/* -*-c-*-
*
- * $Id: mptext.c,v 1.8 2000/12/06 20:32:42 mdw Exp $
+ * $Id: mptext.c,v 1.17 2002/10/19 11:59:04 mdw Exp $
*
* Textual representation of multiprecision numbers
*
/*----- Revision history --------------------------------------------------*
*
* $Log: mptext.c,v $
+ * Revision 1.17 2002/10/19 11:59:04 mdw
+ * Fix leftovers bug in reading.
+ *
+ * Revision 1.16 2002/10/15 22:57:43 mdw
+ * Bug fix: prevent negative zero.
+ *
+ * Revision 1.15 2002/10/15 19:18:15 mdw
+ * Fix fencepost bugs in binary radix writing.
+ *
+ * Revision 1.14 2002/10/09 00:33:44 mdw
+ * Allow `0o' and `0b' prefixes for octal and binary (from Haskell)
+ *
+ * Revision 1.13 2002/10/09 00:21:06 mdw
+ * Allow user-specified `r_xx' bases to be up to 62.
+ *
+ * Revision 1.12 2002/01/13 19:51:18 mdw
+ * Extend the textual format to bases up to 62 by distinguishing case.
+ *
+ * Revision 1.11 2001/06/16 23:42:17 mdw
+ * Typesetting fixes.
+ *
+ * Revision 1.10 2001/06/16 13:22:39 mdw
+ * Added fast-track code for binary output bases, and tests.
+ *
+ * Revision 1.9 2001/02/03 16:05:17 mdw
+ * Make flags be unsigned. Improve the write algorithm: recurse until the
+ * parts are one word long and use single-precision arithmetic from there.
+ * Fix off-by-one bug when breaking the number apart.
+ *
* Revision 1.8 2000/12/06 20:32:42 mdw
* Reduce binary bytes (to allow marker bits to be ignored). Fix error
* message string a bit. Allow leading `+' signs.
*
* This is the number of bits in a @size_t@ object. Why?
*
- * To see this, let %$b = \mathit{MPW\_MAX} + 1$% and let %$Z$% be the
+ * To see this, let %$b = \textit{MPW\_MAX} + 1$% and let %$Z$% be the
* largest @size_t@ value. Then the largest possible @mp@ is %$M - 1$% where
* %$M = b^Z$%. Let %$r$% be a radix to read or write. Since the recursion
* squares the radix at each step, the highest number reached by the
/* --- Flags --- */
- enum {
- f_neg = 1u,
- f_ok = 2u
- };
+#define f_neg 1u
+#define f_ok 2u
+#define f_start 4u
/* --- Initialize the stacks --- */
/* --- If the radix is zero, look for leading zeros --- */
if (radix > 0) {
- assert(((void)"ascii radix must be <= 36", radix <= 36));
+ assert(((void)"ascii radix must be <= 62", radix <= 62));
rd = radix;
r = -1;
} else if (radix < 0) {
r = 0;
} else {
ch = ops->get(p);
- if (ch == 'x') {
- ch = ops->get(p);
- rd = 16;
- } else {
- rd = 8;
- f |= f_ok;
+ switch (ch) {
+ case 'x':
+ rd = 16;
+ goto prefix;
+ case 'o':
+ rd = 8;
+ goto prefix;
+ case 'b':
+ rd = 2;
+ goto prefix;
+ prefix:
+ ch = ops->get(p);
+ break;
+ default:
+ rd = 8;
+ f |= f_ok;
}
r = -1;
}
+ /* --- Use fast algorithm for binary radix --- *
+ *
+ * This is the restart point after having parsed a radix number from the
+ * input. We check whether the radix is binary, and if so use a fast
+ * algorithm which just stacks the bits up in the right order.
+ */
+
+restart:
+ switch (rd) {
+ unsigned bit;
+
+ case 2: bit = 1; goto bin;
+ case 4: bit = 2; goto bin;
+ case 8: bit = 3; goto bin;
+ case 16: bit = 4; goto bin;
+ case 32: bit = 5; goto bin;
+ case 64: bit = 6; goto bin;
+ case 128: bit = 7; goto bin;
+ default:
+ break;
+
+ /* --- The fast binary algorithm --- *
+ *
+ * We stack bits up starting at the top end of a word. When one word is
+ * full, we write it to the integer, and start another with the left-over
+ * bits. When the array in the integer is full, we resize using low-level
+ * calls and copy the current data to the top end. Finally, we do a single
+ * bit-shift when we know where the end of the number is.
+ */
+
+ bin: {
+ mpw a = 0;
+ unsigned b = MPW_BITS;
+ size_t len, n;
+ mpw *v;
+
+ m = mp_dest(MP_NEW, 1, nf);
+ len = n = m->sz;
+ n = len;
+ v = m->v + n;
+ for (;; ch = ops->get(p)) {
+ unsigned x;
+
+ if (ch < 0)
+ break;
+
+ /* --- Check that the character is a digit and in range --- */
+
+ if (radix < 0)
+ x = ch % rd;
+ else {
+ if (!isalnum(ch))
+ break;
+ if (ch >= '0' && ch <= '9')
+ x = ch - '0';
+ else {
+ if (rd <= 36)
+ ch = tolower(ch);
+ if (ch >= 'a' && ch <= 'z') /* ASCII dependent! */
+ x = ch - 'a' + 10;
+ else if (ch >= 'A' && ch <= 'Z')
+ x = ch - 'A' + 36;
+ else
+ break;
+ }
+ }
+ if (x >= rd)
+ break;
+
+ /* --- Feed the digit into the accumulator --- */
+
+ f |= f_ok;
+ if (!x && !(f & f_start))
+ continue;
+ f |= f_start;
+ if (b > bit) {
+ b -= bit;
+ a |= MPW(x) << b;
+ } else {
+ a |= MPW(x) >> (bit - b);
+ b += MPW_BITS - bit;
+ *--v = MPW(a);
+ n--;
+ if (!n) {
+ n = len;
+ len <<= 1;
+ v = mpalloc(m->a, len);
+ memcpy(v + n, m->v, MPWS(n));
+ mpfree(m->a, m->v);
+ m->v = v;
+ v = m->v + n;
+ }
+ a = (b < MPW_BITS) ? MPW(x) << b : 0;
+ }
+ }
+
+ /* --- Finish up --- */
+
+ if (!(f & f_ok)) {
+ mp_drop(m);
+ m = 0;
+ } else {
+ *--v = MPW(a);
+ n--;
+ m->sz = len;
+ m->vl = m->v + len;
+ m->f &= ~MP_UNDEF;
+ m = mp_lsr(m, m, (unsigned long)n * MPW_BITS + b);
+ }
+ ops->unget(ch, p);
+ goto done;
+ }}
+
/* --- Time to start --- */
for (;; ch = ops->get(p)) {
- int x;
+ unsigned x;
if (ch < 0)
break;
/* --- An underscore indicates a numbered base --- */
- if (ch == '_' && r > 0 && r <= 36) {
+ if (ch == '_' && r > 0 && r <= 62) {
unsigned i;
/* --- Clear out the stacks --- */
rd = r;
r = -1;
f &= ~f_ok;
- continue;
+ ch = ops->get(p);
+ goto restart;
}
/* --- Check that the character is a digit and in range --- */
if (ch >= '0' && ch <= '9')
x = ch - '0';
else {
- ch = tolower(ch);
+ if (rd <= 36)
+ ch = tolower(ch);
if (ch >= 'a' && ch <= 'z') /* ASCII dependent! */
x = ch - 'a' + 10;
+ else if (ch >= 'A' && ch <= 'Z')
+ x = ch - 'A' + 36;
else
break;
}
/* --- Bail out if the number was bad --- */
+done:
if (!(f & f_ok))
return (0);
if (f & f_neg)
m->f |= MP_NEG;
+ MP_SHRINK(m);
return (m);
+
+#undef f_start
+#undef f_neg
+#undef f_ok
}
/* --- @mp_write@ --- *
/* --- Simple case --- *
*
- * Use a fixed-sized buffer and the simple single-precision division
- * algorithm to pick off low-order digits. Put each digit in a buffer,
- * working backwards from the end. If the buffer becomes full, recurse to
- * get another one. Ensure that there are at least @z@ digits by writing
- * leading zeroes if there aren't enough real digits.
+ * Use a fixed-sized buffer and single-precision arithmetic to pick off
+ * low-order digits. Put each digit in a buffer, working backwards from the
+ * end. If the buffer becomes full, recurse to get another one. Ensure that
+ * there are at least @z@ digits by writing leading zeroes if there aren't
+ * enough real digits.
*/
-static int simple(mp *m, int radix, unsigned z,
+static int simple(mpw n, int radix, unsigned z,
const mptext_ops *ops, void *p)
{
int rc = 0;
int ch;
mpw x;
- x = mpx_udivn(m->v, m->vl, m->v, m->vl, rd);
- MP_SHRINK(m);
+ x = n % rd;
+ n /= rd;
if (radix < 0)
ch = x;
- else {
- if (x < 10)
- ch = '0' + x;
- else
- ch = 'a' + x - 10;
- }
+ else if (x < 10)
+ ch = '0' + x;
+ else if (x < 36) /* Ascii specific */
+ ch = 'a' + x - 10;
+ else
+ ch = 'A' + x - 36;
buf[--i] = ch;
if (z)
z--;
- } while (i && MP_LEN(m));
+ } while (i && n);
- if (MP_LEN(m))
- rc = simple(m, radix, z, ops, p);
+ if (n)
+ rc = simple(n, radix, z, ops, p);
else {
- static const char zero[32] = "00000000000000000000000000000000";
- while (!rc && z >= sizeof(zero)) {
- rc = ops->put(zero, sizeof(zero), p);
- z -= sizeof(zero);
+ char zbuf[32];
+ memset(zbuf, (radix < 0) ? 0 : '0', sizeof(zbuf));
+ while (!rc && z >= sizeof(zbuf)) {
+ rc = ops->put(zbuf, sizeof(zbuf), p);
+ z -= sizeof(zbuf);
}
if (!rc && z)
- rc = ops->put(zero, z, p);
+ rc = ops->put(zbuf, z, p);
}
if (!rc)
- ops->put(buf + i, sizeof(buf) - i, p);
- if (m->f & MP_BURN)
- BURN(buf);
+ rc = ops->put(buf + i, sizeof(buf) - i, p);
+ BURN(buf);
return (rc);
}
mp *q = MP_NEW;
unsigned d = 1 << i;
- if (MP_LEN(m) < 8)
- return (simple(m, radix, z, ops, p));
+ if (MP_LEN(m) < 2)
+ return (simple(MP_LEN(m) ? m->v[0] : 0, radix, z, ops, p));
+ assert(i);
mp_div(&q, &m, m, pr[i]);
if (!MP_LEN(q))
d = z;
return (rc);
}
+/* --- Binary case --- *
+ *
+ * Special case for binary output. Goes much faster.
+ */
+
+static int binary(mp *m, int bit, int radix, const mptext_ops *ops, void *p)
+{
+ mpw *v;
+ mpw a;
+ int rc = 0;
+ unsigned b;
+ unsigned mask;
+ unsigned long n;
+ unsigned f = 0;
+ char buf[8], *q;
+ unsigned x;
+ int ch;
+
+#define f_out 1u
+
+ /* --- Work out where to start --- */
+
+ n = mp_bits(m);
+ if (n % bit)
+ n += bit - (n % bit);
+ b = n % MPW_BITS;
+ n /= MPW_BITS;
+
+ if (n >= MP_LEN(m)) {
+ n--;
+ b += MPW_BITS;
+ }
+
+ v = m->v + n;
+ a = *v;
+ mask = (1 << bit) - 1;
+ q = buf;
+
+ /* --- Main code --- */
+
+ for (;;) {
+ if (b > bit) {
+ b -= bit;
+ x = a >> b;
+ } else {
+ x = a << (bit - b);
+ b += MPW_BITS - bit;
+ if (v == m->v)
+ break;
+ a = *--v;
+ if (b < MPW_BITS)
+ x |= a >> b;
+ }
+ x &= mask;
+ if (!x && !(f & f_out))
+ continue;
+
+ if (radix < 0)
+ ch = x;
+ else if (x < 10)
+ ch = '0' + x;
+ else if (x < 36)
+ ch = 'a' + x - 10; /* Ascii specific */
+ else
+ ch = 'A' + x - 36;
+ *q++ = ch;
+ if (q >= buf + sizeof(buf)) {
+ if ((rc = ops->put(buf, sizeof(buf), p)) != 0)
+ goto done;
+ q = buf;
+ }
+ f |= f_out;
+ }
+
+ x &= mask;
+ if (radix < 0)
+ ch = x;
+ else if (x < 10)
+ ch = '0' + x;
+ else if (x < 36)
+ ch = 'a' + x - 10; /* Ascii specific */
+ else
+ ch = 'A' + x - 36;
+ *q++ = ch;
+ rc = ops->put(buf, q - buf, p);
+
+done:
+ mp_drop(m);
+ return (rc);
+
+#undef f_out
+}
+
/* --- Main driver code --- */
int mp_write(mp *m, int radix, const mptext_ops *ops, void *p)
{
int rc;
+ if (MP_EQ(m, MP_ZERO))
+ return (ops->put("0", 1, p));
+
/* --- Set various things up --- */
m = MP_COPY(m);
/* --- Check the radix for sensibleness --- */
if (radix > 0)
- assert(((void)"ascii radix must be <= 36", radix <= 36));
+ assert(((void)"ascii radix must be <= 62", radix <= 62));
else if (radix < 0)
assert(((void)"binary radix must fit in a byte", -radix < UCHAR_MAX));
else
m->f &= ~MP_NEG;
}
+ /* --- Handle binary radix --- */
+
+ switch (radix) {
+ case 2: case -2: return (binary(m, 1, radix, ops, p));
+ case 4: case -4: return (binary(m, 2, radix, ops, p));
+ case 8: case -8: return (binary(m, 3, radix, ops, p));
+ case 16: case -16: return (binary(m, 4, radix, ops, p));
+ case 32: case -32: return (binary(m, 5, radix, ops, p));
+ case -64: return (binary(m, 6, radix, ops, p));
+ case -128: return (binary(m, 7, radix, ops, p));
+ }
+
/* --- If the number is small, do it the easy way --- */
- if (MP_LEN(m) < 8)
- rc = simple(m, radix, 0, ops, p);
+ if (MP_LEN(m) < 2)
+ rc = simple(MP_LEN(m) ? m->v[0] : 0, radix, 0, ops, p);
/* --- Use a clever algorithm --- *
*
else {
mp *pr[DEPTH];
- size_t target = MP_LEN(m) / 2;
+ size_t target = (MP_LEN(m) + 1) / 2;
unsigned i = 0;
mp *z = mp_new(1, 0);
int ok = 1;
int ib = *(int *)v[0].buf, ob = *(int *)v[2].buf;
dstr d = DSTR_INIT;
- mp *m = mp_readdstr(MP_NEW, &v[1], 0, ib);
+ size_t off = 0;
+ mp *m = mp_readdstr(MP_NEW, &v[1], &off, ib);
if (m) {
if (!ob) {
fprintf(stderr, "*** unexpected successful parse\n"
- "*** input [%i] = ", ib);
+ "*** input [%2i] = ", ib);
if (ib < 0)
type_hex.dump(&v[1], stderr);
else
mp_writedstr(m, &d, ob);
if (d.len != v[3].len || memcmp(d.buf, v[3].buf, d.len) != 0) {
fprintf(stderr, "*** failed read or write\n"
- "*** input [%i] = ", ib);
+ "*** input [%2i] = ", ib);
if (ib < 0)
type_hex.dump(&v[1], stderr);
else
fputs(v[1].buf, stderr);
- fprintf(stderr, "\n*** output [%i] = ", ob);
+ fprintf(stderr, "\n*** output [%2i] = ", ob);
if (ob < 0)
type_hex.dump(&d, stderr);
else
fputs(d.buf, stderr);
- fprintf(stderr, "\n*** expected [%i] = ", ob);
+ fprintf(stderr, "\n*** expected [%2i] = ", ob);
if (ob < 0)
type_hex.dump(&v[3], stderr);
else
} else {
if (ob) {
fprintf(stderr, "*** unexpected parse failure\n"
- "*** input [%i] = ", ib);
+ "*** input [%2i] = ", ib);
if (ib < 0)
type_hex.dump(&v[1], stderr);
else
fputs(v[1].buf, stderr);
- fprintf(stderr, "\n*** expected [%i] = ", ob);
+ fprintf(stderr, "\n*** expected [%2i] = ", ob);
if (ob < 0)
type_hex.dump(&v[3], stderr);
else
}
}
+ if (v[1].len - off != v[4].len ||
+ memcmp(v[1].buf + off, v[4].buf, v[4].len) != 0) {
+ fprintf(stderr, "*** leftovers incorrect\n"
+ "*** input [%2i] = ", ib);
+ if (ib < 0)
+ type_hex.dump(&v[1], stderr);
+ else
+ fputs(v[1].buf, stderr);
+ fprintf(stderr, "\n*** expected `%s'\n"
+ "*** found `%s'\n",
+ v[4].buf, v[1].buf + off);
+ ok = 0;
+ }
+
dstr_destroy(&d);
assert(mparena_count(MPARENA_GLOBAL) == 0);
return (ok);
static test_chunk tests[] = {
{ "mptext-ascii", verify,
- { &type_int, &type_string, &type_int, &type_string, 0 } },
+ { &type_int, &type_string, &type_int, &type_string, &type_string, 0 } },
{ "mptext-bin-in", verify,
- { &type_int, &type_hex, &type_int, &type_string, 0 } },
+ { &type_int, &type_hex, &type_int, &type_string, &type_string, 0 } },
{ "mptext-bin-out", verify,
- { &type_int, &type_string, &type_int, &type_hex, 0 } },
+ { &type_int, &type_string, &type_int, &type_hex, &type_string, 0 } },
{ 0, 0, { 0 } }
};