Hashlink floating point exception with remainder operator

[PXshadow]

function main() {
	var x = (-1 << 31);
	var y = -1;
	trace(x % y);
}

Haxe nightly: c4868e5
Hashlink: 1.14.0

[Apprentice-Alchemist]

In this case floating point exception actually means integer overflow.
This is an edge case that needs to be worked around either in genhl or in the JIT.

[yuxiaomao]

I confirm that it's an integer overflow, and can be reproduce with hl/c too if the optimization is disabled.
This is indeed an edge case :(

https://wiki.sei.cmu.edu/confluence/display/c/INT33-C.+Ensure+that+division+and+remainder+operations+do+not+result+in+divide-by-zero+errors
Overflow can occur during a remainder operation when the dividend is equal to the minimum (negative) value for the signed integer type and the divisor is equal to −1.

[ncannasse]

Interesting, that's a new class of exception that I just learnt.
In 32 bits it only occurs with these two values exactly, so I guess the best choice is to generate HL/C + JIT code to test divide by (-1) after we checked 0 and if it's the case perform a IMULT instead of a IDIV.

[yuxiaomao]

I can't see a case where % -1 does not return 0, except when overflow occurs; so I'm going to say : r = (b == 0 || b == -1) ? 0 : a % b in HLC(haxe)/JIT(hashlink)

[yuxiaomao]

Nicolas explained to me x)

OSMod: r = (b == 0 || b == -1) ? 0 : a % b
OSDiv: r = (b == 0 || b == -1) ? a * b : a / b with IMUL

OSDiv can be triggered with the following code:

static function main() {
	var x = -1 << 31;
	var y = -1;
	var z = Std.int(x / y);
	trace(z);
}