Coverage Report

Created: 2026-07-14 06:16

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
/src/cpython/Python/dtoa.c
Line
Count
Source
1
/****************************************************************
2
 *
3
 * The author of this software is David M. Gay.
4
 *
5
 * Copyright (c) 1991, 2000, 2001 by Lucent Technologies.
6
 *
7
 * Permission to use, copy, modify, and distribute this software for any
8
 * purpose without fee is hereby granted, provided that this entire notice
9
 * is included in all copies of any software which is or includes a copy
10
 * or modification of this software and in all copies of the supporting
11
 * documentation for such software.
12
 *
13
 * THIS SOFTWARE IS BEING PROVIDED "AS IS", WITHOUT ANY EXPRESS OR IMPLIED
14
 * WARRANTY.  IN PARTICULAR, NEITHER THE AUTHOR NOR LUCENT MAKES ANY
15
 * REPRESENTATION OR WARRANTY OF ANY KIND CONCERNING THE MERCHANTABILITY
16
 * OF THIS SOFTWARE OR ITS FITNESS FOR ANY PARTICULAR PURPOSE.
17
 *
18
 ***************************************************************/
19
20
/****************************************************************
21
 * This is dtoa.c by David M. Gay, downloaded from
22
 * http://www.netlib.org/fp/dtoa.c on April 15, 2009 and modified for
23
 * inclusion into the Python core by Mark E. T. Dickinson and Eric V. Smith.
24
 *
25
 * Please remember to check http://www.netlib.org/fp regularly (and especially
26
 * before any Python release) for bugfixes and updates.
27
 *
28
 * The major modifications from Gay's original code are as follows:
29
 *
30
 *  0. The original code has been specialized to Python's needs by removing
31
 *     many of the #ifdef'd sections.  In particular, code to support VAX and
32
 *     IBM floating-point formats, hex NaNs, hex floats, locale-aware
33
 *     treatment of the decimal point, and setting of the inexact flag have
34
 *     been removed.
35
 *
36
 *  1. We use PyMem_Malloc and PyMem_Free in place of malloc and free.
37
 *
38
 *  2. The public functions strtod, dtoa and freedtoa all now have
39
 *     a _Py_dg_ prefix.
40
 *
41
 *  3. Instead of assuming that PyMem_Malloc always succeeds, we thread
42
 *     PyMem_Malloc failures through the code.  The functions
43
 *
44
 *       Balloc, multadd, s2b, i2b, mult, pow5mult, lshift, diff, d2b
45
 *
46
 *     of return type *Bigint all return NULL to indicate a malloc failure.
47
 *     Similarly, rv_alloc and nrv_alloc (return type char *) return NULL on
48
 *     failure.  bigcomp now has return type int (it used to be void) and
49
 *     returns -1 on failure and 0 otherwise.  _Py_dg_dtoa returns NULL
50
 *     on failure.  _Py_dg_strtod indicates failure due to malloc failure
51
 *     by returning -1.0, setting errno=ENOMEM and *se to s00.
52
 *
53
 *  4. The static variable dtoa_result has been removed.  Callers of
54
 *     _Py_dg_dtoa are expected to call _Py_dg_freedtoa to free
55
 *     the memory allocated by _Py_dg_dtoa.
56
 *
57
 *  5. The code has been reformatted to better fit with Python's
58
 *     C style guide (PEP 7).
59
 *
60
 *  6. A bug in the memory allocation has been fixed: to avoid FREEing memory
61
 *     that hasn't been MALLOC'ed, private_mem should only be used when k <=
62
 *     Kmax.
63
 *
64
 *  7. _Py_dg_strtod has been modified so that it doesn't accept strings with
65
 *     leading whitespace.
66
 *
67
 *  8. A corner case where _Py_dg_dtoa didn't strip trailing zeros has been
68
 *     fixed. (bugs.python.org/issue40780)
69
 *
70
 ***************************************************************/
71
72
/* Please send bug reports for the original dtoa.c code to David M. Gay (dmg
73
 * at acm dot org, with " at " changed at "@" and " dot " changed to ".").
74
 * Please report bugs for this modified version using the Python issue tracker
75
 * as detailed at (https://devguide.python.org/triage/issue-tracker/). */
76
77
/* On a machine with IEEE extended-precision registers, it is
78
 * necessary to specify double-precision (53-bit) rounding precision
79
 * before invoking strtod or dtoa.  If the machine uses (the equivalent
80
 * of) Intel 80x87 arithmetic, the call
81
 *      _control87(PC_53, MCW_PC);
82
 * does this with many compilers.  Whether this or another call is
83
 * appropriate depends on the compiler; for this to work, it may be
84
 * necessary to #include "float.h" or another system-dependent header
85
 * file.
86
 */
87
88
/* strtod for IEEE-, VAX-, and IBM-arithmetic machines.
89
 *
90
 * This strtod returns a nearest machine number to the input decimal
91
 * string (or sets errno to ERANGE).  With IEEE arithmetic, ties are
92
 * broken by the IEEE round-even rule.  Otherwise ties are broken by
93
 * biased rounding (add half and chop).
94
 *
95
 * Inspired loosely by William D. Clinger's paper "How to Read Floating
96
 * Point Numbers Accurately" [Proc. ACM SIGPLAN '90, pp. 92-101].
97
 *
98
 * Modifications:
99
 *
100
 *      1. We only require IEEE, IBM, or VAX double-precision
101
 *              arithmetic (not IEEE double-extended).
102
 *      2. We get by with floating-point arithmetic in a case that
103
 *              Clinger missed -- when we're computing d * 10^n
104
 *              for a small integer d and the integer n is not too
105
 *              much larger than 22 (the maximum integer k for which
106
 *              we can represent 10^k exactly), we may be able to
107
 *              compute (d*10^k) * 10^(e-k) with just one roundoff.
108
 *      3. Rather than a bit-at-a-time adjustment of the binary
109
 *              result in the hard case, we use floating-point
110
 *              arithmetic to determine the adjustment to within
111
 *              one bit; only in really hard cases do we need to
112
 *              compute a second residual.
113
 *      4. Because of 3., we don't need a large table of powers of 10
114
 *              for ten-to-e (just some small tables, e.g. of 10^k
115
 *              for 0 <= k <= 22).
116
 */
117
118
/* Linking of Python's #defines to Gay's #defines starts here. */
119
120
#include "Python.h"
121
#include "pycore_dtoa.h"          // _PY_SHORT_FLOAT_REPR
122
#include "pycore_interp_structs.h"// struct Bigint
123
#include "pycore_pystate.h"       // _PyInterpreterState_GET()
124
#include <stdlib.h>               // exit()
125
126
127
/* if _PY_SHORT_FLOAT_REPR == 0, then don't even try to compile
128
   the following code */
129
#if _PY_SHORT_FLOAT_REPR == 1
130
131
#include "float.h"
132
133
44
#define MALLOC PyMem_Malloc
134
0
#define FREE PyMem_Free
135
136
/* This code should also work for ARM mixed-endian format on little-endian
137
   machines, where doubles have byte order 45670123 (in increasing address
138
   order, 0 being the least significant byte). */
139
#ifdef DOUBLE_IS_LITTLE_ENDIAN_IEEE754
140
#  define IEEE_8087
141
#endif
142
#if defined(DOUBLE_IS_BIG_ENDIAN_IEEE754)
143
#  define IEEE_MC68k
144
#endif
145
#if defined(IEEE_8087) + defined(IEEE_MC68k) != 1
146
#error "Exactly one of IEEE_8087 or IEEE_MC68k should be defined."
147
#endif
148
149
/* The code below assumes that the endianness of integers matches the
150
   endianness of the two 32-bit words of a double.  Check this. */
151
#if defined(WORDS_BIGENDIAN) && defined(DOUBLE_IS_LITTLE_ENDIAN_IEEE754)
152
#error "doubles and ints have incompatible endianness"
153
#endif
154
155
#if !defined(WORDS_BIGENDIAN) && defined(DOUBLE_IS_BIG_ENDIAN_IEEE754)
156
#error "doubles and ints have incompatible endianness"
157
#endif
158
159
160
typedef uint32_t ULong;
161
typedef int32_t Long;
162
typedef uint64_t ULLong;
163
164
#undef DEBUG
165
#ifdef Py_DEBUG
166
#define DEBUG
167
#endif
168
169
/* End Python #define linking */
170
171
#ifdef DEBUG
172
#define Bug(x) {fprintf(stderr, "%s\n", x); exit(1);}
173
#endif
174
175
typedef union { double d; ULong L[2]; } U;
176
177
#ifdef IEEE_8087
178
1.44M
#define word0(x) (x)->L[1]
179
952k
#define word1(x) (x)->L[0]
180
#else
181
#define word0(x) (x)->L[0]
182
#define word1(x) (x)->L[1]
183
#endif
184
3.62M
#define dval(x) (x)->d
185
186
#ifndef STRTOD_DIGLIM
187
72.0k
#define STRTOD_DIGLIM 40
188
#endif
189
190
/* maximum permitted exponent value for strtod; exponents larger than
191
   MAX_ABS_EXP in absolute value get truncated to +-MAX_ABS_EXP.  MAX_ABS_EXP
192
   should fit into an int. */
193
#ifndef MAX_ABS_EXP
194
491k
#define MAX_ABS_EXP 1100000000U
195
#endif
196
/* Bound on length of pieces of input strings in _Py_dg_strtod; specifically,
197
   this is used to bound the total number of digits ignoring leading zeros and
198
   the number of digits that follow the decimal point.  Ideally, MAX_DIGITS
199
   should satisfy MAX_DIGITS + 400 < MAX_ABS_EXP; that ensures that the
200
   exponent clipping in _Py_dg_strtod can't affect the value of the output. */
201
#ifndef MAX_DIGITS
202
1.65M
#define MAX_DIGITS 1000000000U
203
#endif
204
205
/* Guard against trying to use the above values on unusual platforms with ints
206
 * of width less than 32 bits. */
207
#if MAX_ABS_EXP > INT_MAX
208
#error "MAX_ABS_EXP should fit in an int"
209
#endif
210
#if MAX_DIGITS > INT_MAX
211
#error "MAX_DIGITS should fit in an int"
212
#endif
213
214
/* The following definition of Storeinc is appropriate for MIPS processors.
215
 * An alternative that might be better on some machines is
216
 * #define Storeinc(a,b,c) (*a++ = b << 16 | c & 0xffff)
217
 */
218
#if defined(IEEE_8087)
219
#define Storeinc(a,b,c) (((unsigned short *)a)[1] = (unsigned short)b,  \
220
                         ((unsigned short *)a)[0] = (unsigned short)c, a++)
221
#else
222
#define Storeinc(a,b,c) (((unsigned short *)a)[0] = (unsigned short)b,  \
223
                         ((unsigned short *)a)[1] = (unsigned short)c, a++)
224
#endif
225
226
/* #define P DBL_MANT_DIG */
227
/* Ten_pmax = floor(P*log(2)/log(5)) */
228
/* Bletch = (highest power of 2 < DBL_MAX_10_EXP) / 16 */
229
/* Quick_max = floor((P-1)*log(FLT_RADIX)/log(10) - 1) */
230
/* Int_max = floor(P*log(FLT_RADIX)/log(10) - 1) */
231
232
161k
#define Exp_shift  20
233
79.8k
#define Exp_shift1 20
234
463k
#define Exp_msk1    0x100000
235
#define Exp_msk11   0x100000
236
830k
#define Exp_mask  0x7ff00000
237
389k
#define P 53
238
#define Nbits 53
239
199k
#define Bias 1023
240
#define Emax 1023
241
#define Emin (-1022)
242
288k
#define Etiny (-1074)  /* smallest denormal is 2**Etiny */
243
102k
#define Exp_1  0x3ff00000
244
35.6k
#define Exp_11 0x3ff00000
245
214k
#define Ebits 11
246
141k
#define Frac_mask  0xfffff
247
37.4k
#define Frac_mask1 0xfffff
248
961k
#define Ten_pmax 22
249
76
#define Bletch 0x10
250
58.6k
#define Bndry_mask  0xfffff
251
7.59k
#define Bndry_mask1 0xfffff
252
55.9k
#define Sign_bit 0x80000000
253
6.02k
#define Log2P 1
254
#define Tiny0 0
255
29.7k
#define Tiny1 1
256
40.0k
#define Quick_max 14
257
23.8k
#define Int_max 14
258
259
#ifndef Flt_Rounds
260
#ifdef FLT_ROUNDS
261
524k
#define Flt_Rounds FLT_ROUNDS
262
#else
263
#define Flt_Rounds 1
264
#endif
265
#endif /*Flt_Rounds*/
266
267
#define Rounding Flt_Rounds
268
269
2.80k
#define Big0 (Frac_mask1 | Exp_msk1*(DBL_MAX_EXP+Bias-1))
270
1.78k
#define Big1 0xffffffff
271
272
/* Bits of the representation of positive infinity. */
273
274
#define POSINF_WORD0 0x7ff00000
275
#define POSINF_WORD1 0
276
277
/* struct BCinfo is used to pass information from _Py_dg_strtod to bigcomp */
278
279
typedef struct BCinfo BCinfo;
280
struct
281
BCinfo {
282
    int e0, nd, nd0, scale;
283
};
284
285
15.7M
#define FFFFFFFF 0xffffffffUL
286
287
/* struct Bigint is used to represent arbitrary-precision integers.  These
288
   integers are stored in sign-magnitude format, with the magnitude stored as
289
   an array of base 2**32 digits.  Bigints are always normalized: if x is a
290
   Bigint then x->wds >= 1, and either x->wds == 1 or x[wds-1] is nonzero.
291
292
   The Bigint fields are as follows:
293
294
     - next is a header used by Balloc and Bfree to keep track of lists
295
         of freed Bigints;  it's also used for the linked list of
296
         powers of 5 of the form 5**2**i used by pow5mult.
297
     - k indicates which pool this Bigint was allocated from
298
     - maxwds is the maximum number of words space was allocated for
299
       (usually maxwds == 2**k)
300
     - sign is 1 for negative Bigints, 0 for positive.  The sign is unused
301
       (ignored on inputs, set to 0 on outputs) in almost all operations
302
       involving Bigints: a notable exception is the diff function, which
303
       ignores signs on inputs but sets the sign of the output correctly.
304
     - wds is the actual number of significant words
305
     - x contains the vector of words (digits) for this Bigint, from least
306
       significant (x[0]) to most significant (x[wds-1]).
307
*/
308
309
// struct Bigint is defined in pycore_dtoa.h.
310
typedef struct Bigint Bigint;
311
312
#if !defined(Py_GIL_DISABLED) && !defined(Py_USING_MEMORY_DEBUGGER)
313
314
/* Memory management: memory is allocated from, and returned to, Kmax+1 pools
315
   of memory, where pool k (0 <= k <= Kmax) is for Bigints b with b->maxwds ==
316
   1 << k.  These pools are maintained as linked lists, with freelist[k]
317
   pointing to the head of the list for pool k.
318
319
   On allocation, if there's no free slot in the appropriate pool, MALLOC is
320
   called to get more memory.  This memory is not returned to the system until
321
   Python quits.  There's also a private memory pool that's allocated from
322
   in preference to using MALLOC.
323
324
   For Bigints with more than (1 << Kmax) digits (which implies at least 1233
325
   decimal digits), memory is directly allocated using MALLOC, and freed using
326
   FREE.
327
328
   XXX: it would be easy to bypass this memory-management system and
329
   translate each call to Balloc into a call to PyMem_Malloc, and each
330
   Bfree to PyMem_Free.  Investigate whether this has any significant
331
   performance on impact. */
332
333
5.66M
#define freelist interp->dtoa.freelist
334
403
#define private_mem interp->dtoa.preallocated
335
1.12k
#define pmem_next interp->dtoa.preallocated_next
336
337
/* Allocate space for a Bigint with up to 1<<k digits */
338
339
static Bigint *
340
Balloc(int k)
341
1.41M
{
342
1.41M
    int x;
343
1.41M
    Bigint *rv;
344
1.41M
    unsigned int len;
345
1.41M
    PyInterpreterState *interp = _PyInterpreterState_GET();
346
347
1.41M
    if (k <= Bigint_Kmax && (rv = freelist[k]))
348
1.41M
        freelist[k] = rv->next;
349
403
    else {
350
403
        x = 1 << k;
351
403
        len = (sizeof(Bigint) + (x-1)*sizeof(ULong) + sizeof(double) - 1)
352
403
            /sizeof(double);
353
403
        if (k <= Bigint_Kmax &&
354
403
            pmem_next - private_mem + len <= (Py_ssize_t)Bigint_PREALLOC_SIZE
355
403
        ) {
356
359
            rv = (Bigint*)pmem_next;
357
359
            pmem_next += len;
358
359
        }
359
44
        else {
360
44
            rv = (Bigint*)MALLOC(len*sizeof(double));
361
44
            if (rv == NULL)
362
0
                return NULL;
363
44
        }
364
403
        rv->k = k;
365
403
        rv->maxwds = x;
366
403
    }
367
1.41M
    rv->sign = rv->wds = 0;
368
1.41M
    return rv;
369
1.41M
}
370
371
/* Free a Bigint allocated with Balloc */
372
373
static void
374
Bfree(Bigint *v)
375
4.09M
{
376
4.09M
    if (v) {
377
1.41M
        if (v->k > Bigint_Kmax)
378
0
            FREE((void*)v);
379
1.41M
        else {
380
1.41M
            PyInterpreterState *interp = _PyInterpreterState_GET();
381
1.41M
            v->next = freelist[v->k];
382
1.41M
            freelist[v->k] = v;
383
1.41M
        }
384
1.41M
    }
385
4.09M
}
386
387
#undef pmem_next
388
#undef private_mem
389
#undef freelist
390
391
#else
392
393
/* Alternative versions of Balloc and Bfree that use PyMem_Malloc and
394
   PyMem_Free directly in place of the custom memory allocation scheme above.
395
   These are provided for the benefit of memory debugging tools like
396
   Valgrind. */
397
398
/* Allocate space for a Bigint with up to 1<<k digits */
399
400
static Bigint *
401
Balloc(int k)
402
{
403
    int x;
404
    Bigint *rv;
405
    unsigned int len;
406
407
    x = 1 << k;
408
    len = (sizeof(Bigint) + (x-1)*sizeof(ULong) + sizeof(double) - 1)
409
        /sizeof(double);
410
411
    rv = (Bigint*)MALLOC(len*sizeof(double));
412
    if (rv == NULL)
413
        return NULL;
414
415
    rv->k = k;
416
    rv->maxwds = x;
417
    rv->sign = rv->wds = 0;
418
    return rv;
419
}
420
421
/* Free a Bigint allocated with Balloc */
422
423
static void
424
Bfree(Bigint *v)
425
{
426
    if (v) {
427
        FREE((void*)v);
428
    }
429
}
430
431
#endif /* !defined(Py_GIL_DISABLED) && !defined(Py_USING_MEMORY_DEBUGGER) */
432
433
96.4k
#define Bcopy(x,y) memcpy((char *)&x->sign, (char *)&y->sign,   \
434
96.4k
                          y->wds*sizeof(Long) + 2*sizeof(int))
435
436
/* Multiply a Bigint b by m and add a.  Either modifies b in place and returns
437
   a pointer to the modified b, or Bfrees b and returns a pointer to a copy.
438
   On failure, return NULL.  In this case, b will have been already freed. */
439
440
static Bigint *
441
multadd(Bigint *b, int m, int a)       /* multiply by m and add a */
442
676k
{
443
676k
    int i, wds;
444
676k
    ULong *x;
445
676k
    ULLong carry, y;
446
676k
    Bigint *b1;
447
448
676k
    wds = b->wds;
449
676k
    x = b->x;
450
676k
    i = 0;
451
676k
    carry = a;
452
2.39M
    do {
453
2.39M
        y = *x * (ULLong)m + carry;
454
2.39M
        carry = y >> 32;
455
2.39M
        *x++ = (ULong)(y & FFFFFFFF);
456
2.39M
    }
457
2.39M
    while(++i < wds);
458
676k
    if (carry) {
459
44.2k
        if (wds >= b->maxwds) {
460
1.65k
            b1 = Balloc(b->k+1);
461
1.65k
            if (b1 == NULL){
462
0
                Bfree(b);
463
0
                return NULL;
464
0
            }
465
1.65k
            Bcopy(b1, b);
466
1.65k
            Bfree(b);
467
1.65k
            b = b1;
468
1.65k
        }
469
44.2k
        b->x[wds++] = (ULong)carry;
470
44.2k
        b->wds = wds;
471
44.2k
    }
472
676k
    return b;
473
676k
}
474
475
/* convert a string s containing nd decimal digits (possibly containing a
476
   decimal separator at position nd0, which is ignored) to a Bigint.  This
477
   function carries on where the parsing code in _Py_dg_strtod leaves off: on
478
   entry, y9 contains the result of converting the first 9 digits.  Returns
479
   NULL on failure. */
480
481
static Bigint *
482
s2b(const char *s, int nd0, int nd, ULong y9)
483
72.0k
{
484
72.0k
    Bigint *b;
485
72.0k
    int i, k;
486
72.0k
    Long x, y;
487
488
72.0k
    x = (nd + 8) / 9;
489
106k
    for(k = 0, y = 1; x > y; y <<= 1, k++) ;
490
72.0k
    b = Balloc(k);
491
72.0k
    if (b == NULL)
492
0
        return NULL;
493
72.0k
    b->x[0] = y9;
494
72.0k
    b->wds = 1;
495
496
72.0k
    if (nd <= 9)
497
44.2k
      return b;
498
499
27.8k
    s += 9;
500
224k
    for (i = 9; i < nd0; i++) {
501
196k
        b = multadd(b, 10, *s++ - '0');
502
196k
        if (b == NULL)
503
0
            return NULL;
504
196k
    }
505
27.8k
    s++;
506
84.2k
    for(; i < nd; i++) {
507
56.4k
        b = multadd(b, 10, *s++ - '0');
508
56.4k
        if (b == NULL)
509
0
            return NULL;
510
56.4k
    }
511
27.8k
    return b;
512
27.8k
}
513
514
/* count leading 0 bits in the 32-bit integer x. */
515
516
static int
517
hi0bits(ULong x)
518
125k
{
519
125k
    int k = 0;
520
521
125k
    if (!(x & 0xffff0000)) {
522
71.7k
        k = 16;
523
71.7k
        x <<= 16;
524
71.7k
    }
525
125k
    if (!(x & 0xff000000)) {
526
78.8k
        k += 8;
527
78.8k
        x <<= 8;
528
78.8k
    }
529
125k
    if (!(x & 0xf0000000)) {
530
69.4k
        k += 4;
531
69.4k
        x <<= 4;
532
69.4k
    }
533
125k
    if (!(x & 0xc0000000)) {
534
67.0k
        k += 2;
535
67.0k
        x <<= 2;
536
67.0k
    }
537
125k
    if (!(x & 0x80000000)) {
538
72.1k
        k++;
539
72.1k
        if (!(x & 0x40000000))
540
0
            return 32;
541
72.1k
    }
542
125k
    return k;
543
125k
}
544
545
/* count trailing 0 bits in the 32-bit integer y, and shift y right by that
546
   number of bits. */
547
548
static int
549
lo0bits(ULong *y)
550
39.9k
{
551
39.9k
    int k;
552
39.9k
    ULong x = *y;
553
554
39.9k
    if (x & 7) {
555
24.1k
        if (x & 1)
556
11.8k
            return 0;
557
12.3k
        if (x & 2) {
558
6.93k
            *y = x >> 1;
559
6.93k
            return 1;
560
6.93k
        }
561
5.38k
        *y = x >> 2;
562
5.38k
        return 2;
563
12.3k
    }
564
15.8k
    k = 0;
565
15.8k
    if (!(x & 0xffff)) {
566
7.17k
        k = 16;
567
7.17k
        x >>= 16;
568
7.17k
    }
569
15.8k
    if (!(x & 0xff)) {
570
4.08k
        k += 8;
571
4.08k
        x >>= 8;
572
4.08k
    }
573
15.8k
    if (!(x & 0xf)) {
574
8.78k
        k += 4;
575
8.78k
        x >>= 4;
576
8.78k
    }
577
15.8k
    if (!(x & 0x3)) {
578
8.19k
        k += 2;
579
8.19k
        x >>= 2;
580
8.19k
    }
581
15.8k
    if (!(x & 1)) {
582
9.93k
        k++;
583
9.93k
        x >>= 1;
584
9.93k
        if (!x)
585
0
            return 32;
586
9.93k
    }
587
15.8k
    *y = x;
588
15.8k
    return k;
589
15.8k
}
590
591
/* convert a small nonnegative integer to a Bigint */
592
593
static Bigint *
594
i2b(int i)
595
160k
{
596
160k
    Bigint *b;
597
598
160k
    b = Balloc(1);
599
160k
    if (b == NULL)
600
0
        return NULL;
601
160k
    b->x[0] = i;
602
160k
    b->wds = 1;
603
160k
    return b;
604
160k
}
605
606
/* multiply two Bigints.  Returns a new Bigint, or NULL on failure.  Ignores
607
   the signs of a and b. */
608
609
static Bigint *
610
mult(Bigint *a, Bigint *b)
611
391k
{
612
391k
    Bigint *c;
613
391k
    int k, wa, wb, wc;
614
391k
    ULong *x, *xa, *xae, *xb, *xbe, *xc, *xc0;
615
391k
    ULong y;
616
391k
    ULLong carry, z;
617
618
391k
    if ((!a->x[0] && a->wds == 1) || (!b->x[0] && b->wds == 1)) {
619
3.96k
        c = Balloc(0);
620
3.96k
        if (c == NULL)
621
0
            return NULL;
622
3.96k
        c->wds = 1;
623
3.96k
        c->x[0] = 0;
624
3.96k
        return c;
625
3.96k
    }
626
627
387k
    if (a->wds < b->wds) {
628
192k
        c = a;
629
192k
        a = b;
630
192k
        b = c;
631
192k
    }
632
387k
    k = a->k;
633
387k
    wa = a->wds;
634
387k
    wb = b->wds;
635
387k
    wc = wa + wb;
636
387k
    if (wc > a->maxwds)
637
170k
        k++;
638
387k
    c = Balloc(k);
639
387k
    if (c == NULL)
640
0
        return NULL;
641
3.76M
    for(x = c->x, xa = x + wc; x < xa; x++)
642
3.37M
        *x = 0;
643
387k
    xa = a->x;
644
387k
    xae = xa + wa;
645
387k
    xb = b->x;
646
387k
    xbe = xb + wb;
647
387k
    xc0 = c->x;
648
1.23M
    for(; xb < xbe; xc0++) {
649
844k
        if ((y = *xb++)) {
650
838k
            x = xa;
651
838k
            xc = xc0;
652
838k
            carry = 0;
653
8.56M
            do {
654
8.56M
                z = *x++ * (ULLong)y + *xc + carry;
655
8.56M
                carry = z >> 32;
656
8.56M
                *xc++ = (ULong)(z & FFFFFFFF);
657
8.56M
            }
658
8.56M
            while(x < xae);
659
838k
            *xc = (ULong)carry;
660
838k
        }
661
844k
    }
662
663k
    for(xc0 = c->x, xc = xc0 + wc; wc > 0 && !*--xc; --wc) ;
663
387k
    c->wds = wc;
664
387k
    return c;
665
387k
}
666
667
#ifndef Py_USING_MEMORY_DEBUGGER
668
669
/* multiply the Bigint b by 5**k.  Returns a pointer to the result, or NULL on
670
   failure; if the returned pointer is distinct from b then the original
671
   Bigint b will have been Bfree'd.   Ignores the sign of b. */
672
673
static Bigint *
674
pow5mult(Bigint *b, int k)
675
121k
{
676
121k
    Bigint *b1, *p5, **p5s;
677
121k
    int i;
678
121k
    static const int p05[3] = { 5, 25, 125 };
679
680
    // For double-to-string conversion, the maximum value of k is limited by
681
    // DBL_MAX_10_EXP (308), the maximum decimal base-10 exponent for binary64.
682
    // For string-to-double conversion, the extreme case is constrained by our
683
    // hardcoded exponent limit before we underflow of -512, adjusted by
684
    // STRTOD_DIGLIM-DBL_DIG-1, giving a maximum of k=535.
685
121k
    assert(0 <= k && k < 1024);
686
687
121k
    if ((i = k & 3)) {
688
84.0k
        b = multadd(b, p05[i-1], 0);
689
84.0k
        if (b == NULL)
690
0
            return NULL;
691
84.0k
    }
692
693
121k
    if (!(k >>= 2))
694
10.5k
        return b;
695
110k
    PyInterpreterState *interp = _PyInterpreterState_GET();
696
110k
    p5s = interp->dtoa.p5s;
697
573k
    for(;;) {
698
573k
        assert(p5s != interp->dtoa.p5s + Bigint_Pow5size);
699
573k
        p5 = *p5s;
700
573k
        p5s++;
701
573k
        if (k & 1) {
702
330k
            b1 = mult(b, p5);
703
330k
            Bfree(b);
704
330k
            b = b1;
705
330k
            if (b == NULL)
706
0
                return NULL;
707
330k
        }
708
573k
        if (!(k >>= 1))
709
110k
            break;
710
573k
    }
711
110k
    return b;
712
110k
}
713
714
#else
715
716
/* Version of pow5mult that doesn't cache powers of 5. Provided for
717
   the benefit of memory debugging tools like Valgrind. */
718
719
static Bigint *
720
pow5mult(Bigint *b, int k)
721
{
722
    Bigint *b1, *p5, *p51;
723
    int i;
724
    static const int p05[3] = { 5, 25, 125 };
725
726
    if ((i = k & 3)) {
727
        b = multadd(b, p05[i-1], 0);
728
        if (b == NULL)
729
            return NULL;
730
    }
731
732
    if (!(k >>= 2))
733
        return b;
734
    p5 = i2b(625);
735
    if (p5 == NULL) {
736
        Bfree(b);
737
        return NULL;
738
    }
739
740
    for(;;) {
741
        if (k & 1) {
742
            b1 = mult(b, p5);
743
            Bfree(b);
744
            b = b1;
745
            if (b == NULL) {
746
                Bfree(p5);
747
                return NULL;
748
            }
749
        }
750
        if (!(k >>= 1))
751
            break;
752
        p51 = mult(p5, p5);
753
        Bfree(p5);
754
        p5 = p51;
755
        if (p5 == NULL) {
756
            Bfree(b);
757
            return NULL;
758
        }
759
    }
760
    Bfree(p5);
761
    return b;
762
}
763
764
#endif /* Py_USING_MEMORY_DEBUGGER */
765
766
/* shift a Bigint b left by k bits.  Return a pointer to the shifted result,
767
   or NULL on failure.  If the returned pointer is distinct from b then the
768
   original b will have been Bfree'd.   Ignores the sign of b. */
769
770
static Bigint *
771
lshift(Bigint *b, int k)
772
298k
{
773
298k
    int i, k1, n, n1;
774
298k
    Bigint *b1;
775
298k
    ULong *x, *x1, *xe, z;
776
777
298k
    if (!k || (!b->x[0] && b->wds == 1))
778
4.38k
        return b;
779
780
294k
    n = k >> 5;
781
294k
    k1 = b->k;
782
294k
    n1 = n + b->wds + 1;
783
750k
    for(i = b->maxwds; n1 > i; i <<= 1)
784
456k
        k1++;
785
294k
    b1 = Balloc(k1);
786
294k
    if (b1 == NULL) {
787
0
        Bfree(b);
788
0
        return NULL;
789
0
    }
790
294k
    x1 = b1->x;
791
1.86M
    for(i = 0; i < n; i++)
792
1.56M
        *x1++ = 0;
793
294k
    x = b->x;
794
294k
    xe = x + b->wds;
795
294k
    if (k &= 0x1f) {
796
292k
        k1 = 32 - k;
797
292k
        z = 0;
798
1.51M
        do {
799
1.51M
            *x1++ = *x << k | z;
800
1.51M
            z = *x++ >> k1;
801
1.51M
        }
802
1.51M
        while(x < xe);
803
292k
        if ((*x1 = z))
804
48.7k
            ++n1;
805
292k
    }
806
1.81k
    else do
807
3.80k
             *x1++ = *x++;
808
3.80k
        while(x < xe);
809
294k
    b1->wds = n1 - 1;
810
294k
    Bfree(b);
811
294k
    return b1;
812
294k
}
813
814
/* Do a three-way compare of a and b, returning -1 if a < b, 0 if a == b and
815
   1 if a > b.  Ignores signs of a and b. */
816
817
static int
818
cmp(Bigint *a, Bigint *b)
819
843k
{
820
843k
    ULong *xa, *xa0, *xb, *xb0;
821
843k
    int i, j;
822
823
843k
    i = a->wds;
824
843k
    j = b->wds;
825
#ifdef DEBUG
826
    if (i > 1 && !a->x[i-1])
827
        Bug("cmp called with a->x[a->wds-1] == 0");
828
    if (j > 1 && !b->x[j-1])
829
        Bug("cmp called with b->x[b->wds-1] == 0");
830
#endif
831
843k
    if (i -= j)
832
170k
        return i;
833
673k
    xa0 = a->x;
834
673k
    xa = xa0 + j;
835
673k
    xb0 = b->x;
836
673k
    xb = xb0 + j;
837
824k
    for(;;) {
838
824k
        if (*--xa != *--xb)
839
656k
            return *xa < *xb ? -1 : 1;
840
167k
        if (xa <= xa0)
841
17.2k
            break;
842
167k
    }
843
17.2k
    return 0;
844
673k
}
845
846
/* Take the difference of Bigints a and b, returning a new Bigint.  Returns
847
   NULL on failure.  The signs of a and b are ignored, but the sign of the
848
   result is set appropriately. */
849
850
static Bigint *
851
diff(Bigint *a, Bigint *b)
852
215k
{
853
215k
    Bigint *c;
854
215k
    int i, wa, wb;
855
215k
    ULong *xa, *xae, *xb, *xbe, *xc;
856
215k
    ULLong borrow, y;
857
858
215k
    i = cmp(a,b);
859
215k
    if (!i) {
860
3.37k
        c = Balloc(0);
861
3.37k
        if (c == NULL)
862
0
            return NULL;
863
3.37k
        c->wds = 1;
864
3.37k
        c->x[0] = 0;
865
3.37k
        return c;
866
3.37k
    }
867
211k
    if (i < 0) {
868
43.7k
        c = a;
869
43.7k
        a = b;
870
43.7k
        b = c;
871
43.7k
        i = 1;
872
43.7k
    }
873
168k
    else
874
168k
        i = 0;
875
211k
    c = Balloc(a->k);
876
211k
    if (c == NULL)
877
0
        return NULL;
878
211k
    c->sign = i;
879
211k
    wa = a->wds;
880
211k
    xa = a->x;
881
211k
    xae = xa + wa;
882
211k
    wb = b->wds;
883
211k
    xb = b->x;
884
211k
    xbe = xb + wb;
885
211k
    xc = c->x;
886
211k
    borrow = 0;
887
1.75M
    do {
888
1.75M
        y = (ULLong)*xa++ - *xb++ - borrow;
889
1.75M
        borrow = y >> 32 & (ULong)1;
890
1.75M
        *xc++ = (ULong)(y & FFFFFFFF);
891
1.75M
    }
892
1.75M
    while(xb < xbe);
893
408k
    while(xa < xae) {
894
196k
        y = *xa++ - borrow;
895
196k
        borrow = y >> 32 & (ULong)1;
896
196k
        *xc++ = (ULong)(y & FFFFFFFF);
897
196k
    }
898
331k
    while(!*--xc)
899
119k
        wa--;
900
211k
    c->wds = wa;
901
211k
    return c;
902
211k
}
903
904
/* Given a positive normal double x, return the difference between x and the
905
   next double up.  Doesn't give correct results for subnormals. */
906
907
static double
908
ulp(U *x)
909
45.2k
{
910
45.2k
    Long L;
911
45.2k
    U u;
912
913
45.2k
    L = (word0(x) & Exp_mask) - (P-1)*Exp_msk1;
914
45.2k
    word0(&u) = L;
915
45.2k
    word1(&u) = 0;
916
45.2k
    return dval(&u);
917
45.2k
}
918
919
/* Convert a Bigint to a double plus an exponent */
920
921
static double
922
b2d(Bigint *a, int *e)
923
84.2k
{
924
84.2k
    ULong *xa, *xa0, w, y, z;
925
84.2k
    int k;
926
84.2k
    U d;
927
928
84.2k
    xa0 = a->x;
929
84.2k
    xa = xa0 + a->wds;
930
84.2k
    y = *--xa;
931
#ifdef DEBUG
932
    if (!y) Bug("zero y in b2d");
933
#endif
934
84.2k
    k = hi0bits(y);
935
84.2k
    *e = 32 - k;
936
84.2k
    if (k < Ebits) {
937
23.0k
        word0(&d) = Exp_1 | y >> (Ebits - k);
938
23.0k
        w = xa > xa0 ? *--xa : 0;
939
23.0k
        word1(&d) = y << ((32-Ebits) + k) | w >> (Ebits - k);
940
23.0k
        goto ret_d;
941
23.0k
    }
942
61.1k
    z = xa > xa0 ? *--xa : 0;
943
61.1k
    if (k -= Ebits) {
944
57.1k
        word0(&d) = Exp_1 | y << k | z >> (32 - k);
945
57.1k
        y = xa > xa0 ? *--xa : 0;
946
57.1k
        word1(&d) = z << k | y >> (32 - k);
947
57.1k
    }
948
4.03k
    else {
949
4.03k
        word0(&d) = Exp_1 | y;
950
4.03k
        word1(&d) = z;
951
4.03k
    }
952
84.2k
  ret_d:
953
84.2k
    return dval(&d);
954
61.1k
}
955
956
/* Convert a scaled double to a Bigint plus an exponent.  Similar to d2b,
957
   except that it accepts the scale parameter used in _Py_dg_strtod (which
958
   should be either 0 or 2*P), and the normalization for the return value is
959
   different (see below).  On input, d should be finite and nonnegative, and d
960
   / 2**scale should be exactly representable as an IEEE 754 double.
961
962
   Returns a Bigint b and an integer e such that
963
964
     dval(d) / 2**scale = b * 2**e.
965
966
   Unlike d2b, b is not necessarily odd: b and e are normalized so
967
   that either 2**(P-1) <= b < 2**P and e >= Etiny, or b < 2**P
968
   and e == Etiny.  This applies equally to an input of 0.0: in that
969
   case the return values are b = 0 and e = Etiny.
970
971
   The above normalization ensures that for all possible inputs d,
972
   2**e gives ulp(d/2**scale).
973
974
   Returns NULL on failure.
975
*/
976
977
static Bigint *
978
sd2b(U *d, int scale, int *e)
979
101k
{
980
101k
    Bigint *b;
981
982
101k
    b = Balloc(1);
983
101k
    if (b == NULL)
984
0
        return NULL;
985
986
    /* First construct b and e assuming that scale == 0. */
987
101k
    b->wds = 2;
988
101k
    b->x[0] = word1(d);
989
101k
    b->x[1] = word0(d) & Frac_mask;
990
101k
    *e = Etiny - 1 + (int)((word0(d) & Exp_mask) >> Exp_shift);
991
101k
    if (*e < Etiny)
992
4.38k
        *e = Etiny;
993
96.7k
    else
994
96.7k
        b->x[1] |= Exp_msk1;
995
996
    /* Now adjust for scale, provided that b != 0. */
997
101k
    if (scale && (b->x[0] || b->x[1])) {
998
30.6k
        *e -= scale;
999
30.6k
        if (*e < Etiny) {
1000
25.7k
            scale = Etiny - *e;
1001
25.7k
            *e = Etiny;
1002
            /* We can't shift more than P-1 bits without shifting out a 1. */
1003
25.7k
            assert(0 < scale && scale <= P - 1);
1004
25.7k
            if (scale >= 32) {
1005
                /* The bits shifted out should all be zero. */
1006
11.7k
                assert(b->x[0] == 0);
1007
11.7k
                b->x[0] = b->x[1];
1008
11.7k
                b->x[1] = 0;
1009
11.7k
                scale -= 32;
1010
11.7k
            }
1011
25.7k
            if (scale) {
1012
                /* The bits shifted out should all be zero. */
1013
23.9k
                assert(b->x[0] << (32 - scale) == 0);
1014
23.9k
                b->x[0] = (b->x[0] >> scale) | (b->x[1] << (32 - scale));
1015
23.9k
                b->x[1] >>= scale;
1016
23.9k
            }
1017
25.7k
        }
1018
30.6k
    }
1019
    /* Ensure b is normalized. */
1020
101k
    if (!b->x[1])
1021
19.6k
        b->wds = 1;
1022
1023
101k
    return b;
1024
101k
}
1025
1026
/* Convert a double to a Bigint plus an exponent.  Return NULL on failure.
1027
1028
   Given a finite nonzero double d, return an odd Bigint b and exponent *e
1029
   such that fabs(d) = b * 2**e.  On return, *bbits gives the number of
1030
   significant bits of b; that is, 2**(*bbits-1) <= b < 2**(*bbits).
1031
1032
   If d is zero, then b == 0, *e == -1010, *bbits = 0.
1033
 */
1034
1035
static Bigint *
1036
d2b(U *d, int *e, int *bits)
1037
39.9k
{
1038
39.9k
    Bigint *b;
1039
39.9k
    int de, k;
1040
39.9k
    ULong *x, y, z;
1041
39.9k
    int i;
1042
1043
39.9k
    b = Balloc(1);
1044
39.9k
    if (b == NULL)
1045
0
        return NULL;
1046
39.9k
    x = b->x;
1047
1048
39.9k
    z = word0(d) & Frac_mask;
1049
39.9k
    word0(d) &= 0x7fffffff;   /* clear sign bit, which we ignore */
1050
39.9k
    if ((de = (int)(word0(d) >> Exp_shift)))
1051
35.6k
        z |= Exp_msk1;
1052
39.9k
    if ((y = word1(d))) {
1053
29.5k
        if ((k = lo0bits(&y))) {
1054
18.6k
            x[0] = y | z << (32 - k);
1055
18.6k
            z >>= k;
1056
18.6k
        }
1057
10.9k
        else
1058
10.9k
            x[0] = y;
1059
29.5k
        i =
1060
29.5k
            b->wds = (x[1] = z) ? 2 : 1;
1061
29.5k
    }
1062
10.3k
    else {
1063
10.3k
        k = lo0bits(&z);
1064
10.3k
        x[0] = z;
1065
10.3k
        i =
1066
10.3k
            b->wds = 1;
1067
10.3k
        k += 32;
1068
10.3k
    }
1069
39.9k
    if (de) {
1070
35.6k
        *e = de - Bias - (P-1) + k;
1071
35.6k
        *bits = P - k;
1072
35.6k
    }
1073
4.30k
    else {
1074
4.30k
        *e = de - Bias - (P-1) + 1 + k;
1075
4.30k
        *bits = 32*i - hi0bits(x[i-1]);
1076
4.30k
    }
1077
39.9k
    return b;
1078
39.9k
}
1079
1080
/* Compute the ratio of two Bigints, as a double.  The result may have an
1081
   error of up to 2.5 ulps. */
1082
1083
static double
1084
ratio(Bigint *a, Bigint *b)
1085
42.1k
{
1086
42.1k
    U da, db;
1087
42.1k
    int k, ka, kb;
1088
1089
42.1k
    dval(&da) = b2d(a, &ka);
1090
42.1k
    dval(&db) = b2d(b, &kb);
1091
42.1k
    k = ka - kb + 32*(a->wds - b->wds);
1092
42.1k
    if (k > 0)
1093
23.4k
        word0(&da) += k*Exp_msk1;
1094
18.7k
    else {
1095
18.7k
        k = -k;
1096
18.7k
        word0(&db) += k*Exp_msk1;
1097
18.7k
    }
1098
42.1k
    return dval(&da) / dval(&db);
1099
42.1k
}
1100
1101
static const double
1102
tens[] = {
1103
    1e0, 1e1, 1e2, 1e3, 1e4, 1e5, 1e6, 1e7, 1e8, 1e9,
1104
    1e10, 1e11, 1e12, 1e13, 1e14, 1e15, 1e16, 1e17, 1e18, 1e19,
1105
    1e20, 1e21, 1e22
1106
};
1107
1108
static const double
1109
bigtens[] = { 1e16, 1e32, 1e64, 1e128, 1e256 };
1110
static const double tinytens[] = { 1e-16, 1e-32, 1e-64, 1e-128,
1111
                                   9007199254740992.*9007199254740992.e-256
1112
                                   /* = 2^106 * 1e-256 */
1113
};
1114
/* The factor of 2^53 in tinytens[4] helps us avoid setting the underflow */
1115
/* flag unnecessarily.  It leads to a song and dance at the end of strtod. */
1116
23.2k
#define Scale_Bit 0x10
1117
24.7k
#define n_bigtens 5
1118
1119
#define ULbits 32
1120
#define kshift 5
1121
37.0k
#define kmask 31
1122
1123
1124
static int
1125
dshift(Bigint *b, int p2)
1126
37.0k
{
1127
37.0k
    int rv = hi0bits(b->x[b->wds-1]) - 4;
1128
37.0k
    if (p2 > 0)
1129
21.8k
        rv -= p2;
1130
37.0k
    return rv & kmask;
1131
37.0k
}
1132
1133
/* special case of Bigint division.  The quotient is always in the range 0 <=
1134
   quotient < 10, and on entry the divisor S is normalized so that its top 4
1135
   bits (28--31) are zero and bit 27 is set. */
1136
1137
static int
1138
quorem(Bigint *b, Bigint *S)
1139
265k
{
1140
265k
    int n;
1141
265k
    ULong *bx, *bxe, q, *sx, *sxe;
1142
265k
    ULLong borrow, carry, y, ys;
1143
1144
265k
    n = S->wds;
1145
#ifdef DEBUG
1146
    /*debug*/ if (b->wds > n)
1147
        /*debug*/       Bug("oversize b in quorem");
1148
#endif
1149
265k
    if (b->wds < n)
1150
7.26k
        return 0;
1151
258k
    sx = S->x;
1152
258k
    sxe = sx + --n;
1153
258k
    bx = b->x;
1154
258k
    bxe = bx + n;
1155
258k
    q = *bxe / (*sxe + 1);      /* ensure q <= true quotient */
1156
#ifdef DEBUG
1157
    /*debug*/ if (q > 9)
1158
        /*debug*/       Bug("oversized quotient in quorem");
1159
#endif
1160
258k
    if (q) {
1161
212k
        borrow = 0;
1162
212k
        carry = 0;
1163
1.29M
        do {
1164
1.29M
            ys = *sx++ * (ULLong)q + carry;
1165
1.29M
            carry = ys >> 32;
1166
1.29M
            y = *bx - (ys & FFFFFFFF) - borrow;
1167
1.29M
            borrow = y >> 32 & (ULong)1;
1168
1.29M
            *bx++ = (ULong)(y & FFFFFFFF);
1169
1.29M
        }
1170
1.29M
        while(sx <= sxe);
1171
212k
        if (!*bxe) {
1172
850
            bx = b->x;
1173
850
            while(--bxe > bx && !*bxe)
1174
0
                --n;
1175
850
            b->wds = n;
1176
850
        }
1177
212k
    }
1178
258k
    if (cmp(b, S) >= 0) {
1179
15.8k
        q++;
1180
15.8k
        borrow = 0;
1181
15.8k
        carry = 0;
1182
15.8k
        bx = b->x;
1183
15.8k
        sx = S->x;
1184
102k
        do {
1185
102k
            ys = *sx++ + carry;
1186
102k
            carry = ys >> 32;
1187
102k
            y = *bx - (ys & FFFFFFFF) - borrow;
1188
102k
            borrow = y >> 32 & (ULong)1;
1189
102k
            *bx++ = (ULong)(y & FFFFFFFF);
1190
102k
        }
1191
102k
        while(sx <= sxe);
1192
15.8k
        bx = b->x;
1193
15.8k
        bxe = bx + n;
1194
15.8k
        if (!*bxe) {
1195
16.4k
            while(--bxe > bx && !*bxe)
1196
1.40k
                --n;
1197
15.0k
            b->wds = n;
1198
15.0k
        }
1199
15.8k
    }
1200
258k
    return q;
1201
265k
}
1202
1203
/* sulp(x) is a version of ulp(x) that takes bc.scale into account.
1204
1205
   Assuming that x is finite and nonnegative (positive zero is fine
1206
   here) and x / 2^bc.scale is exactly representable as a double,
1207
   sulp(x) is equivalent to 2^bc.scale * ulp(x / 2^bc.scale). */
1208
1209
static double
1210
sulp(U *x, BCinfo *bc)
1211
3.36k
{
1212
3.36k
    U u;
1213
1214
3.36k
    if (bc->scale && 2*P + 1 > (int)((word0(x) & Exp_mask) >> Exp_shift)) {
1215
        /* rv/2^bc->scale is subnormal */
1216
277
        word0(&u) = (P+2)*Exp_msk1;
1217
277
        word1(&u) = 0;
1218
277
        return u.d;
1219
277
    }
1220
3.08k
    else {
1221
3.08k
        assert(word0(x) || word1(x)); /* x != 0.0 */
1222
3.08k
        return ulp(x);
1223
3.08k
    }
1224
3.36k
}
1225
1226
/* The bigcomp function handles some hard cases for strtod, for inputs
1227
   with more than STRTOD_DIGLIM digits.  It's called once an initial
1228
   estimate for the double corresponding to the input string has
1229
   already been obtained by the code in _Py_dg_strtod.
1230
1231
   The bigcomp function is only called after _Py_dg_strtod has found a
1232
   double value rv such that either rv or rv + 1ulp represents the
1233
   correctly rounded value corresponding to the original string.  It
1234
   determines which of these two values is the correct one by
1235
   computing the decimal digits of rv + 0.5ulp and comparing them with
1236
   the corresponding digits of s0.
1237
1238
   In the following, write dv for the absolute value of the number represented
1239
   by the input string.
1240
1241
   Inputs:
1242
1243
     s0 points to the first significant digit of the input string.
1244
1245
     rv is a (possibly scaled) estimate for the closest double value to the
1246
        value represented by the original input to _Py_dg_strtod.  If
1247
        bc->scale is nonzero, then rv/2^(bc->scale) is the approximation to
1248
        the input value.
1249
1250
     bc is a struct containing information gathered during the parsing and
1251
        estimation steps of _Py_dg_strtod.  Description of fields follows:
1252
1253
        bc->e0 gives the exponent of the input value, such that dv = (integer
1254
           given by the bd->nd digits of s0) * 10**e0
1255
1256
        bc->nd gives the total number of significant digits of s0.  It will
1257
           be at least 1.
1258
1259
        bc->nd0 gives the number of significant digits of s0 before the
1260
           decimal separator.  If there's no decimal separator, bc->nd0 ==
1261
           bc->nd.
1262
1263
        bc->scale is the value used to scale rv to avoid doing arithmetic with
1264
           subnormal values.  It's either 0 or 2*P (=106).
1265
1266
   Outputs:
1267
1268
     On successful exit, rv/2^(bc->scale) is the closest double to dv.
1269
1270
     Returns 0 on success, -1 on failure (e.g., due to a failed malloc call). */
1271
1272
static int
1273
bigcomp(U *rv, const char *s0, BCinfo *bc)
1274
7.44k
{
1275
7.44k
    Bigint *b, *d;
1276
7.44k
    int b2, d2, dd, i, nd, nd0, odd, p2, p5;
1277
1278
7.44k
    nd = bc->nd;
1279
7.44k
    nd0 = bc->nd0;
1280
7.44k
    p5 = nd + bc->e0;
1281
7.44k
    b = sd2b(rv, bc->scale, &p2);
1282
7.44k
    if (b == NULL)
1283
0
        return -1;
1284
1285
    /* record whether the lsb of rv/2^(bc->scale) is odd:  in the exact halfway
1286
       case, this is used for round to even. */
1287
7.44k
    odd = b->x[0] & 1;
1288
1289
    /* left shift b by 1 bit and or a 1 into the least significant bit;
1290
       this gives us b * 2**p2 = rv/2^(bc->scale) + 0.5 ulp. */
1291
7.44k
    b = lshift(b, 1);
1292
7.44k
    if (b == NULL)
1293
0
        return -1;
1294
7.44k
    b->x[0] |= 1;
1295
7.44k
    p2--;
1296
1297
7.44k
    p2 -= p5;
1298
7.44k
    d = i2b(1);
1299
7.44k
    if (d == NULL) {
1300
0
        Bfree(b);
1301
0
        return -1;
1302
0
    }
1303
    /* Arrange for convenient computation of quotients:
1304
     * shift left if necessary so divisor has 4 leading 0 bits.
1305
     */
1306
7.44k
    if (p5 > 0) {
1307
5.50k
        d = pow5mult(d, p5);
1308
5.50k
        if (d == NULL) {
1309
0
            Bfree(b);
1310
0
            return -1;
1311
0
        }
1312
5.50k
    }
1313
1.94k
    else if (p5 < 0) {
1314
1.56k
        b = pow5mult(b, -p5);
1315
1.56k
        if (b == NULL) {
1316
0
            Bfree(d);
1317
0
            return -1;
1318
0
        }
1319
1.56k
    }
1320
7.44k
    if (p2 > 0) {
1321
4.12k
        b2 = p2;
1322
4.12k
        d2 = 0;
1323
4.12k
    }
1324
3.32k
    else {
1325
3.32k
        b2 = 0;
1326
3.32k
        d2 = -p2;
1327
3.32k
    }
1328
7.44k
    i = dshift(d, d2);
1329
7.44k
    if ((b2 += i) > 0) {
1330
7.17k
        b = lshift(b, b2);
1331
7.17k
        if (b == NULL) {
1332
0
            Bfree(d);
1333
0
            return -1;
1334
0
        }
1335
7.17k
    }
1336
7.44k
    if ((d2 += i) > 0) {
1337
6.88k
        d = lshift(d, d2);
1338
6.88k
        if (d == NULL) {
1339
0
            Bfree(b);
1340
0
            return -1;
1341
0
        }
1342
6.88k
    }
1343
1344
    /* Compare s0 with b/d: set dd to -1, 0, or 1 according as s0 < b/d, s0 ==
1345
     * b/d, or s0 > b/d.  Here the digits of s0 are thought of as representing
1346
     * a number in the range [0.1, 1). */
1347
7.44k
    if (cmp(b, d) >= 0)
1348
        /* b/d >= 1 */
1349
927
        dd = -1;
1350
6.51k
    else {
1351
6.51k
        i = 0;
1352
143k
        for(;;) {
1353
143k
            b = multadd(b, 10, 0);
1354
143k
            if (b == NULL) {
1355
0
                Bfree(d);
1356
0
                return -1;
1357
0
            }
1358
143k
            dd = s0[i < nd0 ? i : i+1] - '0' - quorem(b, d);
1359
143k
            i++;
1360
1361
143k
            if (dd)
1362
4.50k
                break;
1363
139k
            if (!b->x[0] && b->wds == 1) {
1364
                /* b/d == 0 */
1365
1.04k
                dd = i < nd;
1366
1.04k
                break;
1367
1.04k
            }
1368
138k
            if (!(i < nd)) {
1369
                /* b/d != 0, but digits of s0 exhausted */
1370
976
                dd = -1;
1371
976
                break;
1372
976
            }
1373
138k
        }
1374
6.51k
    }
1375
7.44k
    Bfree(b);
1376
7.44k
    Bfree(d);
1377
7.44k
    if (dd > 0 || (dd == 0 && odd))
1378
1.66k
        dval(rv) += sulp(rv, bc);
1379
7.44k
    return 0;
1380
7.44k
}
1381
1382
1383
double
1384
_Py_dg_strtod(const char *s00, char **se)
1385
608k
{
1386
608k
    int bb2, bb5, bbe, bd2, bd5, bs2, c, dsign, e, e1, error;
1387
608k
    int esign, i, j, k, lz, nd, nd0, odd, sign;
1388
608k
    const char *s, *s0, *s1;
1389
608k
    double aadj, aadj1;
1390
608k
    U aadj2, adj, rv, rv0;
1391
608k
    ULong y, z, abs_exp;
1392
608k
    Long L;
1393
608k
    BCinfo bc;
1394
608k
    Bigint *bb = NULL, *bd = NULL, *bd0 = NULL, *bs = NULL, *delta = NULL;
1395
608k
    size_t ndigits, fraclen;
1396
608k
    double result;
1397
1398
608k
    dval(&rv) = 0.;
1399
1400
    /* Start parsing. */
1401
608k
    c = *(s = s00);
1402
1403
    /* Parse optional sign, if present. */
1404
608k
    sign = 0;
1405
608k
    switch (c) {
1406
423k
    case '-':
1407
423k
        sign = 1;
1408
423k
        _Py_FALLTHROUGH;
1409
423k
    case '+':
1410
423k
        c = *++s;
1411
608k
    }
1412
1413
    /* Skip leading zeros: lz is true iff there were leading zeros. */
1414
608k
    s1 = s;
1415
624k
    while (c == '0')
1416
16.3k
        c = *++s;
1417
608k
    lz = s != s1;
1418
1419
    /* Point s0 at the first nonzero digit (if any).  fraclen will be the
1420
       number of digits between the decimal point and the end of the
1421
       digit string.  ndigits will be the total number of digits ignoring
1422
       leading zeros. */
1423
608k
    s0 = s1 = s;
1424
2.84M
    while ('0' <= c && c <= '9')
1425
2.23M
        c = *++s;
1426
608k
    ndigits = s - s1;
1427
608k
    fraclen = 0;
1428
1429
    /* Parse decimal point and following digits. */
1430
608k
    if (c == '.') {
1431
56.8k
        c = *++s;
1432
56.8k
        if (!ndigits) {
1433
18.3k
            s1 = s;
1434
158k
            while (c == '0')
1435
139k
                c = *++s;
1436
18.3k
            lz = lz || s != s1;
1437
18.3k
            fraclen += (s - s1);
1438
18.3k
            s0 = s;
1439
18.3k
        }
1440
56.8k
        s1 = s;
1441
15.0M
        while ('0' <= c && c <= '9')
1442
14.9M
            c = *++s;
1443
56.8k
        ndigits += s - s1;
1444
56.8k
        fraclen += s - s1;
1445
56.8k
    }
1446
1447
    /* Now lz is true if and only if there were leading zero digits, and
1448
       ndigits gives the total number of digits ignoring leading zeros.  A
1449
       valid input must have at least one digit. */
1450
608k
    if (!ndigits && !lz) {
1451
55.9k
        if (se)
1452
55.9k
            *se = (char *)s00;
1453
55.9k
        goto parse_error;
1454
55.9k
    }
1455
1456
    /* Range check ndigits and fraclen to make sure that they, and values
1457
       computed with them, can safely fit in an int. */
1458
552k
    if (ndigits > MAX_DIGITS || fraclen > MAX_DIGITS) {
1459
0
        if (se)
1460
0
            *se = (char *)s00;
1461
0
        goto parse_error;
1462
0
    }
1463
552k
    nd = (int)ndigits;
1464
552k
    nd0 = (int)ndigits - (int)fraclen;
1465
1466
    /* Parse exponent. */
1467
552k
    e = 0;
1468
552k
    if (c == 'e' || c == 'E') {
1469
491k
        s00 = s;
1470
491k
        c = *++s;
1471
1472
        /* Exponent sign. */
1473
491k
        esign = 0;
1474
491k
        switch (c) {
1475
26.8k
        case '-':
1476
26.8k
            esign = 1;
1477
26.8k
            _Py_FALLTHROUGH;
1478
39.2k
        case '+':
1479
39.2k
            c = *++s;
1480
491k
        }
1481
1482
        /* Skip zeros.  lz is true iff there are leading zeros. */
1483
491k
        s1 = s;
1484
515k
        while (c == '0')
1485
24.2k
            c = *++s;
1486
491k
        lz = s != s1;
1487
1488
        /* Get absolute value of the exponent. */
1489
491k
        s1 = s;
1490
491k
        abs_exp = 0;
1491
2.05M
        while ('0' <= c && c <= '9') {
1492
1.55M
            abs_exp = 10*abs_exp + (c - '0');
1493
1.55M
            c = *++s;
1494
1.55M
        }
1495
1496
        /* abs_exp will be correct modulo 2**32.  But 10**9 < 2**32, so if
1497
           there are at most 9 significant exponent digits then overflow is
1498
           impossible. */
1499
491k
        if (s - s1 > 9 || abs_exp > MAX_ABS_EXP)
1500
3.48k
            e = (int)MAX_ABS_EXP;
1501
487k
        else
1502
487k
            e = (int)abs_exp;
1503
491k
        if (esign)
1504
26.8k
            e = -e;
1505
1506
        /* A valid exponent must have at least one digit. */
1507
491k
        if (s == s1 && !lz)
1508
0
            s = s00;
1509
491k
    }
1510
1511
    /* Adjust exponent to take into account position of the point. */
1512
552k
    e -= nd - nd0;
1513
552k
    if (nd0 <= 0)
1514
22.6k
        nd0 = nd;
1515
1516
    /* Finished parsing.  Set se to indicate how far we parsed */
1517
552k
    if (se)
1518
552k
        *se = (char *)s;
1519
1520
    /* If all digits were zero, exit with return value +-0.0.  Otherwise,
1521
       strip trailing zeros: scan back until we hit a nonzero digit. */
1522
552k
    if (!nd)
1523
9.86k
        goto ret;
1524
2.33M
    for (i = nd; i > 0; ) {
1525
2.33M
        --i;
1526
2.33M
        if (s0[i < nd0 ? i : i+1] != '0') {
1527
542k
            ++i;
1528
542k
            break;
1529
542k
        }
1530
2.33M
    }
1531
542k
    e += nd - i;
1532
542k
    nd = i;
1533
542k
    if (nd0 > nd)
1534
7.83k
        nd0 = nd;
1535
1536
    /* Summary of parsing results.  After parsing, and dealing with zero
1537
     * inputs, we have values s0, nd0, nd, e, sign, where:
1538
     *
1539
     *  - s0 points to the first significant digit of the input string
1540
     *
1541
     *  - nd is the total number of significant digits (here, and
1542
     *    below, 'significant digits' means the set of digits of the
1543
     *    significand of the input that remain after ignoring leading
1544
     *    and trailing zeros).
1545
     *
1546
     *  - nd0 indicates the position of the decimal point, if present; it
1547
     *    satisfies 1 <= nd0 <= nd.  The nd significant digits are in
1548
     *    s0[0:nd0] and s0[nd0+1:nd+1] using the usual Python half-open slice
1549
     *    notation.  (If nd0 < nd, then s0[nd0] contains a '.'  character; if
1550
     *    nd0 == nd, then s0[nd0] could be any non-digit character.)
1551
     *
1552
     *  - e is the adjusted exponent: the absolute value of the number
1553
     *    represented by the original input string is n * 10**e, where
1554
     *    n is the integer represented by the concatenation of
1555
     *    s0[0:nd0] and s0[nd0+1:nd+1]
1556
     *
1557
     *  - sign gives the sign of the input:  1 for negative, 0 for positive
1558
     *
1559
     *  - the first and last significant digits are nonzero
1560
     */
1561
1562
    /* put first DBL_DIG+1 digits into integer y and z.
1563
     *
1564
     *  - y contains the value represented by the first min(9, nd)
1565
     *    significant digits
1566
     *
1567
     *  - if nd > 9, z contains the value represented by significant digits
1568
     *    with indices in [9, min(16, nd)).  So y * 10**(min(16, nd) - 9) + z
1569
     *    gives the value represented by the first min(16, nd) sig. digits.
1570
     */
1571
1572
542k
    bc.e0 = e1 = e;
1573
542k
    y = z = 0;
1574
1.69M
    for (i = 0; i < nd; i++) {
1575
1.17M
        if (i < 9)
1576
925k
            y = 10*y + s0[i < nd0 ? i : i+1] - '0';
1577
249k
        else if (i < DBL_DIG+1)
1578
226k
            z = 10*z + s0[i < nd0 ? i : i+1] - '0';
1579
22.9k
        else
1580
22.9k
            break;
1581
1.17M
    }
1582
1583
542k
    k = nd < DBL_DIG + 1 ? nd : DBL_DIG + 1;
1584
542k
    dval(&rv) = y;
1585
542k
    if (k > 9) {
1586
39.1k
        dval(&rv) = tens[k - 9] * dval(&rv) + z;
1587
39.1k
    }
1588
542k
    if (nd <= DBL_DIG
1589
512k
        && Flt_Rounds == 1
1590
542k
        ) {
1591
512k
        if (!e)
1592
14.0k
            goto ret;
1593
498k
        if (e > 0) {
1594
457k
            if (e <= Ten_pmax) {
1595
22.6k
                dval(&rv) *= tens[e];
1596
22.6k
                goto ret;
1597
22.6k
            }
1598
434k
            i = DBL_DIG - nd;
1599
434k
            if (e <= Ten_pmax + i) {
1600
                /* A fancier test would sometimes let us do
1601
                 * this for larger i values.
1602
                 */
1603
3.09k
                e -= i;
1604
3.09k
                dval(&rv) *= tens[i];
1605
3.09k
                dval(&rv) *= tens[e];
1606
3.09k
                goto ret;
1607
3.09k
            }
1608
434k
        }
1609
41.3k
        else if (e >= -Ten_pmax) {
1610
22.0k
            dval(&rv) /= tens[-e];
1611
22.0k
            goto ret;
1612
22.0k
        }
1613
498k
    }
1614
480k
    e1 += nd - k;
1615
1616
480k
    bc.scale = 0;
1617
1618
    /* Get starting approximation = rv * 10**e1 */
1619
1620
480k
    if (e1 > 0) {
1621
446k
        if ((i = e1 & 15))
1622
436k
            dval(&rv) *= tens[i];
1623
446k
        if (e1 &= ~15) {
1624
439k
            if (e1 > DBL_MAX_10_EXP)
1625
406k
                goto ovfl;
1626
32.3k
            e1 >>= 4;
1627
97.9k
            for(j = 0; e1 > 1; j++, e1 >>= 1)
1628
65.5k
                if (e1 & 1)
1629
33.1k
                    dval(&rv) *= bigtens[j];
1630
            /* The last multiplication could overflow. */
1631
32.3k
            word0(&rv) -= P*Exp_msk1;
1632
32.3k
            dval(&rv) *= bigtens[j];
1633
32.3k
            if ((z = word0(&rv) & Exp_mask)
1634
32.3k
                > Exp_msk1*(DBL_MAX_EXP+Bias-P))
1635
613
                goto ovfl;
1636
31.7k
            if (z > Exp_msk1*(DBL_MAX_EXP+Bias-1-P)) {
1637
                /* set to largest number */
1638
                /* (Can't trust DBL_MAX) */
1639
500
                word0(&rv) = Big0;
1640
500
                word1(&rv) = Big1;
1641
500
            }
1642
31.2k
            else
1643
31.2k
                word0(&rv) += P*Exp_msk1;
1644
31.7k
        }
1645
446k
    }
1646
34.3k
    else if (e1 < 0) {
1647
        /* The input decimal value lies in [10**e1, 10**(e1+16)).
1648
1649
           If e1 <= -512, underflow immediately.
1650
           If e1 <= -256, set bc.scale to 2*P.
1651
1652
           So for input value < 1e-256, bc.scale is always set;
1653
           for input value >= 1e-240, bc.scale is never set.
1654
           For input values in [1e-256, 1e-240), bc.scale may or may
1655
           not be set. */
1656
1657
31.3k
        e1 = -e1;
1658
31.3k
        if ((i = e1 & 15))
1659
27.4k
            dval(&rv) /= tens[i];
1660
31.3k
        if (e1 >>= 4) {
1661
24.7k
            if (e1 >= 1 << n_bigtens)
1662
1.44k
                goto undfl;
1663
23.2k
            if (e1 & Scale_Bit)
1664
18.5k
                bc.scale = 2*P;
1665
123k
            for(j = 0; e1 > 0; j++, e1 >>= 1)
1666
100k
                if (e1 & 1)
1667
57.5k
                    dval(&rv) *= tinytens[j];
1668
23.2k
            if (bc.scale && (j = 2*P + 1 - ((word0(&rv) & Exp_mask)
1669
18.5k
                                            >> Exp_shift)) > 0) {
1670
                /* scaled rv is denormal; clear j low bits */
1671
16.4k
                if (j >= 32) {
1672
8.59k
                    word1(&rv) = 0;
1673
8.59k
                    if (j >= 53)
1674
4.42k
                        word0(&rv) = (P+2)*Exp_msk1;
1675
4.17k
                    else
1676
4.17k
                        word0(&rv) &= 0xffffffff << (j-32);
1677
8.59k
                }
1678
7.88k
                else
1679
7.88k
                    word1(&rv) &= 0xffffffff << j;
1680
16.4k
            }
1681
23.2k
            if (!dval(&rv))
1682
0
                goto undfl;
1683
23.2k
        }
1684
31.3k
    }
1685
1686
    /* Now the hard part -- adjusting rv to the correct value.*/
1687
1688
    /* Put digits into bd: true value = bd * 10^e */
1689
1690
72.0k
    bc.nd = nd;
1691
72.0k
    bc.nd0 = nd0;       /* Only needed if nd > STRTOD_DIGLIM, but done here */
1692
                        /* to silence an erroneous warning about bc.nd0 */
1693
                        /* possibly not being initialized. */
1694
72.0k
    if (nd > STRTOD_DIGLIM) {
1695
        /* ASSERT(STRTOD_DIGLIM >= 18); 18 == one more than the */
1696
        /* minimum number of decimal digits to distinguish double values */
1697
        /* in IEEE arithmetic. */
1698
1699
        /* Truncate input to 18 significant digits, then discard any trailing
1700
           zeros on the result by updating nd, nd0, e and y suitably. (There's
1701
           no need to update z; it's not reused beyond this point.) */
1702
58.4k
        for (i = 18; i > 0; ) {
1703
            /* scan back until we hit a nonzero digit.  significant digit 'i'
1704
            is s0[i] if i < nd0, s0[i+1] if i >= nd0. */
1705
58.4k
            --i;
1706
58.4k
            if (s0[i < nd0 ? i : i+1] != '0') {
1707
10.6k
                ++i;
1708
10.6k
                break;
1709
10.6k
            }
1710
58.4k
        }
1711
10.6k
        e += nd - i;
1712
10.6k
        nd = i;
1713
10.6k
        if (nd0 > nd)
1714
7.95k
            nd0 = nd;
1715
10.6k
        if (nd < 9) { /* must recompute y */
1716
3.51k
            y = 0;
1717
17.1k
            for(i = 0; i < nd0; ++i)
1718
13.6k
                y = 10*y + s0[i] - '0';
1719
9.14k
            for(; i < nd; ++i)
1720
5.63k
                y = 10*y + s0[i+1] - '0';
1721
3.51k
        }
1722
10.6k
    }
1723
72.0k
    bd0 = s2b(s0, nd0, nd, y);
1724
72.0k
    if (bd0 == NULL)
1725
0
        goto failed_malloc;
1726
1727
    /* Notation for the comments below.  Write:
1728
1729
         - dv for the absolute value of the number represented by the original
1730
           decimal input string.
1731
1732
         - if we've truncated dv, write tdv for the truncated value.
1733
           Otherwise, set tdv == dv.
1734
1735
         - srv for the quantity rv/2^bc.scale; so srv is the current binary
1736
           approximation to tdv (and dv).  It should be exactly representable
1737
           in an IEEE 754 double.
1738
    */
1739
1740
93.6k
    for(;;) {
1741
1742
        /* This is the main correction loop for _Py_dg_strtod.
1743
1744
           We've got a decimal value tdv, and a floating-point approximation
1745
           srv=rv/2^bc.scale to tdv.  The aim is to determine whether srv is
1746
           close enough (i.e., within 0.5 ulps) to tdv, and to compute a new
1747
           approximation if not.
1748
1749
           To determine whether srv is close enough to tdv, compute integers
1750
           bd, bb and bs proportional to tdv, srv and 0.5 ulp(srv)
1751
           respectively, and then use integer arithmetic to determine whether
1752
           |tdv - srv| is less than, equal to, or greater than 0.5 ulp(srv).
1753
        */
1754
1755
93.6k
        bd = Balloc(bd0->k);
1756
93.6k
        if (bd == NULL) {
1757
0
            goto failed_malloc;
1758
0
        }
1759
93.6k
        Bcopy(bd, bd0);
1760
93.6k
        bb = sd2b(&rv, bc.scale, &bbe);   /* srv = bb * 2^bbe */
1761
93.6k
        if (bb == NULL) {
1762
0
            goto failed_malloc;
1763
0
        }
1764
        /* Record whether lsb of bb is odd, in case we need this
1765
           for the round-to-even step later. */
1766
93.6k
        odd = bb->x[0] & 1;
1767
1768
        /* tdv = bd * 10**e;  srv = bb * 2**bbe */
1769
93.6k
        bs = i2b(1);
1770
93.6k
        if (bs == NULL) {
1771
0
            goto failed_malloc;
1772
0
        }
1773
1774
93.6k
        if (e >= 0) {
1775
45.3k
            bb2 = bb5 = 0;
1776
45.3k
            bd2 = bd5 = e;
1777
45.3k
        }
1778
48.3k
        else {
1779
48.3k
            bb2 = bb5 = -e;
1780
48.3k
            bd2 = bd5 = 0;
1781
48.3k
        }
1782
93.6k
        if (bbe >= 0)
1783
46.5k
            bb2 += bbe;
1784
47.0k
        else
1785
47.0k
            bd2 -= bbe;
1786
93.6k
        bs2 = bb2;
1787
93.6k
        bb2++;
1788
93.6k
        bd2++;
1789
1790
        /* At this stage bd5 - bb5 == e == bd2 - bb2 + bbe, bb2 - bs2 == 1,
1791
           and bs == 1, so:
1792
1793
              tdv == bd * 10**e = bd * 2**(bbe - bb2 + bd2) * 5**(bd5 - bb5)
1794
              srv == bb * 2**bbe = bb * 2**(bbe - bb2 + bb2)
1795
              0.5 ulp(srv) == 2**(bbe-1) = bs * 2**(bbe - bb2 + bs2)
1796
1797
           It follows that:
1798
1799
              M * tdv = bd * 2**bd2 * 5**bd5
1800
              M * srv = bb * 2**bb2 * 5**bb5
1801
              M * 0.5 ulp(srv) = bs * 2**bs2 * 5**bb5
1802
1803
           for some constant M.  (Actually, M == 2**(bb2 - bbe) * 5**bb5, but
1804
           this fact is not needed below.)
1805
        */
1806
1807
        /* Remove factor of 2**i, where i = min(bb2, bd2, bs2). */
1808
93.6k
        i = bb2 < bd2 ? bb2 : bd2;
1809
93.6k
        if (i > bs2)
1810
47.1k
            i = bs2;
1811
93.6k
        if (i > 0) {
1812
92.4k
            bb2 -= i;
1813
92.4k
            bd2 -= i;
1814
92.4k
            bs2 -= i;
1815
92.4k
        }
1816
1817
        /* Scale bb, bd, bs by the appropriate powers of 2 and 5. */
1818
93.6k
        if (bb5 > 0) {
1819
48.3k
            bs = pow5mult(bs, bb5);
1820
48.3k
            if (bs == NULL) {
1821
0
                goto failed_malloc;
1822
0
            }
1823
48.3k
            Bigint *bb1 = mult(bs, bb);
1824
48.3k
            Bfree(bb);
1825
48.3k
            bb = bb1;
1826
48.3k
            if (bb == NULL) {
1827
0
                goto failed_malloc;
1828
0
            }
1829
48.3k
        }
1830
93.6k
        if (bb2 > 0) {
1831
93.6k
            bb = lshift(bb, bb2);
1832
93.6k
            if (bb == NULL) {
1833
0
                goto failed_malloc;
1834
0
            }
1835
93.6k
        }
1836
93.6k
        if (bd5 > 0) {
1837
38.6k
            bd = pow5mult(bd, bd5);
1838
38.6k
            if (bd == NULL) {
1839
0
                goto failed_malloc;
1840
0
            }
1841
38.6k
        }
1842
93.6k
        if (bd2 > 0) {
1843
47.1k
            bd = lshift(bd, bd2);
1844
47.1k
            if (bd == NULL) {
1845
0
                goto failed_malloc;
1846
0
            }
1847
47.1k
        }
1848
93.6k
        if (bs2 > 0) {
1849
42.7k
            bs = lshift(bs, bs2);
1850
42.7k
            if (bs == NULL) {
1851
0
                goto failed_malloc;
1852
0
            }
1853
42.7k
        }
1854
1855
        /* Now bd, bb and bs are scaled versions of tdv, srv and 0.5 ulp(srv),
1856
           respectively.  Compute the difference |tdv - srv|, and compare
1857
           with 0.5 ulp(srv). */
1858
1859
93.6k
        delta = diff(bb, bd);
1860
93.6k
        if (delta == NULL) {
1861
0
            goto failed_malloc;
1862
0
        }
1863
93.6k
        dsign = delta->sign;
1864
93.6k
        delta->sign = 0;
1865
93.6k
        i = cmp(delta, bs);
1866
93.6k
        if (bc.nd > nd && i <= 0) {
1867
10.6k
            if (dsign)
1868
6.47k
                break;  /* Must use bigcomp(). */
1869
1870
            /* Here rv overestimates the truncated decimal value by at most
1871
               0.5 ulp(rv).  Hence rv either overestimates the true decimal
1872
               value by <= 0.5 ulp(rv), or underestimates it by some small
1873
               amount (< 0.1 ulp(rv)); either way, rv is within 0.5 ulps of
1874
               the true decimal value, so it's possible to exit.
1875
1876
               Exception: if scaled rv is a normal exact power of 2, but not
1877
               DBL_MIN, then rv - 0.5 ulp(rv) takes us all the way down to the
1878
               next double, so the correctly rounded result is either rv - 0.5
1879
               ulp(rv) or rv; in this case, use bigcomp to distinguish. */
1880
1881
4.22k
            if (!word1(&rv) && !(word0(&rv) & Bndry_mask)) {
1882
                /* rv can't be 0, since it's an overestimate for some
1883
                   nonzero value.  So rv is a normal power of 2. */
1884
1.24k
                j = (int)(word0(&rv) & Exp_mask) >> Exp_shift;
1885
                /* rv / 2^bc.scale = 2^(j - 1023 - bc.scale); use bigcomp if
1886
                   rv / 2^bc.scale >= 2^-1021. */
1887
1.24k
                if (j - bc.scale >= 2) {
1888
975
                    dval(&rv) -= 0.5 * sulp(&rv, &bc);
1889
975
                    break; /* Use bigcomp. */
1890
975
                }
1891
1.24k
            }
1892
1893
3.25k
            {
1894
3.25k
                bc.nd = nd;
1895
3.25k
                i = -1; /* Discarded digits make delta smaller. */
1896
3.25k
            }
1897
3.25k
        }
1898
1899
86.2k
        if (i < 0) {
1900
            /* Error is less than half an ulp -- check for
1901
             * special case of mantissa a power of two.
1902
             */
1903
40.4k
            if (dsign || word1(&rv) || word0(&rv) & Bndry_mask
1904
4.72k
                || (word0(&rv) & Exp_mask) <= (2*P+1)*Exp_msk1
1905
40.4k
                ) {
1906
37.2k
                break;
1907
37.2k
            }
1908
3.16k
            if (!delta->x[0] && delta->wds <= 1) {
1909
                /* exact result */
1910
637
                break;
1911
637
            }
1912
2.52k
            delta = lshift(delta,Log2P);
1913
2.52k
            if (delta == NULL) {
1914
0
                goto failed_malloc;
1915
0
            }
1916
2.52k
            if (cmp(delta, bs) > 0)
1917
971
                goto drop_down;
1918
1.55k
            break;
1919
2.52k
        }
1920
45.8k
        if (i == 0) {
1921
            /* exactly half-way between */
1922
3.70k
            if (dsign) {
1923
2.20k
                if ((word0(&rv) & Bndry_mask1) == Bndry_mask1
1924
811
                    &&  word1(&rv) == (
1925
811
                        (bc.scale &&
1926
0
                         (y = word0(&rv) & Exp_mask) <= 2*P*Exp_msk1) ?
1927
0
                        (0xffffffff & (0xffffffff << (2*P+1-(y>>Exp_shift)))) :
1928
811
                        0xffffffff)) {
1929
                    /*boundary case -- increment exponent*/
1930
464
                    word0(&rv) = (word0(&rv) & Exp_mask)
1931
464
                        + Exp_msk1
1932
464
                        ;
1933
464
                    word1(&rv) = 0;
1934
                    /* dsign = 0; */
1935
464
                    break;
1936
464
                }
1937
2.20k
            }
1938
1.49k
            else if (!(word0(&rv) & Bndry_mask) && !word1(&rv)) {
1939
971
              drop_down:
1940
                /* boundary case -- decrement exponent */
1941
971
                if (bc.scale) {
1942
0
                    L = word0(&rv) & Exp_mask;
1943
0
                    if (L <= (2*P+1)*Exp_msk1) {
1944
0
                        if (L > (P+2)*Exp_msk1)
1945
                            /* round even ==> */
1946
                            /* accept rv */
1947
0
                            break;
1948
                        /* rv = smallest denormal */
1949
0
                        if (bc.nd > nd)
1950
0
                            break;
1951
0
                        goto undfl;
1952
0
                    }
1953
0
                }
1954
971
                L = (word0(&rv) & Exp_mask) - Exp_msk1;
1955
971
                word0(&rv) = L | Bndry_mask1;
1956
971
                word1(&rv) = 0xffffffff;
1957
971
                break;
1958
971
            }
1959
3.23k
            if (!odd)
1960
2.51k
                break;
1961
722
            if (dsign)
1962
460
                dval(&rv) += sulp(&rv, &bc);
1963
262
            else {
1964
262
                dval(&rv) -= sulp(&rv, &bc);
1965
262
                if (!dval(&rv)) {
1966
0
                    if (bc.nd >nd)
1967
0
                        break;
1968
0
                    goto undfl;
1969
0
                }
1970
262
            }
1971
            /* dsign = 1 - dsign; */
1972
722
            break;
1973
722
        }
1974
42.1k
        if ((aadj = ratio(delta, bs)) <= 2.) {
1975
31.0k
            if (dsign)
1976
12.1k
                aadj = aadj1 = 1.;
1977
18.8k
            else if (word1(&rv) || word0(&rv) & Bndry_mask) {
1978
14.8k
                if (word1(&rv) == Tiny1 && !word0(&rv)) {
1979
0
                    if (bc.nd >nd)
1980
0
                        break;
1981
0
                    goto undfl;
1982
0
                }
1983
14.8k
                aadj = 1.;
1984
14.8k
                aadj1 = -1.;
1985
14.8k
            }
1986
3.96k
            else {
1987
                /* special case -- power of FLT_RADIX to be */
1988
                /* rounded down... */
1989
1990
3.96k
                if (aadj < 2./FLT_RADIX)
1991
0
                    aadj = 1./FLT_RADIX;
1992
3.96k
                else
1993
3.96k
                    aadj *= 0.5;
1994
3.96k
                aadj1 = -aadj;
1995
3.96k
            }
1996
31.0k
        }
1997
11.1k
        else {
1998
11.1k
            aadj *= 0.5;
1999
11.1k
            aadj1 = dsign ? aadj : -aadj;
2000
11.1k
            if (Flt_Rounds == 0)
2001
0
                aadj1 += 0.5;
2002
11.1k
        }
2003
42.1k
        y = word0(&rv) & Exp_mask;
2004
2005
        /* Check for overflow */
2006
2007
42.1k
        if (y == Exp_msk1*(DBL_MAX_EXP+Bias-1)) {
2008
2.85k
            dval(&rv0) = dval(&rv);
2009
2.85k
            word0(&rv) -= P*Exp_msk1;
2010
2.85k
            adj.d = aadj1 * ulp(&rv);
2011
2.85k
            dval(&rv) += adj.d;
2012
2.85k
            if ((word0(&rv) & Exp_mask) >=
2013
2.85k
                Exp_msk1*(DBL_MAX_EXP+Bias-P)) {
2014
1.02k
                if (word0(&rv0) == Big0 && word1(&rv0) == Big1) {
2015
760
                    goto ovfl;
2016
760
                }
2017
260
                word0(&rv) = Big0;
2018
260
                word1(&rv) = Big1;
2019
260
                goto cont;
2020
1.02k
            }
2021
1.83k
            else
2022
1.83k
                word0(&rv) += P*Exp_msk1;
2023
2.85k
        }
2024
39.2k
        else {
2025
39.2k
            if (bc.scale && y <= 2*P*Exp_msk1) {
2026
13.6k
                if (aadj <= 0x7fffffff) {
2027
13.6k
                    if ((z = (ULong)aadj) <= 0)
2028
789
                        z = 1;
2029
13.6k
                    aadj = z;
2030
13.6k
                    aadj1 = dsign ? aadj : -aadj;
2031
13.6k
                }
2032
13.6k
                dval(&aadj2) = aadj1;
2033
13.6k
                word0(&aadj2) += (2*P+1)*Exp_msk1 - y;
2034
13.6k
                aadj1 = dval(&aadj2);
2035
13.6k
            }
2036
39.2k
            adj.d = aadj1 * ulp(&rv);
2037
39.2k
            dval(&rv) += adj.d;
2038
39.2k
        }
2039
41.0k
        z = word0(&rv) & Exp_mask;
2040
41.0k
        if (bc.nd == nd) {
2041
36.1k
            if (!bc.scale)
2042
22.1k
                if (y == z) {
2043
                    /* Can we stop now? */
2044
20.6k
                    L = (Long)aadj;
2045
20.6k
                    aadj -= L;
2046
                    /* The tolerances below are conservative. */
2047
20.6k
                    if (dsign || word1(&rv) || word0(&rv) & Bndry_mask) {
2048
20.5k
                        if (aadj < .4999999 || aadj > .5000001)
2049
19.6k
                            break;
2050
20.5k
                    }
2051
73
                    else if (aadj < .4999999/FLT_RADIX)
2052
73
                        break;
2053
20.6k
                }
2054
36.1k
        }
2055
21.6k
      cont:
2056
21.6k
        Bfree(bb); bb = NULL;
2057
21.6k
        Bfree(bd); bd = NULL;
2058
21.6k
        Bfree(bs); bs = NULL;
2059
21.6k
        Bfree(delta); delta = NULL;
2060
21.6k
    }
2061
71.2k
    if (bc.nd > nd) {
2062
7.44k
        error = bigcomp(&rv, s0, &bc);
2063
7.44k
        if (error)
2064
0
            goto failed_malloc;
2065
7.44k
    }
2066
2067
71.2k
    if (bc.scale) {
2068
18.5k
        word0(&rv0) = Exp_1 - 2*P*Exp_msk1;
2069
18.5k
        word1(&rv0) = 0;
2070
18.5k
        dval(&rv) *= dval(&rv0);
2071
18.5k
    }
2072
2073
143k
  ret:
2074
143k
    result = sign ? -dval(&rv) : dval(&rv);
2075
143k
    goto done;
2076
2077
55.9k
  parse_error:
2078
55.9k
    result = 0.0;
2079
55.9k
    goto done;
2080
2081
0
  failed_malloc:
2082
0
    errno = ENOMEM;
2083
0
    result = -1.0;
2084
0
    goto done;
2085
2086
1.44k
  undfl:
2087
1.44k
    result = sign ? -0.0 : 0.0;
2088
1.44k
    goto done;
2089
2090
408k
  ovfl:
2091
408k
    errno = ERANGE;
2092
    /* Can't trust HUGE_VAL */
2093
408k
    word0(&rv) = Exp_mask;
2094
408k
    word1(&rv) = 0;
2095
408k
    result = sign ? -dval(&rv) : dval(&rv);
2096
408k
    goto done;
2097
2098
608k
  done:
2099
608k
    Bfree(bb);
2100
608k
    Bfree(bd);
2101
608k
    Bfree(bs);
2102
608k
    Bfree(bd0);
2103
608k
    Bfree(delta);
2104
608k
    return result;
2105
2106
71.2k
}
2107
2108
static char *
2109
rv_alloc(int i)
2110
44.6k
{
2111
44.6k
    int j, k, *r;
2112
2113
44.6k
    j = sizeof(ULong);
2114
44.6k
    for(k = 0;
2115
44.6k
        sizeof(Bigint) - sizeof(ULong) - sizeof(int) + j <= (unsigned)i;
2116
44.6k
        j <<= 1)
2117
0
        k++;
2118
44.6k
    r = (int*)Balloc(k);
2119
44.6k
    if (r == NULL)
2120
0
        return NULL;
2121
44.6k
    *r = k;
2122
44.6k
    return (char *)(r+1);
2123
44.6k
}
2124
2125
static char *
2126
nrv_alloc(const char *s, char **rve, int n)
2127
4.71k
{
2128
4.71k
    char *rv, *t;
2129
2130
4.71k
    rv = rv_alloc(n);
2131
4.71k
    if (rv == NULL)
2132
0
        return NULL;
2133
4.71k
    t = rv;
2134
14.2k
    while((*t = *s++)) t++;
2135
4.71k
    if (rve)
2136
4.71k
        *rve = t;
2137
4.71k
    return rv;
2138
4.71k
}
2139
2140
/* freedtoa(s) must be used to free values s returned by dtoa
2141
 * when MULTIPLE_THREADS is #defined.  It should be used in all cases,
2142
 * but for consistency with earlier versions of dtoa, it is optional
2143
 * when MULTIPLE_THREADS is not defined.
2144
 */
2145
2146
void
2147
_Py_dg_freedtoa(char *s)
2148
44.6k
{
2149
44.6k
    Bigint *b = (Bigint *)((int *)s - 1);
2150
44.6k
    b->maxwds = 1 << (b->k = *(int*)b);
2151
44.6k
    Bfree(b);
2152
44.6k
}
2153
2154
/* dtoa for IEEE arithmetic (dmg): convert double to ASCII string.
2155
 *
2156
 * Inspired by "How to Print Floating-Point Numbers Accurately" by
2157
 * Guy L. Steele, Jr. and Jon L. White [Proc. ACM SIGPLAN '90, pp. 112-126].
2158
 *
2159
 * Modifications:
2160
 *      1. Rather than iterating, we use a simple numeric overestimate
2161
 *         to determine k = floor(log10(d)).  We scale relevant
2162
 *         quantities using O(log2(k)) rather than O(k) multiplications.
2163
 *      2. For some modes > 2 (corresponding to ecvt and fcvt), we don't
2164
 *         try to generate digits strictly left to right.  Instead, we
2165
 *         compute with fewer bits and propagate the carry if necessary
2166
 *         when rounding the final digit up.  This is often faster.
2167
 *      3. Under the assumption that input will be rounded nearest,
2168
 *         mode 0 renders 1e23 as 1e23 rather than 9.999999999999999e22.
2169
 *         That is, we allow equality in stopping tests when the
2170
 *         round-nearest rule will give the same floating-point value
2171
 *         as would satisfaction of the stopping test with strict
2172
 *         inequality.
2173
 *      4. We remove common factors of powers of 2 from relevant
2174
 *         quantities.
2175
 *      5. When converting floating-point integers less than 1e16,
2176
 *         we use floating-point arithmetic rather than resorting
2177
 *         to multiple-precision integers.
2178
 *      6. When asked to produce fewer than 15 digits, we first try
2179
 *         to get by with floating-point arithmetic; we resort to
2180
 *         multiple-precision integer arithmetic only if we cannot
2181
 *         guarantee that the floating-point calculation has given
2182
 *         the correctly rounded result.  For k requested digits and
2183
 *         "uniformly" distributed input, the probability is
2184
 *         something like 10^(k-15) that we must resort to the Long
2185
 *         calculation.
2186
 */
2187
2188
/* Additional notes (METD): (1) returns NULL on failure.  (2) to avoid memory
2189
   leakage, a successful call to _Py_dg_dtoa should always be matched by a
2190
   call to _Py_dg_freedtoa. */
2191
2192
char *
2193
_Py_dg_dtoa(double dd, int mode, int ndigits,
2194
            int *decpt, int *sign, char **rve)
2195
44.6k
{
2196
    /*  Arguments ndigits, decpt, sign are similar to those
2197
        of ecvt and fcvt; trailing zeros are suppressed from
2198
        the returned string.  If not null, *rve is set to point
2199
        to the end of the return value.  If d is +-Infinity or NaN,
2200
        then *decpt is set to 9999.
2201
2202
        mode:
2203
        0 ==> shortest string that yields d when read in
2204
        and rounded to nearest.
2205
        1 ==> like 0, but with Steele & White stopping rule;
2206
        e.g. with IEEE P754 arithmetic , mode 0 gives
2207
        1e23 whereas mode 1 gives 9.999999999999999e22.
2208
        2 ==> max(1,ndigits) significant digits.  This gives a
2209
        return value similar to that of ecvt, except
2210
        that trailing zeros are suppressed.
2211
        3 ==> through ndigits past the decimal point.  This
2212
        gives a return value similar to that from fcvt,
2213
        except that trailing zeros are suppressed, and
2214
        ndigits can be negative.
2215
        4,5 ==> similar to 2 and 3, respectively, but (in
2216
        round-nearest mode) with the tests of mode 0 to
2217
        possibly return a shorter string that rounds to d.
2218
        With IEEE arithmetic and compilation with
2219
        -DHonor_FLT_ROUNDS, modes 4 and 5 behave the same
2220
        as modes 2 and 3 when FLT_ROUNDS != 1.
2221
        6-9 ==> Debugging modes similar to mode - 4:  don't try
2222
        fast floating-point estimate (if applicable).
2223
2224
        Values of mode other than 0-9 are treated as mode 0.
2225
2226
        Sufficient space is allocated to the return value
2227
        to hold the suppressed trailing zeros.
2228
    */
2229
2230
44.6k
    int bbits, b2, b5, be, dig, i, ieps, ilim, ilim0, ilim1,
2231
44.6k
        j, j1, k, k0, k_check, leftright, m2, m5, s2, s5,
2232
44.6k
        spec_case, try_quick;
2233
44.6k
    Long L;
2234
44.6k
    int denorm;
2235
44.6k
    ULong x;
2236
44.6k
    Bigint *b, *b1, *delta, *mlo, *mhi, *S;
2237
44.6k
    U d2, eps, u;
2238
44.6k
    double ds;
2239
44.6k
    char *s, *s0;
2240
2241
    /* set pointers to NULL, to silence gcc compiler warnings and make
2242
       cleanup easier on error */
2243
44.6k
    mlo = mhi = S = 0;
2244
44.6k
    s0 = 0;
2245
2246
44.6k
    u.d = dd;
2247
44.6k
    if (word0(&u) & Sign_bit) {
2248
        /* set sign for everything, including 0's and NaNs */
2249
11.3k
        *sign = 1;
2250
11.3k
        word0(&u) &= ~Sign_bit; /* clear sign bit */
2251
11.3k
    }
2252
33.3k
    else
2253
33.3k
        *sign = 0;
2254
2255
    /* quick return for Infinities, NaNs and zeros */
2256
44.6k
    if ((word0(&u) & Exp_mask) == Exp_mask)
2257
695
    {
2258
        /* Infinity or NaN */
2259
695
        *decpt = 9999;
2260
695
        if (!word1(&u) && !(word0(&u) & 0xfffff))
2261
694
            return nrv_alloc("Infinity", rve, 8);
2262
1
        return nrv_alloc("NaN", rve, 3);
2263
695
    }
2264
43.9k
    if (!dval(&u)) {
2265
4.01k
        *decpt = 1;
2266
4.01k
        return nrv_alloc("0", rve, 1);
2267
4.01k
    }
2268
2269
    /* compute k = floor(log10(d)).  The computation may leave k
2270
       one too large, but should never leave k too small. */
2271
39.9k
    b = d2b(&u, &be, &bbits);
2272
39.9k
    if (b == NULL)
2273
0
        goto failed_malloc;
2274
39.9k
    if ((i = (int)(word0(&u) >> Exp_shift1 & (Exp_mask>>Exp_shift1)))) {
2275
35.6k
        dval(&d2) = dval(&u);
2276
35.6k
        word0(&d2) &= Frac_mask1;
2277
35.6k
        word0(&d2) |= Exp_11;
2278
2279
        /* log(x)       ~=~ log(1.5) + (x-1.5)/1.5
2280
         * log10(x)      =  log(x) / log(10)
2281
         *              ~=~ log(1.5)/log(10) + (x-1.5)/(1.5*log(10))
2282
         * log10(d) = (i-Bias)*log(2)/log(10) + log10(d2)
2283
         *
2284
         * This suggests computing an approximation k to log10(d) by
2285
         *
2286
         * k = (i - Bias)*0.301029995663981
2287
         *      + ( (d2-1.5)*0.289529654602168 + 0.176091259055681 );
2288
         *
2289
         * We want k to be too large rather than too small.
2290
         * The error in the first-order Taylor series approximation
2291
         * is in our favor, so we just round up the constant enough
2292
         * to compensate for any error in the multiplication of
2293
         * (i - Bias) by 0.301029995663981; since |i - Bias| <= 1077,
2294
         * and 1077 * 0.30103 * 2^-52 ~=~ 7.2e-14,
2295
         * adding 1e-13 to the constant term more than suffices.
2296
         * Hence we adjust the constant term to 0.1760912590558.
2297
         * (We could get a more accurate k by invoking log10,
2298
         *  but this is probably not worthwhile.)
2299
         */
2300
2301
35.6k
        i -= Bias;
2302
35.6k
        denorm = 0;
2303
35.6k
    }
2304
4.30k
    else {
2305
        /* d is denormalized */
2306
2307
4.30k
        i = bbits + be + (Bias + (P-1) - 1);
2308
4.30k
        x = i > 32  ? word0(&u) << (64 - i) | word1(&u) >> (i - 32)
2309
4.30k
            : word1(&u) << (32 - i);
2310
4.30k
        dval(&d2) = x;
2311
4.30k
        word0(&d2) -= 31*Exp_msk1; /* adjust exponent */
2312
4.30k
        i -= (Bias + (P-1) - 1) + 1;
2313
4.30k
        denorm = 1;
2314
4.30k
    }
2315
39.9k
    ds = (dval(&d2)-1.5)*0.289529654602168 + 0.1760912590558 +
2316
39.9k
        i*0.301029995663981;
2317
39.9k
    k = (int)ds;
2318
39.9k
    if (ds < 0. && ds != k)
2319
11.6k
        k--;    /* want k = floor(ds) */
2320
39.9k
    k_check = 1;
2321
39.9k
    if (k >= 0 && k <= Ten_pmax) {
2322
17.6k
        if (dval(&u) < tens[k])
2323
1.85k
            k--;
2324
17.6k
        k_check = 0;
2325
17.6k
    }
2326
39.9k
    j = bbits - i - 1;
2327
39.9k
    if (j >= 0) {
2328
17.3k
        b2 = 0;
2329
17.3k
        s2 = j;
2330
17.3k
    }
2331
22.5k
    else {
2332
22.5k
        b2 = -j;
2333
22.5k
        s2 = 0;
2334
22.5k
    }
2335
39.9k
    if (k >= 0) {
2336
27.7k
        b5 = 0;
2337
27.7k
        s5 = k;
2338
27.7k
        s2 += k;
2339
27.7k
    }
2340
12.1k
    else {
2341
12.1k
        b2 -= k;
2342
12.1k
        b5 = -k;
2343
12.1k
        s5 = 0;
2344
12.1k
    }
2345
39.9k
    if (mode < 0 || mode > 9)
2346
0
        mode = 0;
2347
2348
39.9k
    try_quick = 1;
2349
2350
39.9k
    if (mode > 5) {
2351
0
        mode -= 4;
2352
0
        try_quick = 0;
2353
0
    }
2354
39.9k
    leftright = 1;
2355
39.9k
    ilim = ilim1 = -1;  /* Values for cases 0 and 1; done here to */
2356
    /* silence erroneous "gcc -Wall" warning. */
2357
39.9k
    switch(mode) {
2358
39.8k
    case 0:
2359
39.8k
    case 1:
2360
39.8k
        i = 18;
2361
39.8k
        ndigits = 0;
2362
39.8k
        break;
2363
0
    case 2:
2364
0
        leftright = 0;
2365
0
        _Py_FALLTHROUGH;
2366
0
    case 4:
2367
0
        if (ndigits <= 0)
2368
0
            ndigits = 1;
2369
0
        ilim = ilim1 = i = ndigits;
2370
0
        break;
2371
95
    case 3:
2372
95
        leftright = 0;
2373
95
        _Py_FALLTHROUGH;
2374
95
    case 5:
2375
95
        i = ndigits + k + 1;
2376
95
        ilim = i;
2377
95
        ilim1 = i - 1;
2378
95
        if (i <= 0)
2379
0
            i = 1;
2380
39.9k
    }
2381
39.9k
    s0 = rv_alloc(i);
2382
39.9k
    if (s0 == NULL)
2383
0
        goto failed_malloc;
2384
39.9k
    s = s0;
2385
2386
2387
39.9k
    if (ilim >= 0 && ilim <= Quick_max && try_quick) {
2388
2389
        /* Try to get by with floating-point arithmetic. */
2390
2391
95
        i = 0;
2392
95
        dval(&d2) = dval(&u);
2393
95
        k0 = k;
2394
95
        ilim0 = ilim;
2395
95
        ieps = 2; /* conservative */
2396
95
        if (k > 0) {
2397
76
            ds = tens[k&0xf];
2398
76
            j = k >> 4;
2399
76
            if (j & Bletch) {
2400
                /* prevent overflows */
2401
0
                j &= Bletch - 1;
2402
0
                dval(&u) /= bigtens[n_bigtens-1];
2403
0
                ieps++;
2404
0
            }
2405
76
            for(; j; j >>= 1, i++)
2406
0
                if (j & 1) {
2407
0
                    ieps++;
2408
0
                    ds *= bigtens[i];
2409
0
                }
2410
76
            dval(&u) /= ds;
2411
76
        }
2412
19
        else if ((j1 = -k)) {
2413
0
            dval(&u) *= tens[j1 & 0xf];
2414
0
            for(j = j1 >> 4; j; j >>= 1, i++)
2415
0
                if (j & 1) {
2416
0
                    ieps++;
2417
0
                    dval(&u) *= bigtens[i];
2418
0
                }
2419
0
        }
2420
95
        if (k_check && dval(&u) < 1. && ilim > 0) {
2421
0
            if (ilim1 <= 0)
2422
0
                goto fast_failed;
2423
0
            ilim = ilim1;
2424
0
            k--;
2425
0
            dval(&u) *= 10.;
2426
0
            ieps++;
2427
0
        }
2428
95
        dval(&eps) = ieps*dval(&u) + 7.;
2429
95
        word0(&eps) -= (P-1)*Exp_msk1;
2430
95
        if (ilim == 0) {
2431
0
            S = mhi = 0;
2432
0
            dval(&u) -= 5.;
2433
0
            if (dval(&u) > dval(&eps))
2434
0
                goto one_digit;
2435
0
            if (dval(&u) < -dval(&eps))
2436
0
                goto no_digits;
2437
0
            goto fast_failed;
2438
0
        }
2439
95
        if (leftright) {
2440
            /* Use Steele & White method of only
2441
             * generating digits needed.
2442
             */
2443
0
            dval(&eps) = 0.5/tens[ilim-1] - dval(&eps);
2444
0
            for(i = 0;;) {
2445
0
                L = (Long)dval(&u);
2446
0
                dval(&u) -= L;
2447
0
                *s++ = '0' + (int)L;
2448
0
                if (dval(&u) < dval(&eps))
2449
0
                    goto ret1;
2450
0
                if (1. - dval(&u) < dval(&eps))
2451
0
                    goto bump_up;
2452
0
                if (++i >= ilim)
2453
0
                    break;
2454
0
                dval(&eps) *= 10.;
2455
0
                dval(&u) *= 10.;
2456
0
            }
2457
0
        }
2458
95
        else {
2459
            /* Generate ilim digits, then fix them up. */
2460
95
            dval(&eps) *= tens[ilim-1];
2461
261
            for(i = 1;; i++, dval(&u) *= 10.) {
2462
261
                L = (Long)(dval(&u));
2463
261
                if (!(dval(&u) -= L))
2464
11
                    ilim = i;
2465
261
                *s++ = '0' + (int)L;
2466
261
                if (i == ilim) {
2467
95
                    if (dval(&u) > 0.5 + dval(&eps))
2468
47
                        goto bump_up;
2469
48
                    else if (dval(&u) < 0.5 - dval(&eps)) {
2470
56
                        while(*--s == '0');
2471
48
                        s++;
2472
48
                        goto ret1;
2473
48
                    }
2474
0
                    break;
2475
95
                }
2476
261
            }
2477
95
        }
2478
0
      fast_failed:
2479
0
        s = s0;
2480
0
        dval(&u) = dval(&d2);
2481
0
        k = k0;
2482
0
        ilim = ilim0;
2483
0
    }
2484
2485
    /* Do we have a "small" integer? */
2486
2487
39.8k
    if (be >= 0 && k <= Int_max) {
2488
        /* Yes. */
2489
10.1k
        ds = tens[k];
2490
10.1k
        if (ndigits < 0 && ilim <= 0) {
2491
0
            S = mhi = 0;
2492
0
            if (ilim < 0 || dval(&u) <= 5*ds)
2493
0
                goto no_digits;
2494
0
            goto one_digit;
2495
0
        }
2496
13.2k
        for(i = 1;; i++, dval(&u) *= 10.) {
2497
13.2k
            L = (Long)(dval(&u) / ds);
2498
13.2k
            dval(&u) -= L*ds;
2499
13.2k
            *s++ = '0' + (int)L;
2500
13.2k
            if (!dval(&u)) {
2501
10.1k
                break;
2502
10.1k
            }
2503
3.01k
            if (i == ilim) {
2504
0
                dval(&u) += dval(&u);
2505
0
                if (dval(&u) > ds || (dval(&u) == ds && L & 1)) {
2506
47
                  bump_up:
2507
50
                    while(*--s == '9')
2508
3
                        if (s == s0) {
2509
0
                            k++;
2510
0
                            *s = '0';
2511
0
                            break;
2512
0
                        }
2513
47
                    ++*s++;
2514
47
                }
2515
0
                else {
2516
                    /* Strip trailing zeros. This branch was missing from the
2517
                       original dtoa.c, leading to surplus trailing zeros in
2518
                       some cases. See bugs.python.org/issue40780. */
2519
0
                    while (s > s0 && s[-1] == '0') {
2520
0
                        --s;
2521
0
                    }
2522
0
                }
2523
47
                break;
2524
0
            }
2525
3.01k
        }
2526
10.2k
        goto ret1;
2527
10.1k
    }
2528
2529
29.6k
    m2 = b2;
2530
29.6k
    m5 = b5;
2531
29.6k
    if (leftright) {
2532
29.6k
        i =
2533
29.6k
            denorm ? be + (Bias + (P-1) - 1 + 1) :
2534
29.6k
            1 + P - bbits;
2535
29.6k
        b2 += i;
2536
29.6k
        s2 += i;
2537
29.6k
        mhi = i2b(1);
2538
29.6k
        if (mhi == NULL)
2539
0
            goto failed_malloc;
2540
29.6k
    }
2541
29.6k
    if (m2 > 0 && s2 > 0) {
2542
25.8k
        i = m2 < s2 ? m2 : s2;
2543
25.8k
        b2 -= i;
2544
25.8k
        m2 -= i;
2545
25.8k
        s2 -= i;
2546
25.8k
    }
2547
29.6k
    if (b5 > 0) {
2548
12.1k
        if (leftright) {
2549
12.1k
            if (m5 > 0) {
2550
12.1k
                mhi = pow5mult(mhi, m5);
2551
12.1k
                if (mhi == NULL)
2552
0
                    goto failed_malloc;
2553
12.1k
                b1 = mult(mhi, b);
2554
12.1k
                Bfree(b);
2555
12.1k
                b = b1;
2556
12.1k
                if (b == NULL)
2557
0
                    goto failed_malloc;
2558
12.1k
            }
2559
12.1k
            if ((j = b5 - m5)) {
2560
0
                b = pow5mult(b, j);
2561
0
                if (b == NULL)
2562
0
                    goto failed_malloc;
2563
0
            }
2564
12.1k
        }
2565
0
        else {
2566
0
            b = pow5mult(b, b5);
2567
0
            if (b == NULL)
2568
0
                goto failed_malloc;
2569
0
        }
2570
12.1k
    }
2571
29.6k
    S = i2b(1);
2572
29.6k
    if (S == NULL)
2573
0
        goto failed_malloc;
2574
29.6k
    if (s5 > 0) {
2575
15.3k
        S = pow5mult(S, s5);
2576
15.3k
        if (S == NULL)
2577
0
            goto failed_malloc;
2578
15.3k
    }
2579
2580
    /* Check for special case that d is a normalized power of 2. */
2581
2582
29.6k
    spec_case = 0;
2583
29.6k
    if ((mode < 2 || leftright)
2584
29.6k
        ) {
2585
29.6k
        if (!word1(&u) && !(word0(&u) & Bndry_mask)
2586
1.42k
            && word0(&u) & (Exp_mask & ~Exp_msk1)
2587
29.6k
            ) {
2588
            /* The special case */
2589
1.16k
            b2 += Log2P;
2590
1.16k
            s2 += Log2P;
2591
1.16k
            spec_case = 1;
2592
1.16k
        }
2593
29.6k
    }
2594
2595
    /* Arrange for convenient computation of quotients:
2596
     * shift left if necessary so divisor has 4 leading 0 bits.
2597
     *
2598
     * Perhaps we should just compute leading 28 bits of S once
2599
     * and for all and pass them and a shift to quorem, so it
2600
     * can do shifts and ors to compute the numerator for q.
2601
     */
2602
29.6k
#define iInc 28
2603
29.6k
    i = dshift(S, s2);
2604
29.6k
    b2 += i;
2605
29.6k
    m2 += i;
2606
29.6k
    s2 += i;
2607
29.6k
    if (b2 > 0) {
2608
29.6k
        b = lshift(b, b2);
2609
29.6k
        if (b == NULL)
2610
0
            goto failed_malloc;
2611
29.6k
    }
2612
29.6k
    if (s2 > 0) {
2613
29.0k
        S = lshift(S, s2);
2614
29.0k
        if (S == NULL)
2615
0
            goto failed_malloc;
2616
29.0k
    }
2617
29.6k
    if (k_check) {
2618
22.2k
        if (cmp(b,S) < 0) {
2619
3.49k
            k--;
2620
3.49k
            b = multadd(b, 10, 0);      /* we botched the k estimate */
2621
3.49k
            if (b == NULL)
2622
0
                goto failed_malloc;
2623
3.49k
            if (leftright) {
2624
3.49k
                mhi = multadd(mhi, 10, 0);
2625
3.49k
                if (mhi == NULL)
2626
0
                    goto failed_malloc;
2627
3.49k
            }
2628
3.49k
            ilim = ilim1;
2629
3.49k
        }
2630
22.2k
    }
2631
29.6k
    if (ilim <= 0 && (mode == 3 || mode == 5)) {
2632
0
        if (ilim < 0) {
2633
            /* no digits, fcvt style */
2634
0
          no_digits:
2635
0
            k = -1 - ndigits;
2636
0
            goto ret;
2637
0
        }
2638
0
        else {
2639
0
            S = multadd(S, 5, 0);
2640
0
            if (S == NULL)
2641
0
                goto failed_malloc;
2642
0
            if (cmp(b, S) <= 0)
2643
0
                goto no_digits;
2644
0
        }
2645
0
      one_digit:
2646
0
        *s++ = '1';
2647
0
        k++;
2648
0
        goto ret;
2649
0
    }
2650
29.6k
    if (leftright) {
2651
29.6k
        if (m2 > 0) {
2652
28.4k
            mhi = lshift(mhi, m2);
2653
28.4k
            if (mhi == NULL)
2654
0
                goto failed_malloc;
2655
28.4k
        }
2656
2657
        /* Compute mlo -- check for special case
2658
         * that d is a normalized power of 2.
2659
         */
2660
2661
29.6k
        mlo = mhi;
2662
29.6k
        if (spec_case) {
2663
1.16k
            mhi = Balloc(mhi->k);
2664
1.16k
            if (mhi == NULL)
2665
0
                goto failed_malloc;
2666
1.16k
            Bcopy(mhi, mlo);
2667
1.16k
            mhi = lshift(mhi, Log2P);
2668
1.16k
            if (mhi == NULL)
2669
0
                goto failed_malloc;
2670
1.16k
        }
2671
2672
121k
        for(i = 1;;i++) {
2673
121k
            dig = quorem(b,S) + '0';
2674
            /* Do we yet have the shortest decimal string
2675
             * that will round to d?
2676
             */
2677
121k
            j = cmp(b, mlo);
2678
121k
            delta = diff(S, mhi);
2679
121k
            if (delta == NULL)
2680
0
                goto failed_malloc;
2681
121k
            j1 = delta->sign ? 1 : cmp(b, delta);
2682
121k
            Bfree(delta);
2683
121k
            if (j1 == 0 && mode != 1 && !(word1(&u) & 1)
2684
121k
                ) {
2685
2.05k
                if (dig == '9')
2686
540
                    goto round_9_up;
2687
1.51k
                if (j > 0)
2688
745
                    dig++;
2689
1.51k
                *s++ = dig;
2690
1.51k
                goto ret;
2691
2.05k
            }
2692
119k
            if (j < 0 || (j == 0 && mode != 1
2693
1.80k
                          && !(word1(&u) & 1)
2694
103k
                    )) {
2695
16.2k
                if (!b->x[0] && b->wds <= 1) {
2696
2.31k
                    goto accept_dig;
2697
2.31k
                }
2698
13.9k
                if (j1 > 0) {
2699
2.98k
                    b = lshift(b, 1);
2700
2.98k
                    if (b == NULL)
2701
0
                        goto failed_malloc;
2702
2.98k
                    j1 = cmp(b, S);
2703
2.98k
                    if ((j1 > 0 || (j1 == 0 && dig & 1))
2704
1.87k
                        && dig++ == '9')
2705
423
                        goto round_9_up;
2706
2.98k
                }
2707
15.8k
              accept_dig:
2708
15.8k
                *s++ = dig;
2709
15.8k
                goto ret;
2710
13.9k
            }
2711
103k
            if (j1 > 0) {
2712
11.3k
                if (dig == '9') { /* possible if i == 1 */
2713
3.43k
                  round_9_up:
2714
3.43k
                    *s++ = '9';
2715
3.43k
                    goto roundoff;
2716
2.47k
                }
2717
8.82k
                *s++ = dig + 1;
2718
8.82k
                goto ret;
2719
11.3k
            }
2720
92.0k
            *s++ = dig;
2721
92.0k
            if (i == ilim)
2722
0
                break;
2723
92.0k
            b = multadd(b, 10, 0);
2724
92.0k
            if (b == NULL)
2725
0
                goto failed_malloc;
2726
92.0k
            if (mlo == mhi) {
2727
87.0k
                mlo = mhi = multadd(mhi, 10, 0);
2728
87.0k
                if (mlo == NULL)
2729
0
                    goto failed_malloc;
2730
87.0k
            }
2731
4.93k
            else {
2732
4.93k
                mlo = multadd(mlo, 10, 0);
2733
4.93k
                if (mlo == NULL)
2734
0
                    goto failed_malloc;
2735
4.93k
                mhi = multadd(mhi, 10, 0);
2736
4.93k
                if (mhi == NULL)
2737
0
                    goto failed_malloc;
2738
4.93k
            }
2739
92.0k
        }
2740
29.6k
    }
2741
0
    else
2742
0
        for(i = 1;; i++) {
2743
0
            *s++ = dig = quorem(b,S) + '0';
2744
0
            if (!b->x[0] && b->wds <= 1) {
2745
0
                goto ret;
2746
0
            }
2747
0
            if (i >= ilim)
2748
0
                break;
2749
0
            b = multadd(b, 10, 0);
2750
0
            if (b == NULL)
2751
0
                goto failed_malloc;
2752
0
        }
2753
2754
    /* Round off last digit */
2755
2756
0
    b = lshift(b, 1);
2757
0
    if (b == NULL)
2758
0
        goto failed_malloc;
2759
0
    j = cmp(b, S);
2760
0
    if (j > 0 || (j == 0 && dig & 1)) {
2761
3.43k
      roundoff:
2762
3.43k
        while(*--s == '9')
2763
3.43k
            if (s == s0) {
2764
3.43k
                k++;
2765
3.43k
                *s++ = '1';
2766
3.43k
                goto ret;
2767
3.43k
            }
2768
0
        ++*s++;
2769
0
    }
2770
0
    else {
2771
0
        while(*--s == '0');
2772
0
        s++;
2773
0
    }
2774
29.6k
  ret:
2775
29.6k
    Bfree(S);
2776
29.6k
    if (mhi) {
2777
29.6k
        if (mlo && mlo != mhi)
2778
1.16k
            Bfree(mlo);
2779
29.6k
        Bfree(mhi);
2780
29.6k
    }
2781
39.9k
  ret1:
2782
39.9k
    Bfree(b);
2783
39.9k
    *s = 0;
2784
39.9k
    *decpt = k + 1;
2785
39.9k
    if (rve)
2786
39.9k
        *rve = s;
2787
39.9k
    return s0;
2788
0
  failed_malloc:
2789
0
    if (S)
2790
0
        Bfree(S);
2791
0
    if (mlo && mlo != mhi)
2792
0
        Bfree(mlo);
2793
0
    if (mhi)
2794
0
        Bfree(mhi);
2795
0
    if (b)
2796
0
        Bfree(b);
2797
0
    if (s0)
2798
0
        _Py_dg_freedtoa(s0);
2799
0
    return NULL;
2800
29.6k
}
2801
2802
#endif  // _PY_SHORT_FLOAT_REPR == 1
2803
2804
PyStatus
2805
_PyDtoa_Init(PyInterpreterState *interp)
2806
36
{
2807
36
#if _PY_SHORT_FLOAT_REPR == 1 && !defined(Py_USING_MEMORY_DEBUGGER)
2808
36
    Bigint **p5s = interp->dtoa.p5s;
2809
2810
    // 5**4 = 625
2811
36
    Bigint *p5 = i2b(625);
2812
36
    if (p5 == NULL) {
2813
0
        return PyStatus_NoMemory();
2814
0
    }
2815
36
    p5s[0] = p5;
2816
2817
    // compute 5**8, 5**16, 5**32, ..., 5**512
2818
288
    for (Py_ssize_t i = 1; i < Bigint_Pow5size; i++) {
2819
252
        p5 = mult(p5, p5);
2820
252
        if (p5 == NULL) {
2821
0
            return PyStatus_NoMemory();
2822
0
        }
2823
252
        p5s[i] = p5;
2824
252
    }
2825
2826
36
#endif
2827
36
    return PyStatus_Ok();
2828
36
}
2829
2830
void
2831
_PyDtoa_Fini(PyInterpreterState *interp)
2832
0
{
2833
0
#if _PY_SHORT_FLOAT_REPR == 1 && !defined(Py_USING_MEMORY_DEBUGGER)
2834
0
    Bigint **p5s = interp->dtoa.p5s;
2835
0
    for (Py_ssize_t i = 0; i < Bigint_Pow5size; i++) {
2836
0
        Bigint *p5 = p5s[i];
2837
        p5s[i] = NULL;
2838
0
        Bfree(p5);
2839
0
    }
2840
0
#endif
2841
0
}