first attempt at clearing npc events(callbacks) from scriptinterface …

[yamaken93]

…lua events table

Pull Request Prelude

• I have followed proper The Forgotten Server code styling.
• I have read and understood the contribution guidelines before making this PR.
• I am aware that this PR may be closed if the above-mentioned criteria are not fulfilled.

Changes Proposed

Remove/clear the events in the events table so resources can be freed on the lua side.

Issues addressed:
#3552

[yamaken93]

~MonsterInfo() {
			if (!scriptInterface) {
				return;
			}
			std::array<int32_t, 5> eventArray = {creatureSayEvent, creatureDisappearEvent, creatureAppearEvent, creatureMoveEvent, thinkEvent};
			for (int32_t eventId : eventArray) {
				scriptInterface->deleteEvent(eventId);
			}
		};

I still need to find a way to add this kinda of code to MonsterInfo so we can prevent the leak for monsters events too.

[nekiro]

static?

[ranisalt]

@nekiro a static in the destructor means it will always try to delete the events of the first NpcEventsHandler to ever be deleted.

No need for the temporary though

for (int32_t eventId : {creatureSayEvent, creatureDisappearEvent, creatureAppearEvent, creatureMoveEvent, playerCloseChannelEvent, playerEndTradeEvent, thinkEvent}) {

[Erza]

I used an empty NPC script file with these contents

function onCreatureAppear(cid)
end
function onCreatureDisappear(cid)
end

function onCreatureSay(cid, type, msg)
end

function onThink()
end

And data/npc/lib/npc.lua only had the following content, in order to allocate some memory and then see if it will get free'd

local bigTable = {}
local index = 1
for i = 1, 10000 do
    local tbl = {}
    local tmpIndex = 1
    
    for i = 1, 100 do
        tbl[tmpIndex] = ("%s%u%s%u"):format(("a"):rep(100), math.random(1000000000000), ("b"):rep(100), math.random(1000000000000))
        tmpIndex = tmpIndex + 1
    end

    bigTable[index] = tbl
    index = index + 1
end

I then ran the test from #3552 again, with 5000 iterations each, totaling 15k NPC's. This is the memory graph

The second tiny dip in the graph is due to me running /reload npc after the test, so lua_gc will get called. As we can see, the bigTable object's don't get cleaned up. There must be something more to this issue.

[Erza]

New findings: it seems like Lua is reserving a certain amount of memory that it then re-uses, only allocating more when there is no more free room.

With this PR

Callbacks are unreferenced, memory usage stops increasing, GC times stay consistent - 690MB~ of memory used by the process after running the test 6x with 15k NPC's each

Without this PR

Callbacks stay referenced forever, memory usage keeps increasing, GC times increase as well 1.5GB~ of memory used by the process after running the test 4x with 15k NPC's each

Now we just need to unreference all the other places where LuaScriptInterface::getEvent or LuaScriptInterface::getMetaEvent are called

[Erza]

A simple std::cout should be sufficient here since we aren't formatting anything, then we don't need to include the printf header from the fmt lib.

Correction of spelling in the message: It looks like we have a dangling scriptInterface in a combat callback.

[EPuncker]

Suggested change

	fmt::printf("CallBack::~CallBack(): its look like we have a dangling scriptInterface in a combat callback.");
	std::cout << "CallBack::~CallBack(): It looks like we have a dangling scriptInterface in a combat callback." << std::endl;

[EPuncker]

Suggested change

#include "fmt/printf.h"

[EPuncker]

'e': unreferenced local variable