TF2 run_command hook

[quartata]

Hi all. I'm making a TF2 plugin in SourcePython and I'm trying to make a hook using @EntityPreHook for the run_command hook (which should fire every time a player's usercmd is processed if I'm not mistaken). However I keep getting a nasty error:

Code: Select all

Traceback (most recent call last):
  File '../addons/source-python/packages/source-python/entities/hooks.py', line 184, in on_entity_created
    _waiting_entity_hooks.initialize(index)
  File '../addons/source-python/packages/source-python/entities/hooks.py', line 170, in initialize
    if hook.initialize(entity):
  File '../addons/source-python/packages/source-python/entities/hooks.py', line 131, in initialize
    self.hooked_function = getattr(entity, self.function)
  File '../addons/source-python/packages/source-python/entities/entity.py', line 102, in __getattr__
    raise AttributeError('Attribute '{0}' not found'.format(attr))
 
AttributeError: Attribute 'run_command' not found


Here's the really minimal code I was testing with:

Syntax: Select all

from entities.hooks import EntityCondition, EntityPreHook
from players import UserCmd
from memory import make_object

@EntityPreHook(EntityCondition.is_player, "run_command")
def classmenu(stack):
  print(make_object(UserCmd, stack[1]).buttons)

What's going wrong here?

(I'm using the latest version of SourcePython)

[L'In20Cible]

I guess this: https://github.com/Source-Python-Dev-Te ... ini#L39-43

Should be moved into ../orangebox/CBasePlayer.ini as it is most likely the same for all OB games. I will have to update some of my servers before confirming it is and push that but try to replace the content of that file with the following:

Syntax: Select all

# ../orangebox/CBasePlayer.ini

[function]

    [[increment_frag_count]]
        identifier_windows = 55 8B EC 8B 45 08 01 81 F8 0C 00 00
        identifier_linux = _ZN11CBasePlayer18IncrementFragCountEi
        arguments = INT

    [[increment_death_count]]
        identifier_windows = 55 8B EC 8B 45 08 01 81 FC 0C 00 00
        identifier_linux = _ZN11CBasePlayer19IncrementDeathCountEi
        arguments = INT


[virtual_function]

    # _ZN11CBasePlayer11Weapon_DropEP17CBaseCombatWeaponPK6VectorS4_
    [[drop_weapon]]
        offset_linux = 264
        offset_windows = 263
        arguments = POINTER, POINTER, POINTER

    # _ZN11CBasePlayer8PreThinkEv
    [[pre_think]]
        offset_linux = 333
        offset_windows = 332
    
    # _ZN11CBasePlayer9PostThinkEv
    [[post_think]]
        offset_linux = 334
        offset_windows = 333

    # _ZN11CBasePlayer16PlayerRunCommandEP8CUserCmdP11IMoveHelper
    [[run_command]]
        offset_linux = 420
        offset_windows = 419
        arguments = POINTER, POINTER

[satoon101]

No it isn't, at least not as of the time this was added:
https://github.com/Source-Python-Dev-Te ... yer.ini#L4

We do need to add it for all engines/games, though. I am finally getting up and running at my new place, but my computer has been offline for 2 months and has a lot of catching up to do before I can look to add those in.

[L'In20Cible]

You are right, HL2:MP/DOD:S are one offset lesser than CS:S/TF2. I've pushed for those 2 missing games, don't have updated servers for the others.

[quartata]

So wait, was the Python binding for CBasePlayer::PlayerRunCommand missing in the TF2 build but not in the others? That explains why I saw a forum thread about it working in HL2DM... I assumed I was just doing it wrong :P

[Ayuto]

Yep, that was the reason.

[quartata]

Just tried it with the new build, when I added a puppet bot to test it (I was too lazy to actually connect to the server) it segfaults. Give me a moment to attach a debug.log

[quartata]

Sorry for the delay:

[quartata]

Probably should have mentioned this earlier, but I'm on Linux (Ubuntu 14.04). I do have two other plugins on my testing server (TFTrue and TF2Items) but looking at the stack trace they're definitely not the culprits.

[Ayuto]

Looks like Sourcemod is also hooking this method and now they are conflicting when calling make_object().

[quartata]

Interesting; I'll take out Sourcemod for the time and see if it works

[quartata]

Actually, I'll just whip up a clean new server to test it with. Can't do that right now though.

[quartata]

No luck even on a clean server without Sourcemod or any of that:

[quartata]

I should probably open this as an issue on Github instead of here

[L'In20Cible]

What if, instead of using EntityCondition.is_player, you try with EntityCondition.is_human_player and you actually goes on the server yourself instead of adding bots?

[quartata]

Stranger and stranger. It works perfectly fine like that, but adding a bot still crashes it even though it's using is_human_player now (which means the hook shouldn't even get called for a bot). It's only when this plugin is loaded so it isn't a general bug in SP or anything. NextBots don't do this, just puppet bots. Not a big deal, just somewhat strange.

[L'In20Cible]

even though it's using is_human_player now (which means the hook shouldn't even get called).

That's not right. The hook will gets called if both classes share the same address in their dispatch table. The EntityCondition only makes sure that the address is retrieved from an human player before registering the hook but if both classes are sharing the same pointer, it will get fired no matter what the actual "this" pointer is (done this way so you can fully control which method to hook in case of multiple inheritances but filtering the calls, is your responsability).

Now it is time to debug, what gets printed to the console before it crashes?

Syntax: Select all

from entities.hooks import EntityCondition, EntityPreHook
from players import UserCmd
from memory import make_object

@EntityPreHook(EntityCondition.is_human_player, 'run_command')
def classmenu(stack):
    print('Callback executed...')
    print('Arguments =', list(stack))
    usercmd = make_object(UserCmd, stack[1])
    print('UserCmd =', usercmd)
    print('Buttons =', usercmd.buttons)

Additional question, does it crashes even if you didn't join the server first? In such case, the hook should not be registered so if it does, we may have to add a specific check in either the hook API or the conversion functions which may get a NULL pointer somewhere for those specific bots.

Also, I originally proposed EntityCondition.is_human_player cause I thought the bot class may override that method and pass a CBotCmd instance instead of a CUserCmd.

[quartata]

It does not crash if I don't join the server first. When I do join, it prints the Callback executed and the Arguments before crashing. So the bots using a different class for their usercmd seems plausible; it's obviously crashing when dereferencing the pointer to the usercmd.

[L'In20Cible]

Yeah, I think so. What if you use the following:

Syntax: Select all

from entities.hooks import EntityCondition, EntityPreHook
from players.bots import BotCmd
from players import UserCmd
from players.entity import Player
from memory import make_object

@EntityPreHook(EntityCondition.is_player, "run_command")
def classmenu(stack):
    player = make_object(Player, stack[0])
    if player.is_fake_client():
        cls = BotCmd
    else:
        cls = UserCmd
    usercmd = make_object(cls, stack[1])
    print(usercmd.buttons)

[quartata]

Did you mean

Syntax: Select all

from players.bots import BotCmd

?