Someone can add FLT_MIN DBL_MIN detection ?

[Daniel Marjamäki]

TBD

[Absol]

FLT_MIN DBL_MIN - is positive value.
Many developers use it as minimum (negative) floating point value(like INT_MIN) and it fully wrong and can cause undefined behavior.
possible to add warn if it found in code?

Many examples from Valve(thanks for this examples for bad valve developers :) :) )
Examples: https://github.com/ValveSoftware/source-sdk-2013/issues/519 or https://github.com/perilouswithadollarsign/cstrike15_src/search?q=FLT_MIN

[Daniel Marjamäki]

I would like an example code that has a bug .. if possible 4-5 lines of code.. the smaller the better.

I wonder how do we detect that FLT_MIN is used wrongly? If it cause undefined behavior I think we can warn about that. But if the code just says std::cout << FLT_MIN; or something like that then what is wrong with that?

[Absol]

You can see examples, and many examples, above )
or it simple example:

BAD

#include <algorithm>
#include <iostream>
#include <string>
#include <cfloat>

float test[3]{0.0,-1.5,-5.5};

int main()
{
    float minvalue = FLT_MAX;
    float maxvalue = FLT_MIN;

    for(int i = 0; i < 3; i++)
    {
        minvalue = std::min(test[i],minvalue);
        maxvalue = std::max(test[i],maxvalue);
    }

    std::cout << minvalue << std::endl << maxvalue ;

}

TRUE

#include <algorithm>
#include <iostream>
#include <string>
#include <cfloat>

float test[3]{0.0,-1.5,-5.5};

int main()
{
    float minvalue = FLT_MAX;
    float maxvalue = -FLT_MAX;

    for(int i = 0; i < 3; i++)
    {
        minvalue = std::min(test[i],minvalue);
        maxvalue = std::max(test[i],maxvalue);
    }

    std::cout << minvalue << std::endl << maxvalue ;

}

In this example, vecMax can not be

[Daniel Marjamäki]

I feel a bit skeptic about this. So the heuristic that we should warn about is when a float variable is assigned FLT_MIN and then the variable is compared against some float values in a loop.. and the max value is stored in the variable..

Even this feels inconclusive for me. If you replace FLT_MIN with 0.0f in your example code I assume no warning should be written. However the code would technically be almost exactly the same. Maybe there are some cases when it is intentional to use FLT_MIN.

[Daniel Marjamäki]

I have created https://trac.cppcheck.net/ticket/10411

When this check is added I think a careful investigation will be needed to see that the noise ratio is acceptable. I don't want to warn about some FLT_MIN pattern if that is used by intention in various places.

[Absol]

I think 'if DBL_MAX and DBL_MIN found in same code block, write warn about DBL_MIN is 'Minimum representable positive floating-point number.' (or FLT_MIN)

[Daniel Marjamäki]

Thanks! I mentioned that heuristic in the ticket.

[Absol]

You can see examples above in valve repositories (This error happens very often with them)

https://github.com/ValveSoftware/source-sdk-2013/issues/519
https://github.com/perilouswithadollarsign/cstrike15_src/search?q=FLT_MIN

[Absol]

https://www.cplusplus.com/reference/cfloat/
https://stackoverflow.com/questions/7973737/why-are-flt-max-and-flt-min-not-positive-and-negative-infinity-and-what-is-thei
https://stackoverflow.com/questions/2528039/why-is-flt-min-equal-to-zero

[Absol]

You can use http://cpp.sh/ for test it