CulturedCait

That was good suggestion. I placed as much "warns" as possible and discovered that in my OnInitGame trigger: Trigger.prototype.TestLearningStart = function() { warn("Init"); //zapytajmy czy jest plik startowy inicjalizujacy nasz algorytm wstepnymi entities itp let cmpGuiInterface = Engine.QueryInterface(SYSTEM_ENTITY, IID_GuiInterface); cmpGuiInterface.PushNotification({ "type": "load-json", "players": [0], "name": "testlearning-last", "param": "init" }); warn("Init finish"); //this.TestLearningStartFromScratch(); } PushNotification sends the request without errors, but OnCommand never receives it. It is was clear that my custom message "load-json", which loads JSON file, is simply not being push further. This is my load-json msg: "load-json": function(notification, player) { warn("Received load-json"); let name = notification.name || "testlearning"; let data = null; if (Engine.FileExists("simulation/" + name + ".json")) data = Engine.ReadJSONFile("simulation/" + name + ".json"); Engine.PostNetworkCommand({ "type": "dialog-answer", "dialog": "json", "answer": { "param": notification.param, "data": data, }, }); warn("finished load-json"); }, And this is how I run the game with loading mod with customized "messages.js": $ pyrogenesis -mod="public" -mod="train" -autostart-victory="endless" -autostart="skirmishes/test_sparta" -autostart-disable-replay -quickstart -autostart-nonvisual Mod "train" contains just one single file with modified "messages.js". Since warning "Received load-json" never appears, we can be sure this message never hits in headless mode. So the only explanation is that under -autostart-nonvisual, custom "messages.js" coming from my mod is either ignored or, which is more likely, overlapped with different set of "messages.js" designated for -autostart-nonvisual. Since I need custom messages to communicate with some GUI APIs, I can't get rid of this pattern... I think it is not worth investigating further. I'm satisfied with my current solution as it is and although headless mode certainly could be nice improvement it is not that important for these type of calculations. Thank you again for your help.

Nope. I'm not sure if I can attach full logs here in public. They include sensitive data like API keys. I can send you both logs in PM if needed. In -autostart-nonvisual "NonVisualTrigger.js" is loaded additionally as opposed to regular run. I don't think its the culprit, but who knows.

With -autostart-nonvisual: Without:

Well, as I mentioned earlier my output is not appearing at all. I have plenty of warns. Script also reads/writes JSON files and does save game. None of this is executed if -autostart-nonvisual is on. Just "Turn (...)..." appears instead. Looks like it is impossible to use headless mode just to run map with scripts. Probably -autostart-nonvisual is not just turning off graphics and sound, but rather special mode designated to work with -rl-interface only.

That explains a lot. Hmm.. Weird idea, but is it possible that in nonvisual mode "OnInitGame" is not invoked?? This is my entry point: { let learnTrigger = Engine.QueryInterface(SYSTEM_ENTITY, IID_Trigger); learnTrigger.RegisterTrigger("OnDeserialized", "TestLearningLoad", { "enabled": true }); learnTrigger.RegisterTrigger("OnInitGame", "TestLearningStart", { "enabled": true }); learnTrigger.RegisterTrigger("OnPlayerCommand", "TestLearningCommand", { "enabled": true }); ////learnTrigger.DoAfterDelay(1000, "TestLearningSpeed", { "enabled": true }); } In single run my script ends in about 20 minutes,then Engine.Exit is issued to close the game. If it works much faster in -nonvisual mode, then we should expect it to finish in less than 20 minutes. What I did was to test it again with -autostart-nonvisual and just waited. Although it didn't produce save games and JSON files, we could explain this is somehow forbidden in this mode. But considering it has still not finished after 30 minutes, maybe the whole thing is just that the script is not triggering in -nonvisual mode??

Nope, this is not the issue. Script works correctly. It was just redundancy but didn't affect anything. I removed "TestLearningCommand" so you could see it had nothing to do with this matter. Problem is that --autostart-nonvisual does *something*, but certainly not what user selected with "-autostart". Here are some proofs: --autostart-nonvisual causes high CPU usage and uses all cores. While regular game uses only one thread (~6-7%). My script should generate "debug.json" after start, and about each minute performs the save game. With --autostart-nonvisual nothing happens. Even if we assumed --autostart-nonvisual suppressed all warnings, but still performed map and scripts, I should expect to JSON files and savegame to be created. --autostart-nonvisual uses only ~250Mb of memory. My script after start uses ~1G and after 15min consumes around 5G. It's combat between ~300units performed each 5sec. Screen below pertains --autostart-nonvisual: Certainly pyrogenesis performs some heavy computations on all cores. And selected map is not being run. While this screen below is without -autostart-nonvisual. One core used, JSON files produced, output produced, and all works as intended: I checked on windows and it's same story. The only difference is that if you use --autostart-nonvisual, "Turn (....)" output is not being produced at all, program exits immediately spawning another process. Rest is the same. Spawned process *does something* on all cores, but not running the map that it was asked.

I see. I'll try to run same command on windows build to see if it works there. That should hopefully give me the answer if the problem is my version of the game or something else.

While quick searching the repository I encountered this: https://gitea.wildfiregames.com/0ad/0ad/src/commit/b77ee3c5dcde432787b0bdf0386fbf766e19f25f/source/ps/Replay.cpp#L279 It's part of the "Replay" method and looks precisely as part of the output I got. Is it possible that --autostart-nonvisual is designed only for replays and you simply can't start the match from scratch when this parameter is present?

Yes, I checked other maps. Same result. Always executes "Turn 0 (200).." script etc ignoring the map. Doesn't matter if mods are used or not. If I remove --autostart-nonvisual everything is OK and maps are loaded normally. Maybe my version of the game is the issue? ~/.local/share/0ad/mods/user/simulation$ pyrogenesis -autostart="skirmishes/tarim_basin_2p" -quickstart --autostart-nonvisual TIMER| InitVfs: 160.709 us FILES| Main log written to '/home/przemek/.config/0ad/logs/mainlog.html' FILES| Interesting log written to '/home/przemek/.config/0ad/logs/interestinglog.html' TIMER| CONFIG_Init: 2.53454 ms FILES| Replay written to '/home/przemek/.local/share/0ad/replays/0.27.1/2026-01-06_0007' GAME STARTED, ALL INIT COMPLETE Turn 0 (200)... Turn 1 (200)... Turn 2 (200)... Turn 3 (200)... Turn 4 (200)... Turn 5 (200)... Turn 6 (200)... Turn 7 (200)... Turn 8 (200)... Turn 9 (200)... Turn 10 (200)... ~/.local/share/0ad/mods/user/simulation$ pyrogenesis -autostart="scenarios/polynesia" -quickstart --autostart-nonvisual TIMER| InitVfs: 194.705 us FILES| Main log written to '/home/przemek/.config/0ad/logs/mainlog.html' FILES| Interesting log written to '/home/przemek/.config/0ad/logs/interestinglog.html' TIMER| CONFIG_Init: 3.4274 ms FILES| Replay written to '/home/przemek/.local/share/0ad/replays/0.27.1/2026-01-06_0008' GAME STARTED, ALL INIT COMPLETE Turn 0 (200)... Turn 1 (200)... Turn 2 (200)... Turn 3 (200)... Turn 4 (200)...

OK, I've completed environment setup. I managed to expose SetEngineRate, WriteJSONFile and incorporate other useful APIs to JS triggers according to your suggestions. To make it work creating a very tiny mod was necessary. However -autostart-nonvisual seems not to work as intended. According to readme: -autostart-nonvisual disable any graphics and sounds Apparently it ignores all other settings and runs the game with hardcoded options like if it was tailored to --rl-interface only. This is the command I run the game. It includes my custom map with all algorithms, and mod name with exposed functions. pyrogenesis -mod="public" -mod="train" -autostart-victory="endless" -autostart="skirmishes/test_sparta" -autostart-disable-replay -quickstart Typically map starts automatically, then training is performed, and after one generation script closes the game due to high memory usage. Then bash script runs it again indefinitely to iterate etc. Expected output in my case after pyrogenesis start is something like: (....) TIMER| session/setup.xml: 82.173 us TIMER| session/sprites.xml: 381.305 us TIMER| session/styles.xml: 72.086 us TIMER| session/session.xml: 31.4006 ms GAME STARTED, ALL INIT COMPLETE WARNING: Beginning from JSON file WARNING: Trigger.js: Trigger "TestLearningCommand" has been registered before. Aborting... WARNING: Load WARNING: currentIteration: 9 WARNING: entitiesToDo: 0 WARNING: entitiesDone: 250 WARNING: crossover chance: 50% WARNING: mutation chance: 40% WARNING: 512 (....) However if I just add "-autostart-nonvisual", then map isn't being loaded, and some other scripts or computations are being performed instead (look after "GAME STARTED, ALL INIT COMPLETE): bash:~/.local/share/0ad/mods/user/simulation$ pyrogenesis -mod="public" -mod="train" -autostart-victory="endless" -autostart="skirmishes/test_sparta" -autostart-disable-replay -quickstart -autostart-nonvisual TIMER| InitVfs: 197.269 us FILES| Main log written to '/home/przemek/.config/0ad/logs/mainlog.html' FILES| Interesting log written to '/home/przemek/.config/0ad/logs/interestinglog.html' TIMER| CONFIG_Init: 1.27741 ms WARNING: Init GAME STARTED, ALL INIT COMPLETE Turn 0 (200)... Turn 1 (200)... Turn 2 (200)... Turn 3 (200)... Turn 4 (200)... Turn 5 (200)... Turn 6 (200)... (...) This may be a bug if (=>readme) is accurate. Is it possible to disable screen and sound in a different way? Thank you!

OK! Thank you guys. I'll investigate and let you know if I encounter any obstacles. Cheers.

I checked the python repo and it picked my interest. Particularly headless option you mentioned would be very useful. I have question regarding this cmd: pyrogenesis --rl-interface=0.0.0.0:6000 --autostart-nonvisual --mod=rl-scenarios --mod=public Is there any api documentation regarding how to communicate with this interface? I.e. what should you send to this port to execute some part of the game (assuming I understand this interface correctly, sorry I didn't have much time to examine python scripts in detail..). In my case what I needed would be to actually load given map remotely, and save this map after period of time.

Nope. Very interesting. This would be precisely what I'm looking for. Thank you.

Well, I wanted to avoid creating a mod. Actually I just wanted to make scripted map in plain "vanilla" 0 A.D. SetSimRate, as helpful as it would be for automating few things, is not that critical in the end. I can always invoke the command from the console. Thank you. Ok. This is very long story, but to the details what I'm working on... I'll try to explain. I've always wanted to know which types of units and formations are best in a given case. Also, how my formation should look like considering shape and direction vector of the enemy during tactical clash (in second stage of the project my idea would be also to find precise polynomials that should represent formation lines that should be used). On the one hand these are pointless questions (what we're dealing with are differential equations... that's why some say that strategy is rather an art than something you can calculate). But for a long time it's been bugging me and I had some ideas how to crack this problem with genetic algorithms. Or at least to try finding approximate or partial answers in some cases. Since I found some time during the Christmas I developed very tiny genetic algorithm framework in JS that operates only on 0 AD triggers in asynchronous way. Since the objective function in my problem is the result of single combat, and GA requires to perform many of such clashes for single generation, this is "the lengthy calculation" I was talking about. GA has to wait for each of such "entities" (i.e. clash of two armies) results to pick parameter candidates for next generation. That's why I was interested in SetSimRate to speed up the process a little. As you can see there are many asynchronous calculations (actually thousands...) that need to be processed sequentially. Speeding up simulation helps much. I know what you say. You could always grab units statistics, calculations etc from the code and write a program in faster environment to achieve similar results. But I wanted to utilize exact 0 A.D. engine as it is. There would be many mistakes in this process if I did otherwise. Maybe it would be easier, and you wouldn't have to resort to some quirks, but since scripting support in 0 A.D. is very comprehensive it is possible to achieve all I described above. GA's are very slow by definition so technology doesn't matter that much. The algorithm itself is the bottleneck. This is of course very lengthy project. Absolutely nothing to brag about, more like a hobby actually Hope this now sheds little light where my question came from. Now, to simulate how SetSimRate helps, just try to invoke something simple that spawns itself each 100ms in a sequence using DoAfterDelay. You can assume this "100ms" is aforementioned sequenced lengthy atomic operation that you don't know the result yet. You'll be notified of the result later when it finishes, and you'll utilize the result later to spawn next iteration "somehow". testTrigger.DoAfterDelay(100, .....) You'll see the difference if you speed up simulation to x20. This "100" would become ~5 in real life time. Maybe not really 5, but much faster.

Hi phosit, I checked PushNotification and it seems to be limited to several possible message types. Which of course makes sense from the usage perspective... Does it mean the only way to achieve what I wanted would be to mod messages.js by adding new message type that registers function executing SetSimRate? Did I get your idea right? Thanks.