_matherr doesn't get invoked when sscanf is called.

[ishekara]

Hi,

As per the msdn knowledge base.. i find the following

"In an application developed with Microsoft C or C/C++, the sscanf()
function is a good alternative to the atof() function to convert a string of
digits into a floating-point number. If a string does not represent a valid
number, atof() returns the value zero; sscanf() returns more useful error
information. The application can use the error value from sscanf() with the
matherr() function to perform error handling. The atof() function does not
call matherr() unless an actual math exception occurs. "

But even if the string is not correctly converted to float the _matherr()
function is not being called. Is this a bug or something wrong?

float f;

char temp[]="1mmm1";

int i=sscanf(temp,"%g",&f);

Here f get assigned with 1 instead i was looking for an error.

Any ideas how to solve this



Thanks

ishekara

 

[Emmanuel Delahaye]

As per the msdn knowledge base.. i find the following

Huh! Why not reading a good C-book in the first place?

www.accu.org

"In an application developed with Microsoft C or C/C++, the sscanf()
function is a good alternative to the atof() function to convert a
string of digits into a floating-point number. If a string does not
represent a valid number, atof() returns the value zero; sscanf()
returns more useful error information. The application can use the error
value from sscanf() with the matherr() function to perform error
handling. The atof() function does not call matherr() unless an actual
math exception occurs. "

But even if the string is not correctly converted to float the
_matherr() function is not being called. Is this a bug or something
wrong?

float f;

char temp[]="1mmm1";

int i=sscanf(temp,"%g",&f);

Here f get assigned with 1 instead i was looking for an error.

It's not really an error. But you could pre-check the domain with

{
char checked[128];
int ret = sscanf (checked, "%[0-9-+.eE]s", temp);

if (ret == 1)
{
int i = sscanf (checked, "%g", &f);
<...>
}
}

but it wont prevent against malformed expressions like

"12+3" or "45e"

Any ideas how to solve this

Yes. Better to use strtod() and to learn how to use the 2nd parameter. It's
worthy.

 

[Eric Sosman]

Emmanuel

Yes, it is.

Well, no: it's only good for *some* `float' values.
strtod() will happily and successfully convert a string
to a `double' that can't in turn be converted to `float'.
For example, on many implementations the string "1E300"
is acceptable to strtod(), but yields a value larger
than FLT_MAX.

C99 introduced strtof(), which might be available on
ishekara's implementation (it's worth looking, anyhow).
If it isn't, I think the best he can do is to use strtod()
with the usual error checks, and then either hope for the
best or add further checks for range and perhaps precision:

float flt;
double dbl = strtod(...);
/* usual checks here */
if (fabs(dbl) > FLT_MAX
|| (fabs(dbl) < FLT_MIN && dbl != 0.0)) {
/* `double' value outside `float' range */
}
else {
flt = dbl;
}

.... but on systems that support infinities and/or NaNs,
even this much isn't quite enough.