[Snippet] GetUnitCollision

[Nestharus]

This resource has moved here.

--

gone

 

[Bribe]

This script is still using the unsafe ObjectMerger one-liner, if you plan to use this as an alternate library to http://www.wc3c.net/showthread.php?t=101309 it could stand fixing.

 

[Nestharus]

I'll update it when I can... and that's the first time I've seen that alternative library... wow, it's really bad o-o.

You could always be a pal and update it for me ^)^.

 

[Bribe]

I think the whole thing should just hash all the collision sizes using a LUA generator, since those are static Object Editor things anyway. Requiring placing the location in a certain "perfect" spot is less appealing than just running a simple loop like Vexorian has, to be honest. It won't run more than 50% slower but it doesn't require a location, unit, external libraries... a lot easier to implement. As it stands, it would be fine as an alternative but probably not going to turn any heads.

 

[themerion]

• Triggers Enter/Leave region events
• Returns the wrong sign (- instead of +)
• Is less accurate than Vexorian's version
• (Breaks channels...?)

 

[Nestharus]

I'm starting to think that 128 refers to the diameter rather than the radius, which would explain the negative sign return as well as how it is less accurate. The current algorithm would return something 100% accurate since it uses wc3's collision detection.

So -128 should be -64 =D.

 

[themerion]

Nah, that gave me collision size ~50... Using -128 gave (-13.5) which seem closer (if you do abs). GetUnitCollisionSize gives 15.9...

The desired output is 16, btw.

Oh, and I'm almost certain that moving a unit will make it stop channeling (besides from possibly triggering region events).

 

[Magtheridon96]

This should be GetWidgetCollision =P

 

[Bribe]

Item collision is always 16 last I checked, not sure what destructable collision is.

 

[WaterKnight]

Destructables may only have a pathing map.

 

[Bribe]

So does this have a future?

 

[Nestharus]

So does this have a future?

At some point yes

 

[Weep]

Comments from a quick look:
1. It won't work for flying units, will it?
2. Moving the test unit is dangerous, because it might trip exit-region events or stop Drain Life, not to mention interrupting that unit with SetUnitPosition. Instead...
3. ...shouldn't it take unit-type and create a new test subject?
4. It'll consider pathing maps, and whether that's desired or not, they won't necessarily be a circle (no single collision value).

5. Or, all that could be avoided, including the need for a special tester unit-type at all, by using the method used in this script: iterating distances from the unit's position until IsUnitInRangeXY returns false, since that function considers a unit's collision size.

 

[Nestharus]

That one has a lot higher overhead :\, like I said, I'll get back to this when I can.

 

[Bribe]

This is about as light as it gets - a hybrid of Vexorian's GeUnitCollisionSize with some of Bribe's crazy ideas. It iterates 8 or 9 times at most. The only disadvantage is that it works entirely in integers, so decimals will be cropped off.

library GetUnitCollision requires Table
//Created by Vexorian with some improvements by Bribe
    
    globals
        //-------------------------------------------------------------------
        // You can optimize this to be the max collision in your map
        // 172 is max natural collision - from tier 3 town halls.
        private constant integer MAX_COLLISION = 172
        
        // A Table for an optimization perk.
        private Table t = 0
    endglobals
    
    //=======================================================================
    private function Check takes integer id, unit u returns integer
        local integer z = MAX_COLLISION
        local integer a = 0
        local integer i = z
        local real x = GetUnitX(u)
        local real y = GetUnitY(u)
        loop
            if IsUnitInRangeXY(u, x + i, y, 0) then
                set a = i
            else
                set z = i
            endif
            exitwhen z - a <= 1
            set i = (z + a) / 2
        endloop
        set i = i - z + a
        set t[id] = i
        return i
    endfunction
    
    //=======================================================================
    // Golden function - optimized for fast repeat returns on type id's.
    //
    function GetUnitCollision takes unit u returns integer
        local integer id = GetUnitTypeId(u)
        if not t.has(id) then
            return Check(id, u)
        endif
        return t[id]
    endfunction
    
    //=======================================================================
    private module I
        private static method onInit takes nothing returns nothing
            set t = Table.create()
        endmethod
    endmodule
    private struct S extends array
        implement I
    endstruct
    
endlibrary

 

[Bribe]

Opening thread

 

[Nestharus]

Updated so that this now actually works.

 

[Magtheridon96]

        set uc.real[i] = R2I(m*10)/10.
            
        return R2I(m*10)/10.

->

set m = R2I(m*10)/10.

set uc.real[i] = m

return m

I think that should be faster ;D (and cleaner)

 

[Nestharus]

Done, tx

 

[Bribe]

I think you should split this into two functions like I did - one for fast returns if the collision size already had been retrieved, if not then call the function which calculates the collision size again.

 

[Nestharus]

Is it ok now?

 

[Bribe]

~ Approved

Need I say more?

 

[Dirac]

Could you add a constant for the max unit collision so we can pick groups in range + max collision and then decide if the unit's collision matches the actual range?

edit
to be more specific

function TestGroup takes real x, real y, real r returns nothing
	local unit u
	local real ux
	local real uy
	call GroupEnumUnitsInRange(bj_lastCreatedGroup,x,y,r+MAX_COLLISION_SIZE,null)
	loop
		set u = FirstOfGroup(bj_lastCreatedGroup)
		exitwhen u==null
		set ux = GetUnitX(u)-x
		set uy = GetUnitY(u)-y
		// check if the unit collides with the picking radius
		if SquareRoot(ux*ux+uy*uy)-GetUnitCollision(u) <= r then
			// Do Actions...
		endif
		call GroupRemoveUnit(bj_lastCreatedGroup,u)
	endloop
	set u = null
endfunction

 

[WaterKnight]

Is there any mean to change the collision size ingame? I just cache it in my map.

 

[Bribe]

The unit type ID collision has never changed from my experience. You can disable pathing, or add locust, but otherwise it's not going to change much.

 

[Nestharus]

Could you add a constant for the max unit collision so we can pick groups in range + max collision and then decide if the unit's collision matches the actual range?

max realistic collision is 160, so 300 is fair.

 

[Bribe]

@Nes, human town halls have 197 collision by default. That is the largest default object editor collision size.

 

[Nestharus]

@Nes, human town halls have 197 collision by default. That is the largest default object editor collision size.

but the pathing map is 160.

 

[BBQ]

but the pathing map is 160.

What the hell? Their pathing map is 512x512 WC3 units. Each pixel in the *.tga file represents 32 WC3 units.

 

[Nestharus]

Eh, you are right ; P. I wasn't thinking while reading default object editor collision size for the castles. Just did the math really quick based on the stuff I did in IsPathBlocked and it indeed comes out to 512 ; ).

Castle collision size doesn't match its pathing map ; D. I should probably fix that in IsPathBlocked ^)^.


Figured Blizzard would at least do the collision sizes right relative to the pathing maps, but guess I was wrong >.>.


However, the collision size for the 512 would still be 256 (radius), so 300 is still ok.


Thought that 160 radius was pretty strange while writing IsPathBlocked ; P.

 

[Nestharus]

Added the GetUnitCollision req at top ^)^.

 

[Magtheridon96]

O_O
Nes, you posted the wrong code in the wrong thread.

 

[Nestharus]

Yea, I saw that just now... can't put code back up until later since I don't have GetUnitCollision on this comp.

edit
thanks for putting up the code in pastebin mag ;p