[Bug c/99707] New: missing -Woverflow in floating-point to integer conversion for known but non-constant value

[Bug c/99707] New: missing -Woverflow in floating-point to integer conversion for known but non-constant value

[vincent-gcc at vinc17 dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99707

            Bug ID: 99707
           Summary: missing -Woverflow in floating-point to integer
                    conversion for known but non-constant value
           Product: gcc
           Version: 10.2.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: vincent-gcc at vinc17 dot net
  Target Milestone: ---

Consider the following code (from PR93806 Comment 29):

#include <stdio.h>

int main (void)
{
  volatile double d = -1.0;
  double x = d;
  unsigned int i = x;

  printf ("%u\n", i);
  if (x == -1.0)
    printf ("%u\n", i);
  return 0;
}

First note that if instead of "unsigned int i = x;", one has "unsigned int i =
-1.0;", then one gets a warning with GCC 10.2.1:

conv-warn.c: In function ‘main’:
conv-warn.c:7:20: warning: overflow in conversion from ‘double’ to ‘unsigned
in’ changes value from ‘-1.0e+0’ to ‘0’ [-Woverflow]
    7 |   unsigned int i = -1.0;
      |                    ^

But the original code does not give any warning, even though with -O1, GCC uses
the fact that x == -1.0 to optimize and give a strange result (because the
variable i appears to have two different values):

4294967295
0

Thus GCC should know about the undefined behavior and warn.

[Bug c/99707] missing -Woverflow warning in floating-point to integer conversion for known but non-constant value

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99707

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Gcc does not know that x will be negative. 

Note: x == -1.0 is well defined. As x is converted to double from unsigned int.

-Wconversion might warn about this but I have not tried.

[Bug c/99707] missing -Woverflow warning in floating-point to integer conversion for known but non-constant value

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99707

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Also this might be caught at runtime with -fsanitizer=undefined but I have not
tried yet. Since this is undefined behavior in this case unlike the original
PR93806.

[Bug c/99707] missing -Woverflow warning in floating-point to integer conversion for known but non-constant value

[vincent-gcc at vinc17 dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99707

--- Comment #3 from Vincent Lefèvre <vincent-gcc at vinc17 dot net> ---
(In reply to Andrew Pinski from comment #1)
> Gcc does not know that x will be negative.

Actually, the code should have been (without the volatile):

#include <stdio.h>

int main (void)
{
  double x = -1.0;
  unsigned int i = x;

  printf ("%u\n", i);
  if (x == -1.0)
    printf ("%u\n", i);
  return 0;
}

I don't get the warning either, and due to the "double x = -1.0;", GCC knows
that x is negative.

> -Wconversion might warn about this but I have not tried.

The test is done with -Wconversion.

(In reply to Andrew Pinski from comment #2)
> Also this might be caught at runtime with -fsanitizer=undefined but I have
> not tried yet. Since this is undefined behavior in this case unlike the
> original PR93806.

After checking... This is not undefined behavior, but the value is unspecified.
F.4: "[...] if the floating value is infinite or NaN or if the integral part of
the floating value exceeds the range of the integer type, then the “invalid”
floating-point exception is raised and the resulting value is unspecified."

[Bug c/99707] missing -Woverflow warning in floating-point to integer conversion for known but non-constant value

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99707

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The front end does not normally do constant prop if the variable is not a
const/constexpr so the warning would be done in the middle end and might have
false warnings so it might not be useful.