Factorio Forums

Zeblote wrote:

ratchetfreak wrote: Also on the more technical side; the replay file must contain the file that was pulled in. That can cause the replay file to grow quite a bit if a mod is dumb and pulls a file every tick.

I see no reason for a mod to do that. If a file is needed regularly, it can be cached in lua.

As I said "if a mod is dumb" and there will be dumb mods.

Doing that in multiplayer for each player will be even more hazardous because the bandwidth cost will explode.

So a rate limiter on it would be nice.

ratchetfreak wrote: As I said "if a mod is dumb" and there will be dumb mods.

Doing that in multiplayer for each player will be even more hazardous because the bandwidth cost will explode.

So a rate limiter on it would be nice.

Well, just don't install dumb mods. There shouldn't be a limit just because someone might be unable to use it correctly.

I assume text box synchronization happens between ticks, so doing something like: would wait for the next tick, which would wait for the script to finish, which would wait for the next tick. Infinite loop!

It would be need to be at least slightly asynchronous, like: And since it is asynchronous anyway, it is no problem to cap the file transfer speed to prevent it from eating all the bandwidth.

Hmm. That is a good point. Loading a larger file might also take many ticks to sync.

Now the real question is, how long does it take to implement these and can we get them in 12.11 or 12.12?

Seems like you've found a way to prevent a script from changing the game state, so I don't see what's stopping client sided scripts.

Not running everything on all clients makes way for many client sided functions in the future, especially regarding guis and files.

Zeblote wrote:Seems like you've found a way to prevent a script from changing the game state, so I don't see what's stopping client sided scripts.

Not running everything on all clients makes way for many client sided functions in the future, especially regarding guis and files.

By not allowing access to the game state you can't change the game state. Allowing reading files in while you had no access to the game state or the global table would also prevent modifying the game state but it would be useless as you couldn't *do* anything with the file.

GUIs are part of the game state. If the mod is able to read from the GUI then it is changes the game state.

Rseding91 wrote: By not allowing access to the game state you can't change the game state. Allowing reading files in while you had no access to the game state or the global table would also prevent modifying the game state but it would be useless as you couldn't *do* anything with the file.

That's why you send a custom event to the game that can then be processed deterministically. What you're doing with a gui and files and whatever does not matter to other clients until you actually want to do something with it that changes the game state. Client sided scripts are required for many advanced APIs like files, maybe internet access, etc...

Rseding91 wrote:GUIs are part of the game state. If the mod is able to read from the GUI then it is changes the game state.

They shouldn't be, it makes no sense. If I open the escape menu and then the Save game gui, do other client's know what I'm doing with them? Does it send the preview images to other clients when I select a save game in the list? Do other client's know whether I'm typing a chat message until I've actually sent it? If not, why do you want to force modders to use this system?

Has any more thought been put into this?

Looked around a bit at how the default guis work, and your internal system is a lot more advanced than the few parts you expose to modding. I can only assume this is because it is (obviously) client-sided and doesn't have to deal with synchronizing everything.

The high level gui system is fine for many simple things, but it quickly reaches its limit. I'd like to make more advanced mods with seperate game/sim sided and client sided (ui, keybinds, files, network, etc) parts but can't because there's no such system available.

Zeblote wrote:They shouldn't be, it makes no sense. If I open the escape menu and then the Save game gui, do other client's know what I'm doing with them? Does it send the preview images to other clients when I select a save game in the list? Do other client's know whether I'm typing a chat message until I've actually sent it? If not, why do you want to force modders to use this system?

Mods have read/write access to the GUI elements they and other mods create. Because they have read access the GUIs mods create have to be persisted between saves.

Without persisting the GUI elements mods create you easily end up with a desync: Create GUI element through mod, MP peer connects, GUI element wasn't persisted between save/load and now peer 1 has the GUI element and peer 2 does not. The mod is identical on both ends and does different stuff based on the GUI existing on one peer and not on the other.

I think you're misunderstanding what I'm asking for. It's not a change to the existing system, but rather a new system. Lower level than the existing one.

Currently, we have one script: control.lua. This runs as part of the simulation and is fully deterministic. However, certain actions simply are not deterministic or don't have to be, because they don't affect anyone else.


I'm asking for a second script: client.lua. This runs on a client and is local to that client only.

It's executed/called at these times:
- Started single player game, before (!) initializing the control.lua scripts
- Loaded existing save
- Started multiplayer game, before (!) initializing the control.lua scripts
- Loaded existing save as multiplayer game
- Joined multiplayer game

This script does not need to be synchronized with other clients, or store any persistent data in the save files.
It would introduce a new namespace "client", similar to the existing "script" and "game".

Because it's local to each client, it can't change the game state on it's own. (It does have read access, though. As in, you would be able to run find_entities but not create_entity.)
That means there needs to be a way to communicate between these two scripts. It could be as simple as client.transmit_event(name, args). The sent events will be processed deterministically by the control.lua script, which applies the game state change we need.

Now that we have the client script framework, what's the point? For example, you can now implement full file IO under the client namespace. You don't have to worry about synchronization, because the file is handled on the client side and sent to the server using transmit_event when it's actually needed. You can implement all of the internal gui elements under the client namespace, because you don't have to worry about synchronization. Once a client does something in the gui that should trigger a game state change, it is simply sent to control.lua with transmit_event and replicated on all clients. I'm sure you can think of more.

By offering lower level control over handling multiplayer you let modders make much more advanced features that otherwise have to be part of the engine.
For example, the blueprint gallery. I'm imagining a fancy gui that has all your blueprints sorted in folders and even with preview images. From there, you can add network functions to the client namespace so we can make an online blueprint sharing site. There's really no limit!