Neko overflow problem

Ali Hassan

hello everyone ..

so I needed FlxNapeTileMap into my current project .. and mostly it worked prefectly (great thanks for those who worked on it :smiley: )

but it seemed to crash on neko for certain maps in a very weird way althoug it will work nicely on linux cpp & flash. this crash only happens on certain layouts though . I tried to find a pattern to it but I couldn't. so whenever I faced it I kept changing different tiles until it worked again (which is not acceptable of course).

here is the neko crash log ->

Invalid field access : elem
Called from zpp_nape/shape/Polygon.hx line 923
Called from zpp_nape/shape/Polygon.hx line 1115
Called from zpp_nape/space/Space.hx line 4944
Called from zpp_nape/space/Space.hx line 2405
Called from zpp_nape/space/Space.hx line 1821
Called from nape/phys/BodyList.hx line 455
Called from flixel/addons/nape/FlxNapeTilemap.hx line 256
Called from flixel/addons/nape/FlxNapeTilemap.hx line 143
Called from map/TiledLevel.hx line 106
Called from a C function
Called from map/TiledLevel.hx line 29
Called from state/mode/collect/CollectCoinsSinglePlayerMode.hx line 30
Called from flixel/FlxGame.hx line 623
Called from flixel/FlxGame.hx line 713
Called from flixel/FlxGame.hx line 663
Called from flixel/FlxGame.hx line 530
Called from openfl/_legacy/events/EventDispatcher.hx line 98
Called from a C function
Called from openfl/_legacy/display/DisplayObject.hx line 161
Called from a C function
Called from openfl/_legacy/display/DisplayObjectContainer.hx line 286
Called from openfl/_legacy/display/Stage.hx line 1103
Called from openfl/_legacy/display/Stage.hx line 351
Called from openfl/_legacy/display/Stage.hx line 1084
Called from openfl/_legacy/display/Stage.hx line 430

so today I was planning to fix that problem .. I started to follow the call stack.
and then three hours and 40 traces() later .. I found the line causing the problem.
without going into much details this check should evaluate to true

if(crs*crs>=Config.epsilon*Config.epsilon)

tracing both crs & config.epsilon results in 49152 and1e-08 respectively
now you would think that the last condition would result in true easily, right ?

WRONG because in neko -> 49152*49152 = -1879048192 ._.
overflowing seems to be a know issue with Neko .. Ref

so I was finally able to get it to work by just using crs for comparison instead of crc*crs
but I would like to know better ways to do it. also I would like to know what should I do about that change .. should I push it somewhere for example ?
thanks

starry-abyss

@Ali-Hassan I think one should use Int64 or Float (which is actually double in Haxe, thanks, didn't know it before your question :D).

alihassan

for the sake of completeness i made a simple test class in Neko and here are the results

class Test {
    static function main() {
        var x:Int        = 49152;
        var y:Float      = 49152;
        var z:haxe.Int32 = 49152;
        var w:haxe.Int64 = 49152;
        trace(x*x);
        trace(y*y);
        trace(z*z);
        trace(w*w);
    	trace(x*1.0*x);
        trace(y*1.0*y);
        trace(z*1.0*z);
        trace(w*1.0*w);
    }
}

here is the logs :

Test.hx:7: -1879048192
Test.hx:8: -1879048192
Test.hx:9: -1879048192
Test.hx:10: 2415919104
Test.hx:11: 2415919104
Test.hx:12: -1879048192
Test.hx:13: 2415919104
Test.hx:14: 2415919104