Bug 22527 – Casting out-of-range floating point value to signed integer overflows

As of DMD 2.098.0, the following program fails to compile: --- static assert(cast(int) float.max > 0); --- The error message is: --- Error: static assert: `-2147483648 > 0` is false --- According to the language spec: > Casting a floating point value to an integral type is the equivalent of converting to an integer using truncation. Since float.max is a positive number, truncation of its value should result in a positive integer, not a negative integer.

The problem here is that the integral value of float.max exceeds int.max, so it's a truncation followed by an integer overflow.

Compiling this example in C: ``` #include <float.h> int main() { float f = FLT_MAX; int i = (int)f; return 0; } ``` and then looking into the assembly, you can see the cvttss2si instruction is used to perform the cast. Reading from the second paragraph of https://www.felixcloutier.com/x86/cvttss2si#description we can find the answer: 80000000H (or INT_MIN) is the result in case of floating-point invalid exception. As this is defined directly in the x86 ISA and is the default behavior in C, this will also be the case for D.

@sorin-gabriel created dlang/dlang.org pull request #3140 "fix Issue 22527 - Casting out-of-range floating point value to signed…" fixing this issue: - fix Issue 22527 - Casting out-of-range floating point value to signed integer overflows https://github.com/dlang/dlang.org/pull/3140

dlang/dlang.org pull request #3140 "fix Issue 22527 - Casting out-of-range floating point value to signed…" was merged into master: - 05cb1035324a62ffab1467fad7eea8d68ca41520 by Gabriel: fix Issue 22527 - Casting out-of-range floating point value to signed integer overflows https://github.com/dlang/dlang.org/pull/3140