Which would be faster?

05-31-2006, 03:59 AM

Let say I wanted to make a list.

I could use gamecache, or I could hijack a location.
The list api is retardedly simple and it would work as follows

If I used a gc, then I need to make it a node class, with two functions Head (car/getdata), and Tail (cdr/getnext). As far as a null list, I could simply create a special gc object that when passed to a function isNull returns true.

For locations its actually a bit easier:
function car takes location l returns integer
return GetLocatonX(l)
return 0
endfunction
Tail=GetY
function cdr takes location l returns location
return GetLocationY(l)
return null
endfunction
function IsNull takes location l returns boolean
return l==null
endfunction

Vex says most of the slowdown of gamecache is return bug, but what about other data structures. I'm only wrapping GetX with integer since almost everybody uses H2I.

05-31-2006, 04:27 AM

weaaddar

Okay actually locations seem to be pretty fucked up, check out this sample code:

JASS:

function H2R takes handle h returns real
    return h
    return 0.0
endfunction

function ItoR takes integer i returns real
    return i
    return 0.
endfunction

function car takes location list returns integer
    return GetLocationX(list)
    return 0
endfunction

function cdr takes location list returns location
    return GetLocationY(list)
    return null
endfunction

function IsNull takes location list returns boolean
    return list==null
endfunction

function cons takes integer d, location l returns location
    return Location(ItoR(d),H2R(l))
endfunction

function PrintList takes location l returns nothing
    if(not IsNull(l))then
        call BJDebugMsg(I2S(car(l)))
        call PrintList(cdr(l))
    endif
endfunction

function Test takes nothing returns nothing
    local location l=cons(5,null)
    set l=cons(7,l)
    set l=cons(23,l)
    set l=cons(0,l)
    if(car(cdr(l))==203 and car(cdr(l))==0)then
        call BJDebugMsg("Wow this is fubared!")
    else
        call PrintList(l)
    endif
endfunction

call function test, it seems that when using typecasting to reals the value could equal Anything and the system considers it peachy. How IsNull works perfeclty fine may just be a fantastic mystery. This is what pipedream posted earlier. I do find it pretty lol worthy that you can violate the law of contradiction and all sorts of other nasty things with this "wonderful" quirk.

05-31-2006, 04:35 AM

weaaddar

Well the fix is pretty silly but I read what pipedream said, and I figured it out I realize that this is a triple post but I figure this is worth nothing.

The idea is you have to do store the values as explicit whatevers whenever you use real as your dummy data spot. To get around this I simply built generic typecasters, and then I create a local version in the calls car and cdr this fixes the problem. I still have no idea WHY IsNull worked perfectly, when car and cdr would cause such bizzare problems.

JASS:

function H2R takes handle h returns real
    return h
    return 0.0
endfunction

function R2H takes real r returns handle
    return r
    return null
endfunction    
function ItoR takes integer i returns real
    return i
    return 0.
endfunction

function RtoI takes real r returns integer
    return r
    return 0
endfunction
function car takes location list returns integer
    local integer i= RtoI(GetLocationX(list))
    return i
endfunction

function cdr takes location list returns location
    local handle h=R2H(GetLocationY(list))
    return h
endfunction

function IsNull takes location list returns boolean
    return list==null
endfunction

function cons takes integer d, location l returns location
    return Location(ItoR(d),H2R(l))
endfunction

function PrintList takes location l returns nothing
    if(not IsNull(l))then
        call BJDebugMsg(I2S(car(l)))
        call PrintList(cdr(l))
    endif
endfunction

05-31-2006, 05:32 AM

PipeDream

Maybe it's the not operator that makes it happy.
If you want list behavior, it should be about the same speedwise. I'd go with the locations since you can deallocate 100% of the memory and you won't have a first time I2S hit.

JASS:

function cdr takes location list returns location
    local handle h=R2H(GetLocationY(list))
    return h
endfunction

Pretty sure this will take hostages. Probably got to

JASS:

function cdr takes location list returns location
    local integer h=RtoI(GetLocationY(list))
    return h
        return null
endfunction

For sanity and consistency I've started using "to" instead of "2" for all return bugging casting.

06-01-2006, 03:24 AM	#5
Vexorian	the closest thing we had to return bug exploiter naming standard was the 2

06-01-2006, 03:28 AM	#6
weaaddar	yeah, but then what about Real->Int bugged caster? We can't really call it R2I.

06-01-2006, 04:08 AM

Vexorian

no wonder why I hate return bug exploiters. to is just to long so long that we could go for more verbose and readable ones instead like RealToInt , we could go r2i and use lower case for return bug exploiters. We could see lower case as for reference type conversions and Upper case for the value conversions. But well not like we would ever have an standard for this.

Edit: If you only need a single linked list, locations would be better. Is the most slow part of the handle variables and similar behaviours, but gamecache is still very slow and a list that uses locations should be faster than one that uses gamecache.