[kossolax]

Originally Posted by Potato Uno Isn't pawn still single threaded? (On that note, how can SQL queries run on a second thread?)

Extention can run threaded code, then a callback is called into your plugin.

[Potato Uno]

So if you want a function to run on a separate thread, you would have to write that function as a C++ extension?

[Neuro Toxin]

I don't think you would want any threaded code (function) updating or accessing shared plugin memory (via natives) as it wouldn't be thread safe without you checking for such conditions.

Eg. You could update a char[] while a Sourcemod Plugin is calling StrEqual at the same time for a game event that just occurred in the main game thread.
Eg. You could have a loop condition on a global int that you change during the loop iteration.

These scenarios cause undesirable effects that are generally inconsistent in outcome and are really tough to debug. You need locking and all this other BS that I could rant on for ages about.

As kossolax said, for MySQL you want these queries to be run on worker threads so the main thread doesn't wait in a locked state for the SQL query to complete. The SQL query data doesn't need to be thread safe as when it completes, it's results are executed on the main thread via a callback.

[nergal]

Hey bail, small question on the methodmaps.

I used Voided's source engine like methodmaps to handle entity stuff a little easier but when I tried using them with loops and other time-critical parts, they'd lag the hell out of the server CPU compared to simply using entities as ints...

Voided's methodmaps don't do anything except store the entity's reference ID.

why are they lagging the entity loops?

What I did exactly was loop through ALL entities, including validity checks, using a methodmap instance for an entity lags the server but a normal entity index loop with validity checks don't lag at all...

I'm assuming the speed issue is due to the backward compatibility with the older syntax?

[BAILOPAN]

Neuro Toxin is correct in that we want to make sure any sort of asynchronous API we provide does not have implicit shared state. That is pretty tricky if we have arbitrary concurrency in the language. Also note that (1) most SourceMod API is not thread-safe and (2) most Source Engine function calls are not thread-safe.

So the worker-model for things like MySQL works best for now.

[BAILOPAN]

@nergal (Without having read your code or Voided's), that sounds very unlikely. But if it was the case, it was a bug somewhere. Methodmaps are just syntactic sugar and do not materially change the performance of code.

[K.K.Lv]

#include <sourcemod>

#pragma semicolon 1
#pragma newdecls required

public Plugin myinfo = {
    name = "",
    author = "",
    description = "",
    version = ""
};

enum Struct {
    ArrayList:arrTest,
};

Struct g_ArrTest[Struct];


public void OnPluginStart() {
    AddCommandListener(CommandListener_Callback, "test");

    if (g_ArrTest[arrTest] == null) {
        g_ArrTest[arrTest] = new ArrayList(1);
    }

    g_ArrTest[arrTest].Clear();
}

public void OnPluginEnd() {
    if (g_ArrTest[arrTest] != null) {
        delete g_ArrTest[arrTest];
        g_ArrTest[arrTest] = null;
    }
}

public Action CommandListener_Callback(int client, const char [] command, int argc) {
    if (!IsPlayerAlive(client))
        return Plugin_Handled;

    static int iTimes = 0;
    iTimes++;
    PrintToServer("0x%x", g_ArrTest[arrTest]);

    if (g_ArrTest[arrTest] != null) {

        g_ArrTest[arrTest].Push(iTimes);

        for (int x = 0; x < g_ArrTest[arrTest].Length; x++) {

            PrintToServer("iTimes is \"%d\"", g_ArrTest[arrTest].Get(x));
        }

        PrintToServer("======================End of loop======================\n");
    }

    return Plugin_Handled;
}

handle 0x10000f
Error:

L 06/30/2015 - 11:32:54: [SM] Native "ArrayList.Get" reported: Invalid Handle 0 (error: 4)
L 06/30/2015 - 11:32:54: [SM] Displaying call stack trace for plugin "test.smx":
L 06/30/2015 - 11:32:54: [SM]   [0]  Line 51, test.sp::CommandListener_Callback()

Is it a bug ?

[BAILOPAN]

Does it happen without using enum structs? I.e.,

const int Struct = 0;
ArrayList g_ArrTest[1];

[K.K.Lv]

Originally Posted by BAILOPAN Does it happen without using enum structs? I.e., const int Struct = 0; ArrayList g_ArrTest[1];

The same as enum struct

Edited:
And for sure I use the last official version fo SourceMod and MetaMod:Source

[BAILOPAN]

@K.K.Lv - Can you file @ https://bugs.alliedmods.net/ ? Seems legit.