Fwd: mpx_set_str2(mpx_t, const char*, const char*, int) suggestion
Pierre Chatelier
pierre at chachatelier.fr
Tue Feb 2 19:00:46 UTC 2016
Hello,
It's an "up" from a previous message without answer.
I don't see guidelines on gmplib.org to know how to submit a patch, so I do not know how to submit otherwise.
Here is a patch for the following feature :
Currently the various mpx structures (mpz_t, mpq_t, mpf_t) have an input function similar to :
mpx_set_str(mpx_t r, const char *sp, int base)
For the sake of performance, I suggest a new prototype mpx_set_str2(mpx_t r, const char *start, const char* end, int base) (where [start;end[ is the string to consider)
For me, the interest is :
-It saves calls to strlen() in the current implementation
-users of gmplib (like me) do not need to create intermediate strings with '\0' terminator when parsing substring from complex input data
Regards,
Pierre Chatelier
# HG changeset patch
# User Pierre Chatelier <pierre at chachatelier.fr>
# Date 1449003901 -3600
# Tue Dec 01 22:05:01 2015 +0100
# Node ID 89fd519f1bac6846bfa54319f416c6c75716ea71
# Parent bb072477386f1a513937aec31cd6fa3e925f59f6
added mpz_set_str2
added mpq_set_str2
added mpf_set_str2
diff -r bb072477386f -r 89fd519f1bac gmp-h.in
--- a/gmp-h.in Tue Nov 24 15:31:26 2015 +0100
+++ b/gmp-h.in Tue Dec 01 22:05:01 2015 +0100
@@ -1032,6 +1032,9 @@
#define mpz_set_str __gmpz_set_str
__GMP_DECLSPEC int mpz_set_str (mpz_ptr, const char *, int);
+#define mpz_set_str2 __gmpz_set_str2
+__GMP_DECLSPEC int mpz_set_str2 (mpz_ptr, const char *, const char *, int);
+
#define mpz_set_ui __gmpz_set_ui
__GMP_DECLSPEC void mpz_set_ui (mpz_ptr, unsigned long int);
@@ -1233,6 +1236,9 @@
#define mpq_set_str __gmpq_set_str
__GMP_DECLSPEC int mpq_set_str (mpq_ptr, const char *, int);
+#define mpq_set_str2 __gmpq_set_str2
+__GMP_DECLSPEC int mpq_set_str2 (mpq_ptr, const char *, const char *, int);
+
#define mpq_set_ui __gmpq_set_ui
__GMP_DECLSPEC void mpq_set_ui (mpq_ptr, unsigned long int, unsigned long int);
@@ -1419,6 +1425,9 @@
#define mpf_set_str __gmpf_set_str
__GMP_DECLSPEC int mpf_set_str (mpf_ptr, const char *, int);
+#define mpf_set_str2 __gmpf_set_str2
+__GMP_DECLSPEC int mpf_set_str2 (mpf_ptr, const char *, const char *, int);
+
#define mpf_set_ui __gmpf_set_ui
__GMP_DECLSPEC void mpf_set_ui (mpf_ptr, unsigned long int);
diff -r bb072477386f -r 89fd519f1bac mpf/set_str.c
--- a/mpf/set_str.c Tue Nov 24 15:31:26 2015 +0100
+++ b/mpf/set_str.c Tue Dec 01 22:05:01 2015 +0100
@@ -122,6 +122,12 @@
int
mpf_set_str (mpf_ptr x, const char *str, int base)
{
+ return mpf_set_str2(x, str, str+strlen(str), base);
+}
+
+int
+mpf_set_str2 (mpf_ptr x, const char *start, const char *end, int base)
+{
size_t str_size;
char *s, *begs;
size_t i, j;
@@ -133,16 +139,21 @@
const char *point = GMP_DECIMAL_POINT;
size_t pointlen = strlen (point);
const unsigned char *digit_value;
+ const char* str = start;
TMP_DECL;
+
+ if (!start || (end <= start))
+ return -1;
- c = (unsigned char) *str;
+ if (str<end)
+ c = (unsigned char) *str;
/* Skip whitespace. */
- while (isspace (c))
+ while ((str<end) && isspace (c))
c = (unsigned char) *++str;
negative = 0;
- if (c == '-')
+ if ((c == '-') && (str<end))
{
negative = 1;
c = (unsigned char) *++str;
@@ -175,16 +186,16 @@
{
/* not a digit, must be a decimal point */
for (i = 0; i < pointlen; i++)
- if (str[i] != point[i])
+ if ((str+i >= end) || (str[i] != point[i]))
return -1;
- if (digit_value[(unsigned char) str[pointlen]] >= base)
+ if ((str+pointlen >= end) || (digit_value[(unsigned char) str[pointlen]] >= base))
return -1;
}
/* Locate exponent part of the input. Look from the right of the string,
since the exponent is usually a lot shorter than the mantissa. */
expptr = NULL;
- str_size = strlen (str);
+ str_size = end-str;
for (i = str_size - 1; i > 0; i--)
{
c = (unsigned char) str[i];
diff -r bb072477386f -r 89fd519f1bac mpq/set_str.c
--- a/mpq/set_str.c Tue Nov 24 15:31:26 2015 +0100
+++ b/mpq/set_str.c Tue Dec 01 22:05:01 2015 +0100
@@ -41,29 +41,32 @@
int
mpq_set_str (mpq_ptr q, const char *str, int base)
{
+ return mpq_set_str2(q, str, str+strlen(str), base);
+}
+
+int
+mpq_set_str2 (mpq_ptr q, const char *start, const char *end, int base)
+{
const char *slash;
- char *num;
- size_t numlen;
int ret;
+ const char* str = start;
+
+ if (!start || (end<=start))
+ return -1;
- slash = strchr (str, '/');
+ slash = memchr (str, '/', end-str);
if (slash == NULL)
{
SIZ(DEN(q)) = 1;
PTR(DEN(q))[0] = 1;
- return mpz_set_str (mpq_numref(q), str, base);
+ return mpz_set_str2 (mpq_numref(q), str, end, base);
}
- numlen = slash - str;
- num = __GMP_ALLOCATE_FUNC_TYPE (numlen+1, char);
- memcpy (num, str, numlen);
- num[numlen] = '\0';
- ret = mpz_set_str (mpq_numref(q), num, base);
- (*__gmp_free_func) (num, numlen+1);
+ ret = mpz_set_str2 (mpq_numref(q), str, slash, base);
if (ret != 0)
return ret;
- return mpz_set_str (mpq_denref(q), slash+1, base);
+ return mpz_set_str2 (mpq_denref(q), slash+1, end, base);
}
diff -r bb072477386f -r 89fd519f1bac mpz/set_str.c
--- a/mpz/set_str.c Tue Nov 24 15:31:26 2015 +0100
+++ b/mpz/set_str.c Tue Dec 01 22:05:01 2015 +0100
@@ -44,6 +44,12 @@
int
mpz_set_str (mpz_ptr x, const char *str, int base)
{
+ return (mpz_set_str2(x, str, str+strlen(str), base));
+}
+
+int
+mpz_set_str2 (mpz_ptr x, const char *start, const char* end, int base)
+{
size_t str_size;
char *s, *begs;
size_t i;
@@ -51,7 +57,11 @@
int c;
int negative;
const unsigned char *digit_value;
+ const char* str = start;
TMP_DECL;
+
+ if (!start || (end<=start))
+ return -1;
digit_value = digit_value_tab;
if (base > 36)
@@ -64,12 +74,15 @@
}
/* Skip whitespace. */
- do
- c = (unsigned char) *str++;
- while (isspace (c));
+ if (str<end)
+ {
+ do
+ c = (unsigned char) *str++;
+ while ((isspace (c)) && (str<end));
+ }
negative = 0;
- if (c == '-')
+ if ((c == '-') && (str<end))
{
negative = 1;
c = (unsigned char) *str++;
@@ -80,19 +93,19 @@
/* If BASE is 0, try to find out the base by looking at the initial
characters. */
- if (base == 0)
+ if ((base == 0) && (str<end))
{
base = 10;
if (c == '0')
{
base = 8;
c = (unsigned char) *str++;
- if (c == 'x' || c == 'X')
+ if ((c == 'x' || c == 'X') && (str<end))
{
base = 16;
c = (unsigned char) *str++;
}
- else if (c == 'b' || c == 'B')
+ else if ((c == 'b' || c == 'B') && (str<end))
{
base = 2;
c = (unsigned char) *str++;
@@ -101,7 +114,7 @@
}
/* Skip leading zeros and white space. */
- while (c == '0' || isspace (c))
+ while ((str<end) && (c == '0' || isspace (c)))
c = (unsigned char) *str++;
/* Make sure the string does not become empty, mpn_set_str would fail. */
if (c == 0)
@@ -111,7 +124,7 @@
}
TMP_MARK;
- str_size = strlen (str - 1);
+ str_size = end-str+1;
s = begs = (char *) TMP_ALLOC (str_size + 1);
/* Remove spaces from the string and convert the result from ASCII to a
Regards,
Pierre Chatelier
Le 29 nov. 2015 à 07:26, Marco Bodrato <bodrato at mail.dm.unipi.it> a écrit :
> Ciao,
>
> Il Ven, 27 Novembre 2015 8:53 pm, Pierre Chatelier ha scritto:
>> Currently the various mpx structures (mpz_t, mpq_t, mpf_t) have an
>
>> For the sake of performance, I suggest a new prototype
>> mpx_set_str2(mpx_t r, const char *start, const char* end, int base)
>> (where [start;end[ is the string to consider)
>
> The prototype of the underlying mpn function is
> mp_size_t mpn_set_str (mp_limb_t *rp, const unsigned char *str, size_t
> strsize, int base)
>
> If you use start:end, you need to explain: *end is included or not?
>
>> For me, the interest is :
>
>> -It saves a call to strlen() in the current implementation
>> -users of gmplib do not need to create intermediate strings with
>> '\0' terminator when parsing substring from complex input data
>
> From mpq/set_str.c:
> /* FIXME: Would like an mpz_set_mem (or similar) accepting a pointer and
> length so we wouldn't have to copy the numerator just to null-terminate
> it. */
>
> Regards,
> m
>
> --
> http://bodrato.it/papers/
>
_______________________________________________
gmp-discuss mailing list
gmp-discuss at gmplib.org
https://gmplib.org/mailman/listinfo/gmp-discuss
More information about the gmp-discuss
mailing list