HomeUser Control Panel (unavailable in archive)ForumsTutorialsArt GalleryResourcesMaps

Automatic Memory Leak Destroyer

08-30-2009, 03:20 PM#1
Mr.Malte
Well, this is a system I wrote mainly for GUI users.
It is not very useful for JASS, but it makes coding in GUI so much easier.
And when you make maps, you dan't do everything in JASS.
You shold write the systems in vJass, but not everything.

This is a tool that automatically detects and kills memory leaks caused by these handle types:

Unitgroup, Effect, Location

which are mainly used in GUI.

How does it work?
Using native hooking it responses to any function that uses groups, effects or locations (in Jass you don't need locations, don't use the BJ functions for effects and never destroy groups) and registers them. After a few seconds they get destroyed.
I could also make it work with Jass, but that would be senseless, because you don't want your groups in JASS to be destroyed automatically. Actually that would break many systems.

And when you don't want the data to be destroyed, you can simply call
ProtectHandle or ProtectVariable on them.
That will prevent the system catching the data that the variables are filled with.

So this catches only BJ functions!
It will not break JASS Systems!
It will just remove memory leaks the user didn't destroy on his own.

Here is the code:

Collapse JASS:
library MemoryLeakHelper initializer Init requires Table
// ==================================
// Give credits to Mr.Malte when used!
//===========================================================================
// Information: 
//==============
//
//  There are things called 'memory Leaks'. When you create a group or use a location
//  without destroying it, you will cause lag that stays in the whole game.
//  If you implement this library into your map it will automatically fix a big part
//  of those memory leaks, so reduce the lag extremely.
//  This should mainfully be used by GUI-users because the system is designed for them.
//
//  Of course no system can work totally automatically.
//  But there is only one thing you have to do in order to prevent bugs:
//  If you make groups or locations that have to be filled for more than CLEAN_UP_INTERVAL seconds
//  you have to save them with the code:
//
//  call ProtectHandle(XXX)
//
//  Where XXX is filled with your variable. Otherwise that variable gets destroyed
//  automatically. You can also fill in XXX with
//
//  GetLastCaughtHandle()
//
//  But if you save the things in a variable, I'd recommend to directly put the 
//  variable into the brackets. Note: GUI variables have the prefix 'udg_' in JASS.
//
//     This gives the 'Do Nothing' function in GUI a sense!
//     When you call DoNothing, all data that were caught by this system
//     will be destroyed in CLEAN_UP_INTERVAL seconds, ignoring how big
//     the number of caught handles is. This will not work, if the system is
//     already cleaning up.
//===========================================================================
// Implementation: 
//===============
//
//  The easiest thing is to directly implement this thing into your map, when you start making
//  it, so you don't have to look over your globals and use ProtectHandle on them.
//  These are the steps you have to do to clear the memory leaks:
//
//  1. Download a tool called 'JassNewGen', and unpack it somewhere. You need that
//     edit to use this tool. The 'JassNewGen' is used very commonly and offers other
//     nice features. You can find it at:
//     [url]http://www.wc3c.net/showthread.php?t=90999[/url]

//  2. Make a new trigger, and convert it to custom text. Insert everything
//     the library contains into that trigger.
//
//  3. Download a system called 'Table' from this link:
//     [url]http://www.wc3c.net/showthread.php?t=101246[/url]
//     Do the same installation stuff for 'Table' as for this system.
//
//  4. Save your map and enjoy :-)
//    
//  Note: Instead of doing 2 and 3 you can also copy and paste the folder 'MemoryLeakHelper' 
//        from the example map.
//===========================================================================
// How bad are memory leaks? 
//==========================
//  If you don't remove memory leaks, they suck memory:
//
//  Location:  0.361 kb
//  Group:     0.62 kb + 0.040 kb for each unit in the group.
//  Effect:    11.631 kb
// 
// Both, locations and groups are used very frequently. So when you don't fix those memory leaks,
// you will experience lag.

//        When you want to see, how useful this is for your map, implement it
//        and write 'call DisplayLeaks()' into a custom script that is fired when
//        they game ends.
//===========================================================================
// Changelog:
//===========
// v1.00 --- first version
// v1.01 --- able to detect special effects, too now.
// v1.02 --- made the system safer and reduced the number of variables to protect greatly.
// v1.03 --- Gave a sense to 'DoNothing'* GUI function and made the Pass Data part
//           more accurate, so the time until data get destroyed are much more explicit
//           now.
// v1.04 --- Added the very important constant MAX_LEAK_INSTANCES
//
// *if you don't want it to be hooked, comment line 350.
//===========================================================================
// FAQ:
// ====
//  1. Why don't you hook functions like GetLocationX or the ForGroup without BJ?
//
//      Answer: Well, in jass you would never destroy groups, rather have one global group
//              and clear/recycle it. But GUI always creates new groups with the functions.
//              So actually, jass groups don't have to be destroyed.
//              And special effects are mostly instantly destroyed and locations are never used.
//              So I don't want to endanger breaking jass systems, I rather make it for GUI, where it is
//              really useful and neccessary
//
//  2. Why should I protect my variables instead of killing my leaks?
//
//      Answer: In GUI, unitgroup effect and location variables are actually just used
//              for destroying stuff. It is rare that you really want to keep the groups.
//              So in fact, it is like one-hundred times less frequent that you want to keep
//              your data instead of destroying it.
//
//  3. I can't use jass. How can this system be useful for me?
//
//      Answer: This system works mainly automatically; You just have to use jass very rarely (and then a simple function).
//              The functions you need are:
//
//                  ProtectVariable(udg_###)
//                  GetLastCaughtHandle()
//
//              where ProtectVariable saves something you want to keep from getting destroyed. Just replace ### with your
//              variable name. NOTE: You don't have to protect the variable, when you want to keep the data inside less than
//              CLEAN_UP_INTERVAL seconds.
//
//              and where GetLastCaughtHandle responses to the handle* that was used lastly.
//              That can be for example the point where you just spawned a unit.
//              
//      * 'handle' means a specialeffect, a location or a unitgroup
//
//  4. If you give functions like 'DelayMMH', why don't you add functions like 'Disable/EnableMMH'?
//
//      Answer: Well, I want to protect the user. You do not need Disable/Enable functions.
//              To me the danger is too big, that you forget to activate it again or do
//              something like that. DelayMMH is totally enough if you don't want this system
//              to affect the code that comes next. Also that prevents, that there are
//              spells like 'Trojan Horses' that get too much access to this and can
//              change it's infrastructure.
//
//===========================================================================
// Functions:
//==========
// ProtectHandle              : Saves a handle from getting destroyed 
// ProtectVariable            : Same.
// DoNothing()                : Destroys all data caught by the system right now in X seconds.
// DelayMMD()                 : Stops the system working until the trigger ends/next wait *
//
// * This is as fast as an automatic memory leak destroyer can get. Why should
// you want to disable the system? Because it offers the possibilty to make things
// more efficient. I don't want to say, this is unefficient, because it is not.
// But this will destroy leaks like 10% slower.
//
//===========================================================================


    globals
        // The system fires when you do something that creates a leak.
        // The data that cause leak are saved in a variable then.
        // And every CLEAN_UP_INTERVAL seconds those data are destroyed.
        // This shouldn't be too high, or too low.
        private constant real CLEAN_UP_INTERVAL = 120.
        // If this is set to true, the system will work more slowly (but you wont notice)
        // and count, how much memory this system was able to save.
        // This value is display by the function DisplayLeaks() then.
        // WARNING: This sucks a lot of performance. I would ONLY use it when you want
        // to test, if this is useful for your map. Later set it to false.
        private constant boolean DISPLAY_SAVED_MEMORY = false
        // The Data are only cleaned up, when that many handles were caught
        private constant integer MIN_LEAK_NUMBER = 1750
        // How often are data passed to the destroyer?
        // Leaks stay for a random time between CLEAN_UP_INTERVAL and CLEAN_UP_INTERVAL+PASS_INTERVAL
        // in the game
        private constant real PASS_INTERVAL = 2.5
        // Memory leaks occur pretty frequently. When a leak is caught it is saved in
        // an array. But the array can't have more than MAX_LEAK_INSTANCES instances, so
        // if more than MAX_LEAK_INSTANCES memory leaks occur during a destroy interval,
        // the system fails.
        private constant integer MAX_LEAK_INSTANCES = 60000
    endglobals
    
    globals
        private HandleTable IndexData
        private HandleTable IsSaved
        
        //! textmacro MemoryLeakVars takes NAME, TYPE
        private integer Caught$NAME$Leaks = 0
        private $TYPE$ array $NAME$LeakData[MAX_LEAK_INSTANCES]
        private integer $NAME$DestroyCount = 0
        private $TYPE$ array $NAME$DestroyData[MAX_LEAK_INSTANCES]
        //! endtextmacro
        //! runtextmacro MemoryLeakVars("Location","location")
        //! runtextmacro MemoryLeakVars("Effect","effect")
        //! runtextmacro MemoryLeakVars("Group","group")
        
        
        private integer DestroyedLeaks = 0
        private integer CaughtLeaks = 0
        private integer DestroyedLeaksUser = 0
        private handle LastCaught
        private timer PassTimer = CreateTimer()
        private timer CleanTimer = CreateTimer()
        private timer DelayTimer = CreateTimer()
        private boolean IsDestroying = false
        private real SavedMemory = 0.
        private real LastCheckedGroupMemoryUsage = 0.
        private boolean DestroyThreadRunning = false
        private boolean Disabled = false
        
        // These values were found out in a big leak test by gekko.
        private constant real LOCATION_MEMORY_USAGE        = 0.361
        private constant real GROUP_MEMORY_USAGE           = 0.62
        private constant real GROUP_UNIT_MEMORY_USAGE      = 0.040
        private constant real EFFECT_MEMORY_USAGE          = 11.631
        private constant real REMOVED_EFFECT_MEMORY_USAGE  = 0.066
    endglobals
    
    // ======================================
    // ============= Basic Code =============
    // ======================================
    
    function GetLastCaughtHandle takes nothing returns handle
        return LastCaught
    endfunction
    
    function ProtectHandle takes handle h returns nothing
        set IsSaved[h] = 1
    endfunction
    
    function ProtectVariable takes handle h returns nothing
        set IsSaved[h] = 1
    endfunction

    private function EnableMMH takes nothing returns nothing
        set Disabled = false
    endfunction

    function DelayMMH takes nothing returns nothing
        set Disabled = true
        call TimerStart(DelayTimer,0.00,false,function EnableMMH)
    endfunction
    
    function DisplayLeaks takes nothing returns nothing
        call ClearTextMessages()
        call BJDebugMsg("======= MemoryLeakHelper =======")
        call BJDebugMsg("Destroyed Leaks: "+I2S(DestroyedLeaks))
        call BJDebugMsg("Destroyed Leaks by user: "+I2S(DestroyedLeaksUser))
        call BJDebugMsg("Percentage System: "+R2S(I2R(DestroyedLeaks)/I2R(DestroyedLeaks+DestroyedLeaksUser)*100.)+"%")
        call BJDebugMsg("Percentage User: "+R2S(I2R(DestroyedLeaksUser)/I2R(DestroyedLeaks+DestroyedLeaksUser)*100.)+"%")
        call BJDebugMsg("Leaks until next destroy: "+I2S(MIN_LEAK_NUMBER-CaughtLeaks))
        call BJDebugMsg(" === In Destroy Queue === ")
        call BJDebugMsg("   Group Leaks: "+I2S(GroupDestroyCount))
        call BJDebugMsg("   Location Leaks: "+I2S(LocationDestroyCount))
        call BJDebugMsg("   Effect Leaks: "+I2S(EffectDestroyCount))
        call BJDebugMsg(" === Not in Destroy Queue yet === ")
        call BJDebugMsg("   Group Leaks: "+I2S(CaughtGroupLeaks))
        call BJDebugMsg("   Location Leaks: "+I2S(CaughtLocationLeaks))
        call BJDebugMsg("   Effect Leaks: "+I2S(CaughtEffectLeaks))
        call BJDebugMsg("Time until next PassSequence: "+I2S(R2I(TimerGetRemaining(PassTimer)+0.5))+" seconds.")
        call BJDebugMsg(" ")
        if DISPLAY_SAVED_MEMORY then
            call BJDebugMsg("All in all the MemoryLeakHelper could release "+R2S(SavedMemory)+" kb of memory.")
        endif
        call BJDebugMsg("================================")
    endfunction
    
    private function GroupGetMemoryUsageEnum takes nothing returns nothing
        set LastCheckedGroupMemoryUsage = LastCheckedGroupMemoryUsage + GROUP_UNIT_MEMORY_USAGE
    endfunction
    
    function GroupGetMemoryUsage takes group g returns real
        set LastCheckedGroupMemoryUsage = 0.
        call ForGroup(g,function GroupGetMemoryUsageEnum)
        return LastCheckedGroupMemoryUsage + GROUP_MEMORY_USAGE
    endfunction
    
    //! textmacro ResponseOnLeak takes NAME, VALUE
    private function Catch$NAME$ takes $VALUE$ l returns nothing
        set LastCaught = l
        
        if Disabled then
            return
        elseif Caught$NAME$Leaks == MAX_LEAK_INSTANCES then
            debug call BJDebugMsg("MemoryLeakHelper: Failed to store leak because of size limitations")
            return
        endif
        
        if IndexData.exists(l) == false then
            //call BJDebugMsg("Caught $NAME$")
            set Caught$NAME$Leaks = Caught$NAME$Leaks + 1
            set $NAME$LeakData[Caught$NAME$Leaks] = l
            set IndexData[l] = Caught$NAME$Leaks
        endif
    endfunction
    
    private function AddTo$NAME$DestroyQueue takes $VALUE$ l returns nothing
        set $NAME$DestroyCount = $NAME$DestroyCount + 1
        set $NAME$DestroyData[$NAME$DestroyCount] = l
        set IndexData[l] = $NAME$DestroyCount*-1 // Put his to negative, so we know that this is used in the DestroyQueue now.
    endfunction
    
    private function Release$NAME$ takes $VALUE$ l returns nothing
        local integer index
        if IsDestroying == false and IndexData.exists(l) then
            set index = IndexData[l]
            // If this is true, the index wasn't put to a destroy queue yet.
            if index > 0 then
                set $NAME$LeakData[index] = $NAME$LeakData[Caught$NAME$Leaks]
                set Caught$NAME$Leaks = Caught$NAME$Leaks - 1
            else
                set index = index * -1
                set $NAME$DestroyData[index] = $NAME$DestroyData[$NAME$DestroyCount]
                set $NAME$DestroyCount = $NAME$DestroyCount - 1
            endif
            call IndexData.flush(l)
            set DestroyedLeaksUser = DestroyedLeaksUser + 1
        endif
    endfunction
    //! endtextmacro
    //! runtextmacro ResponseOnLeak("Location","location")
    //! runtextmacro ResponseOnLeak("Group","group")
    //! runtextmacro ResponseOnLeak("Effect","effect")
    
    private function DestroyMemoryLeaks takes nothing returns nothing
        set IsDestroying = true
        
        //call BJDebugMsg("DESTROYING Memory Leaks")
        //! textmacro DestroyLeaks takes NAME, DESTROYCALL, MEMORYUSAGE
        set DestroyedLeaks = DestroyedLeaks + $NAME$DestroyCount
        loop
            exitwhen $NAME$DestroyCount == 0
            
            if DISPLAY_SAVED_MEMORY then
                set SavedMemory = SavedMemory + $MEMORYUSAGE$
            endif
            
            call $DESTROYCALL$($NAME$DestroyData[$NAME$DestroyCount])
            call IndexData.flush($NAME$DestroyData[$NAME$DestroyCount])
            set $NAME$DestroyCount = $NAME$DestroyCount - 1
        endloop
        //! endtextmacro
        //! runtextmacro DestroyLeaks ("Group","DestroyGroup","GroupGetMemoryUsage(GroupDestroyData[GroupDestroyCount])")
        //! runtextmacro DestroyLeaks ("Location","RemoveLocation","LOCATION_MEMORY_USAGE")
        //! runtextmacro DestroyLeaks ("Effect","DestroyEffect","EFFECT_MEMORY_USAGE")
        
        set IsDestroying = false
        set DestroyThreadRunning = false
        //call StartPassTimer.execute() // Strange. This causes bugs sometimes and the function isn't called
        // This is slower, but safe.
        call ExecuteFunc("StartPassTimer")
    endfunction
    
    function StartDestroyThread takes nothing returns nothing
        if DestroyThreadRunning == false then
            set DestroyThreadRunning = true
            call TimerStart(CleanTimer,CLEAN_UP_INTERVAL,false,function DestroyMemoryLeaks)
            call PauseTimer(PassTimer)
        endif
    endfunction
    
    hook DoNothing StartDestroyThread
    
    // We want that the user doesn't have to protect too many variables, but all the variables that are filled longer
    // than CLEAN_UP_INTERVAL seconds. But what, when the handle thing is put into the destroy stack and the next destroy is
    // in 5 seconds, because the last one was 15 seconds ago? We can simply avoid something like that by using a 2-step-system
    // that goes sure, the handle is only destroyed when it passed the CLEAN_UP_INTERVAL twice.
    // Having two kinds of variables is simply easier and more efficient than having another variable that refers to
    // how many times the handle passed the timer; If it isn't passed/cleared in the Interval then, we can't loop
    // that easily through the data and we'd have to fix gaps later; That would suck a lot of performacne.
    private function PassMemoryLeaks takes nothing returns nothing
        //call BJDebugMsg("PassMemoryLeaks")
        //! textmacro PassLeaks takes NAME
        set CaughtLeaks = CaughtLeaks + Caught$NAME$Leaks
        //call BJDebugMsg("Caught $NAME$s: "+I2S(Caught$NAME$Leaks))
        loop
            exitwhen Caught$NAME$Leaks < 1
            if IsSaved.exists($NAME$LeakData[Caught$NAME$Leaks]) == false and $NAME$LeakData[Caught$NAME$Leaks] != null then
                call AddTo$NAME$DestroyQueue($NAME$LeakData[Caught$NAME$Leaks])
            endif
            set $NAME$LeakData[Caught$NAME$Leaks] = null
            set Caught$NAME$Leaks = Caught$NAME$Leaks - 1
        endloop
        //! endtextmacro
        //! runtextmacro PassLeaks ("Group")
        //! runtextmacro PassLeaks ("Location")
        //! runtextmacro PassLeaks ("Effect")
        
        if CaughtLeaks > MIN_LEAK_NUMBER then
            set CaughtLeaks = 0
            //call BJDebugMsg("Caught Leaks: "+I2S(MIN_LEAK_NUMBER))
            //call BJDebugMsg("Now start Destroy Timer")
            set DestroyThreadRunning = true
            call TimerStart(CleanTimer,CLEAN_UP_INTERVAL,false,function DestroyMemoryLeaks)
            // We have to pause this timer a bit; Otherwise it would break the CLEAN_UP_INTERVAL rule.
            call PauseTimer(PassTimer)
        endif
        
    endfunction
    
    // =================================
    // ============= Usage =============
    // =================================
    
    private function PP takes location source, real dist, real angle returns nothing
        call CatchLocation(source)
    endfunction
    
    private function CU takes integer count, integer unitId, player p, location l, real face returns nothing
        call CatchLocation(l)
    endfunction
    
    private function IPO takes unit k, string order, location l returns nothing
        call CatchLocation(l)
    endfunction
    
    private function SUP takes unit who, location l returns nothing
        call CatchLocation(l)
    endfunction
    
    private function SUF takes unit who, location l, real dur returns nothing
        call CatchLocation(l)
    endfunction
    
    private function GUR takes real radius, location l, boolexpr filter returns nothing
        call CatchLocation(l)
    endfunction
    
    private function CUF takes integer count, integer unitId, player whichPlayer, location loc, location lookAt returns nothing
        call CatchLocation(loc)
        call CatchLocation(lookAt)
    endfunction
    
    hook PolarProjectionBJ PP
    hook CreateNUnitsAtLoc CU
    hook CreateNUnitsAtLocFacingLocBJ CUF
    hook IssuePointOrderLocBJ IPO
    hook SetUnitPositionLoc SUP
    hook SetUnitFacingToFaceLocTimed SUF
    hook GetUnitsInRangeOfLocMatching GUR

    hook RemoveLocation ReleaseLocation
    
    
    private function FG takes group g, code callback returns nothing
        call CatchGroup(g)
    endfunction
    
    hook ForGroupBJ FG // :D This should catch all GUI usages for groups.
    hook GroupPickRandomUnit CatchGroup
    hook CountUnitsInGroup CatchGroup
    
    hook DestroyGroup ReleaseGroup
    
    private function ASETU takes string bla, widget d, string blu returns nothing
        // We can not catch THIS effect, but the effect that was created before.
        // So we can destroy all SpecialEffects excpet one.
        call CatchEffect(GetLastCreatedEffectBJ())
    endfunction
    
    private function ASE takes location where, string modelName returns nothing
        call CatchLocation(where)
        call CatchEffect(GetLastCreatedEffectBJ())
    endfunction

    hook AddSpecialEffectLocBJ ASE
    hook AddSpecialEffectTargetUnitBJ ASETU
    hook DestroyEffect ReleaseEffect
    hook DestroyEffectBJ ReleaseEffect
    
    // When I want to make the timer run the PassMemoryLeaks things, I have to use an .execute command which requires an extra func.
    function StartPassTimer takes nothing returns nothing
        //call BJDebugMsg("Restarting PassTimer")
        call TimerStart(PassTimer,PASS_INTERVAL,true,function PassMemoryLeaks)
    endfunction
    
    private function Init takes nothing returns nothing
        set IndexData = HandleTable.create()
        set IsSaved = HandleTable.create()
        call StartPassTimer()
    endfunction
endlibrary

Attached Files
File type: w3xMemoryLeakHelper.w3x (51.3 KB)
08-30-2009, 03:37 PM#2
Rising_Dusk
Hrm, so you just need to make sure that CLEAN_UP_INTERVAL is greater than the longest time that you hold a var in your GUI and this will work fine? That's kind of cool, honestly.
08-30-2009, 04:09 PM#3
Mr.Malte
Well, not quietly
The data are destroyed every CLEAN_UP_INTERVAL seconds.
But what, if something is detected, the last clan up was 15 seconds ago and the next one is in 5 seconds, but your variable has to stay 19 seconds?
That would bug,

No, you have to call ProtectHandle on the variables you don't want to become destroyed.
08-30-2009, 04:29 PM#4
Opossum
Although I have to say I don't really like ecouraging GUI users to not use Jass, this is a pretty neat idea.
An alternative would be to hook a complete cleanup to DoNothing. That function is not used for anything really afaik, so you can just overload it with the possibility of cleaning up leaks.
08-30-2009, 04:43 PM#5
Mr.Malte
Hey, thanks,
but I didn't really understand that part:

Quote:
An alternative would be to hook a complete cleanup to DoNothing. That function is not used for anything really afaik, so you can just overload it with the possibility of cleaning up leaks.

How would that work?
08-30-2009, 04:48 PM#6
TKF
Nice system.


Collapse JASS:
....
//  Location:  0.361 kb
//  Group:     0.62 kb + 0.040 kb for each unit in the group.
//  Effect:    13.61 kb
....


That's quite much. Especially effects. I thought the memory leaks did take 0.004 (Maybe only to chat?)

But how much memory does it take for each ingame projectile and units? Do you know that?
08-30-2009, 04:55 PM#7
Rising_Dusk
Quote:
Originally Posted by Opossum
An alternative would be to hook a complete cleanup to DoNothing. That function is not used for anything really afaik, so you can just overload it with the possibility of cleaning up leaks.
Vex's Optimizer kills DoNothing calls anyways under the "Removing useless script" header.
08-30-2009, 04:56 PM#8
Opossum
Quote:
Originally Posted by Mr.Malte
How would that work?
Well, whenever someone uses
Trigger:
Do nothing
(= call DoNothing) you just destroy all the leaks like you do currently in your periodic function DestroyMemoryLeaks. So "Do nothing" finally has a use for GUI users.

Edit:
Quote:
Originally Posted by Rising_Dusk
Vex's Optimizer kills DoNothing calls anyways under the "Removing useless script" header.
Even if they're hooked to other functions?
08-30-2009, 05:15 PM#9
Rising_Dusk
Quote:
Originally Posted by Opossum
Even if they're hooked to other functions?
I don't know the specifics of how hooking works, but if you just have a DoNothing() call somewhere, say in a BJ, then it will clear them happily and normally.
08-30-2009, 05:17 PM#10
uberfoop
When you hook a function it does something like this:

Collapse JASS:
hook ThisFunction SomeFunction
-->
Collapse JASS:
function MakeSomeFunctionHappen takes WhatItNeedsTo returns MostLikelyNothing
//makes some function happen
endfunction

function ProcessThisStuffForThisFunctionHook takes whatever returns whateverreturned
call  MakeSomeFunctionHappen(stuff)
call/return ThisFunction(whatever)
endfunction

Then, ThisFunction calls are replaced by ProcessThisStuffForThisFunctionHook calls. So it probably wouldn't be an issue; DoNothing would just be removed from the end of a function that calls the hooked stuff.

The only issue I can think of is GUI's throwing DoNothing calls in if statements, which could totally be a problem.
08-30-2009, 05:26 PM#11
Mr.Malte
Hm.
But why would users want to destroy the data by hand if it is also done periodically?
08-30-2009, 05:33 PM#12
Opossum
Quote:
Originally Posted by Mr.Malte
Hm.
But why would users want to destroy the data by hand if it is also done periodically?
I thought of it as a replacement for the periodic removing which features some problems like you mentioned above. You wouldn't need protected handles anymore and the GUI user only has to throw in a DoNothing every once in a while, not worrying about custom scripts.
08-30-2009, 05:57 PM#13
Mr.Malte
But that would still kill all data:
It is not possible to detect which data are stored in globals.
So when you out sth. into a variable and call DoNothing in the next seconds, it would still clear up your variable.

But I think the protect thing isn't a big deal anyways.
I mean: Which reason exists for saving data like these in globals in GUI except for deleting them by hand?
No reason! And global shadowing isn't possible any longer anyways.

Really, it would be stupid to have such vars in GUI.
And if there are still users having them, probably as 'constants' defined somewhere in a Map Initialization script. (Like Location 'SpawnPoint')
And that is very easy to find and protect.
08-30-2009, 06:32 PM#14
Vexorian
If you hook DoNothing, the optimizer would not be able to clean up anymore. I can confirm so.

Quote:
Well, not quietly
The data are destroyed every CLEAN_UP_INTERVAL seconds.
But what, if something is detected, the last clan up was 15 seconds ago and the next one is in 5 seconds, but your variable has to stay 19 seconds?
That would bug,

No, you have to call ProtectHandle on the variables you don't want to become destroyed.
These things are making me feel like graveyarding it.

First, although no sane Jasser would use those BJs, some jassers do use them, and then they do clean 'the leaks', forcing people not to use BJs is probably nice though.

Second, I think the whole "Protect" thing needs to be thrown out. I also think that it is possible to ensure that no handle will be destroyed before CLEAN_UP_INTERVAL seconds, this is simple by making it require the handle to go through two of these intervals before cleaning it up.

To make it safer, add a something that will wait until you have N handles before doing any cleanup? Leaks are not an issue until you hit the thousands, and making the auto remover more conservative seems like a good idea. If you think about it, cleaning too early would be much more catastrophic than cleaning too late.

Also, you shouldn't rely on catching ForGroupBJ, GUI's won't necessarily pass only new groups to it, they may pass variables to it, and you'd risk having some double frees. You probably need to catch the Get****UnitsIn**** BJ functions instead.
08-30-2009, 08:11 PM#15
Mr.Malte
Quote:
First, although no sane Jasser would use those BJs, some jassers do use them, and then they do clean 'the leaks', forcing people not to use BJs is probably nice though.

Well, if they clean the stuff up, it's no problem for the system: The system can handle it when you clan things up on your own.

Quote:
Second, I think the whole "Protect" thing needs to be thrown out.

I do not agree.
What if users have things like location that are filled with a point and will never change in their game? Something like a constant (for example sth. that defines where units spawn).

Quote:
I also think that it is possible to ensure that no handle will be destroyed before CLEAN_UP_INTERVAL seconds, this is simple by making it require the handle to go through two of these intervals before cleaning it up.

This is a nice idea, thanks :-)

Quote:
To make it safer, add a something that will wait until you have N handles before doing any cleanup? Leaks are not an issue until you hit the thousands, and making the auto remover more conservative seems like a good idea. If you think about it, cleaning too early would be much more catastrophic than cleaning too late.

Another good idea.
I will implement them tomorrow.

Quote:
Also, you shouldn't rely on catching ForGroupBJ, GUI's won't necessarily pass only new groups to it, they may pass variables to it, and you'd risk having some double frees. You probably need to catch the Get****UnitsIn**** BJ functions instead.

I don't agree.
The system will never make double frees, because it checks whether data are registered or not. That's, where I use Table:

Code:
if IndexData.exists(l) == false then

Also it is very sure that GUI users will use ForGroupBJ on their groups, because Blizzard implements it automatically.

Quote:
These things are making me feel like graveyarding it.

Even if I make it safe?