Skip to content

Commit 84895f1

Browse files
dpgeorgedpgeorge
committed
py/parsenum: Improve parsing of floating point numbers.
This patch improves parsing of floating point numbers by converting all the digits (integer and fractional) together into a number 1 or greater, and then applying the correct power of 10 at the very end. In particular the multiple "multiply by 0.1" operations to build a fraction are now combined together and applied at the same time as the exponent, at the very end. This helps to retain precision during parsing of floats, and also includes a check that the number doesn't overflow during the parsing. One benefit is that a float will have the same value no matter where the decimal point is located, eg 1.23 == 123e-2.
1 parent f59c6b4 commit 84895f1

4 files changed

Lines changed: 60 additions & 6 deletions

File tree

‎py/parsenum.c‎

Lines changed: 21 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -170,6 +170,14 @@ typedef enum {
170170

171171
mp_obj_t mp_parse_num_decimal(const char *str, size_t len, bool allow_imag, bool force_complex, mp_lexer_t *lex) {
172172
#if MICROPY_PY_BUILTINS_FLOAT
173+
174+
// DEC_VAL_MAX only needs to be rough and is used to retain precision while not overflowing
175+
#if MICROPY_FLOAT_IMPL == MICROPY_FLOAT_IMPL_FLOAT
176+
#define DEC_VAL_MAX 1e20F
177+
#elif MICROPY_FLOAT_IMPL == MICROPY_FLOAT_IMPL_DOUBLE
178+
#define DEC_VAL_MAX 1e200
179+
#endif
180+
173181
const char *top = str + len;
174182
mp_float_t dec_val = 0;
175183
bool dec_neg = false;
@@ -214,20 +222,27 @@ mp_obj_t mp_parse_num_decimal(const char *str, size_t len, bool allow_imag, bool
214222
// string should be a decimal number
215223
parse_dec_in_t in = PARSE_DEC_IN_INTG;
216224
bool exp_neg = false;
217-
mp_float_t frac_mult = 0.1;
218225
mp_int_t exp_val = 0;
226+
mp_int_t exp_extra = 0;
219227
while (str < top) {
220228
mp_uint_t dig = *str++;
221229
if ('0' <= dig && dig <= '9') {
222230
dig -= '0';
223231
if (in == PARSE_DEC_IN_EXP) {
224232
exp_val = 10 * exp_val + dig;
225233
} else {
226-
if (in == PARSE_DEC_IN_FRAC) {
227-
dec_val += dig * frac_mult;
228-
frac_mult *= MICROPY_FLOAT_CONST(0.1);
229-
} else {
234+
if (dec_val < DEC_VAL_MAX) {
235+
// dec_val won't overflow so keep accumulating
230236
dec_val = 10 * dec_val + dig;
237+
if (in == PARSE_DEC_IN_FRAC) {
238+
--exp_extra;
239+
}
240+
} else {
241+
// dec_val might overflow and we anyway can't represent more digits
242+
// of precision, so ignore the digit and just adjust the exponent
243+
if (in == PARSE_DEC_IN_INTG) {
244+
++exp_extra;
245+
}
231246
}
232247
}
233248
} else if (in == PARSE_DEC_IN_INTG && dig == '.') {
@@ -261,7 +276,7 @@ mp_obj_t mp_parse_num_decimal(const char *str, size_t len, bool allow_imag, bool
261276
}
262277

263278
// apply the exponent
264-
dec_val *= MICROPY_FLOAT_C_FUN(pow)(10, exp_val);
279+
dec_val *= MICROPY_FLOAT_C_FUN(pow)(10, exp_val + exp_extra);
265280
}
266281

267282
// negate value if needed

‎tests/float/float_parse.py‎

Lines changed: 22 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,22 @@
1+
# test parsing of floats
2+
3+
inf = float('inf')
4+
5+
# it shouldn't matter where the decimal point is if the exponent balances the value
6+
print(float('1234') - float('0.1234e4'))
7+
print(float('1.015625') - float('1015625e-6'))
8+
9+
# very large integer part with a very negative exponent should cancel out
10+
print(float('9' * 60 + 'e-60'))
11+
print(float('9' * 60 + 'e-40'))
12+
print(float('9' * 60 + 'e-20') == float('1e40'))
13+
14+
# many fractional digits
15+
print(float('.' + '9' * 70))
16+
print(float('.' + '9' * 70 + 'e20'))
17+
print(float('.' + '9' * 70 + 'e-50') == float('1e-50'))
18+
19+
# tiny fraction with large exponent
20+
print(float('.' + '0' * 60 + '1e10') == float('1e-51'))
21+
print(float('.' + '0' * 60 + '9e25'))
22+
print(float('.' + '0' * 60 + '9e40'))

‎tests/float/float_parse_doubleprec.py‎

Lines changed: 16 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,16 @@
1+
# test parsing of floats, requiring double-precision
2+
3+
# very large integer part with a very negative exponent should cancel out
4+
print(float('9' * 400 + 'e-100'))
5+
print(float('9' * 400 + 'e-200'))
6+
print(float('9' * 400 + 'e-400'))
7+
8+
# many fractional digits
9+
print(float('.' + '9' * 400))
10+
print(float('.' + '9' * 400 + 'e100'))
11+
print(float('.' + '9' * 400 + 'e-100'))
12+
13+
# tiny fraction with large exponent
14+
print(float('.' + '0' * 400 + '9e100'))
15+
print(float('.' + '0' * 400 + '9e200'))
16+
print(float('.' + '0' * 400 + '9e400'))

‎tests/run-tests‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -271,6 +271,7 @@ def run_tests(pyb, tests, args, base_path="."):
271271
if upy_float_precision < 64:
272272
skip_tests.add('float/float_divmod.py') # tested by float/float_divmod_relaxed.py instead
273273
skip_tests.add('float/float2int_doubleprec_intbig.py')
274+
skip_tests.add('float/float_parse_doubleprec.py')
274275

275276
if not has_complex:
276277
skip_tests.add('float/complex1.py')

0 commit comments

Comments
 (0)