf.sz + rv->buf.m - 1);
    n = (size_t)p%rv->buf.m;
    rv->buf.off = (rv->buf.a - n + rv->buf.m)%rv->buf.m;
    rv->buf.p = p + rv->buf.off;
  }
  memset(rv->buf.p, '?', rv->buf.sz);
}

/*----- That's all, folks -------------------------------------------------*/