Warnings in presence of Top floats

ID0000997: This issue was created automatically from Mantis Issue 997. Further discussion may take place here.

---

Id	Project	Category	View	Due Date	Updated
ID0000997	Frama-C	Plug-in > Eva	public	2011-10-23	2012-09-19

Reporter	yakobowski	Assigned To	yakobowski	Resolution	fixed
Priority	normal	Severity	minor	Reproducibility	always
Platform	-	OS	-	OS Version	-
Product Version	Frama-C Nitrogen-20111001	Target Version	-	Fixed in Version	Frama-C Oxygen-20120901

Description :

Consider the following program analyzed with frama-c -val -absolute-valid-range 0-100

const short max = 255; const short min = 0;

extern float u[256];

float main () { short r;

float f = *((float *)18);

// assert -3.40282346639e+38 <= f <= 3.40282346639e+38; int j = 2; // To be able to see the value of f in the gui

if(f >= max) r = max; else if (f <= min) r = min; else r = f + 0.5;

return u[r]; }

The results are not optimal, as the access to u[r] is not proven correct. The reason is as follows: with a "normal" float, the reductions within the 'if' take place normally, and r belongs to [0..255]. However, here we have a Topint inside a float, and the reduction does not happen.

If we activate the assertion to help the user, the result is worse: the assertion is incorrectly proven correct, as the conversion to a finite float happens before the truth value of the assertion is computed. Moreover, since the assertion is "correct", no reduction by the assertion occurs, and the access to u[r] is still not correct.