Skip to main content

self_hosted_node/engine_backend/
java_backend.rs

1#![allow(dead_code)]
2
3#[cfg(forge_backend)]
4use std::collections::HashMap;
5use std::env;
6#[cfg(feature = "java-forge")]
7use std::io::{BufRead, BufReader, BufWriter, Write};
8use std::path::{Path, PathBuf};
9#[cfg(feature = "java-forge")]
10use std::process::{Child, ChildStdin, Command, Stdio};
11use std::sync::atomic::AtomicBool;
12use std::sync::mpsc as std_mpsc;
13#[cfg(feature = "java-forge")]
14use std::sync::mpsc::RecvTimeoutError;
15#[cfg(forge_backend)]
16use std::sync::mpsc::TryRecvError;
17use std::sync::Arc;
18#[cfg(feature = "java-forge")]
19use std::sync::Mutex;
20#[cfg(forge_backend)]
21use std::time::Duration;
22#[cfg(forge_backend)]
23use std::time::Instant;
24
25use manabrew_protocol::deck_dto::{Deck, DeckCardIdentity};
26
27use crate::config::DeckSelection;
28#[cfg(feature = "java-forge")]
29use manabot::{BotAgent, SimpleAi};
30#[cfg(forge_backend)]
31use manabrew_agent_interface::game_view_dto::GameViewDto;
32use manabrew_agent_interface::prompt::{AgentMessage, ClientToServerMessage};
33#[cfg(forge_backend)]
34use manabrew_agent_interface::prompt::{
35    AgentPrompt, ChooseActionOutput, DiceRolledOutput, DirectiveInput, GameOverInput, PromptInput,
36    PromptOutput, ProtocolError, ProtocolErrorCode, ResponseViolation, StateUpdate,
37};
38#[cfg(feature = "java-forge")]
39use manabrew_agent_interface::prompt::{MulliganOutput, MulliganPutBackOutput};
40use serde::Serialize;
41#[cfg(feature = "java-forge")]
42use serde_json::json;
43#[cfg(feature = "java-forge")]
44use serde_json::Value;
45#[cfg(forge_backend)]
46use tracing::warn;
47#[cfg(forge_backend)]
48use tracing::{debug, info};
49
50use super::HostedGameOver;
51use crate::config::workspace_root;
52
53pub fn unsupported_message() -> &'static str {
54    "hosted java-forge backend is unavailable; rebuild self-hosted-node with --features java-forge"
55}
56
57#[cfg(feature = "java-forge")]
58pub fn run_smoke_game(max_prompts: usize) -> Result<(), String> {
59    let config = JavaRuntimeConfig::from_env();
60    let assets_dir = config.assets_dir.to_string_lossy().to_string();
61    let bridge = SubprocessBridge::spawn(&config)?;
62    let mut session = JavaForgeSession::new(bridge);
63    session.initialize(&assets_dir)?;
64
65    let deck_a = smoke_deck("Mountain", "Lightning Bolt");
66    let deck_b = smoke_deck("Forest", "Grizzly Bears");
67    let request = StartGameRequest::new(
68        "self-hosted-java-smoke".to_string(),
69        String::new(),
70        20,
71        42,
72        vec![
73            PlayerConfig::new("Smoke A".to_string(), &deck_a, Vec::new()),
74            PlayerConfig::new("Smoke B".to_string(), &deck_b, Vec::new()),
75        ],
76    );
77    let session_id = session.start_game(&request)?;
78    info!(session_id, "java-forge smoke session started");
79
80    let mut prompts_seen = 0usize;
81    while prompts_seen < max_prompts {
82        let Some(prompt_json) = wait_for_prompt(&mut session, 600)? else {
83            session.end_game()?;
84            return Err("timed out waiting for java-forge smoke prompt".to_string());
85        };
86        let prompt: AgentPrompt = serde_json::from_str(&prompt_json)
87            .map_err(|err| format!("failed to parse java-forge smoke prompt: {err}"))?;
88        let player = player_index(&prompt.deciding_player_id);
89        info!(prompts_seen, player, "java-forge smoke prompt");
90        let pass = PromptOutput::ChooseAction(ChooseActionOutput::Pass {
91            until: None,
92            exhaust_stack: false,
93        });
94        session.submit_action(&serde_json::to_string(&pass).map_err(|err| err.to_string())?)?;
95        prompts_seen += 1;
96    }
97
98    let snapshot_json = session.get_snapshot(Some(0))?;
99    let snapshot: Value = serde_json::from_str(&snapshot_json)
100        .map_err(|err| format!("failed to parse java-forge smoke snapshot: {err}"))?;
101    info!(
102        turn = snapshot
103            .get("turn")
104            .and_then(|value| value.as_i64())
105            .unwrap_or_default(),
106        phase = snapshot
107            .get("phase")
108            .and_then(|value| value.as_str())
109            .unwrap_or("<missing>"),
110        "java-forge smoke snapshot"
111    );
112    session.end_game()?;
113    Ok(())
114}
115
116#[cfg(not(feature = "java-forge"))]
117pub fn run_smoke_game(_max_prompts: usize) -> Result<(), String> {
118    Err(
119        "java-forge smoke requires building self-hosted-node with --features java-forge"
120            .to_string(),
121    )
122}
123
124// Unlike run_smoke_game, which spawns the JAR in a subprocess, this drives the
125// GraalVM native library in-process: isolate creation, forge_initialize (which
126// loads the card database) and a real game start. CI runs it after restoring
127// forge-harness/native/build from cache, where nothing else would notice a
128// stale or broken libforgeharness.
129#[cfg(feature = "graal-forge")]
130pub fn run_graal_smoke() -> Result<(), String> {
131    let config = JavaRuntimeConfig::from_env();
132    let engine = GraalEngineHandle::create(&config.assets_dir)?;
133
134    let deck_a = smoke_deck("Mountain", "Lightning Bolt");
135    let deck_b = smoke_deck("Forest", "Grizzly Bears");
136    let request = StartGameRequest::new(
137        "self-hosted-graal-smoke".to_string(),
138        String::new(),
139        20,
140        42,
141        vec![
142            PlayerConfig::new("Smoke A".to_string(), &deck_a, Vec::new()),
143            PlayerConfig::new("Smoke B".to_string(), &deck_b, Vec::new()),
144        ],
145    );
146
147    let session_id = engine.start_game(&request.to_json().map_err(|err| err.to_string())?)?;
148    info!(session_id, "graal-forge smoke session started");
149    engine.end_game(&session_id)?;
150    Ok(())
151}
152
153#[cfg(not(feature = "graal-forge"))]
154pub fn run_graal_smoke() -> Result<(), String> {
155    Err(
156        "graal-forge smoke requires building self-hosted-node with --features graal-forge"
157            .to_string(),
158    )
159}
160
161#[cfg(feature = "java-forge")]
162pub fn run_scenario(name: &str, max_prompts: usize) -> Result<(), String> {
163    let scenario = JavaScenario::from_name(name)?;
164    let config = JavaRuntimeConfig::from_env();
165    let assets_dir = config.assets_dir.to_string_lossy().to_string();
166    let bridge = SubprocessBridge::spawn(&config)?;
167    let mut session = JavaForgeSession::new(bridge);
168    session.initialize(&assets_dir)?;
169
170    let request = StartGameRequest::new(
171        format!("self-hosted-java-scenario-{}", scenario.name()),
172        String::new(),
173        20,
174        42,
175        vec![
176            PlayerConfig::new(
177                "Scenario A".to_string(),
178                &scenario_deck("Swamp"),
179                Vec::new(),
180            ),
181            PlayerConfig::new(
182                "Scenario B".to_string(),
183                &scenario_deck("Forest"),
184                Vec::new(),
185            ),
186        ],
187    );
188    let session_id = session.start_game(&request)?;
189    info!(
190        session_id,
191        scenario = scenario.name(),
192        "java-forge scenario started"
193    );
194
195    let result = run_scenario_loop(&mut session, scenario, max_prompts);
196    let end_result = session.end_game();
197    result.and(end_result)
198}
199
200#[cfg(not(feature = "java-forge"))]
201pub fn run_scenario(_name: &str, _max_prompts: usize) -> Result<(), String> {
202    Err(
203        "java-forge scenarios require building self-hosted-node with --features java-forge"
204            .to_string(),
205    )
206}
207
208#[cfg(feature = "java-forge")]
209pub fn run_self_play(
210    seats: &[DeckSelection],
211    starting_life: i32,
212    seed: u64,
213    max_prompts: usize,
214    games: usize,
215) -> Result<(), String> {
216    let config = JavaRuntimeConfig::from_env();
217    let assets_dir = config.assets_dir.to_string_lossy().to_string();
218    let bridge = SubprocessBridge::spawn(&config)?;
219    let mut session = JavaForgeSession::new(bridge);
220    session.initialize(&assets_dir)?;
221
222    let mut players = Vec::with_capacity(seats.len());
223    for (i, seat) in seats.iter().enumerate() {
224        let identities = deck_card_identities(&seat.deck);
225        players.push(PlayerConfig::new(
226            format!("Self-Play {}", i + 1),
227            &identities,
228            commander_names_for_java(&seat.deck, seat.commander_name.as_deref()),
229        ));
230    }
231
232    for game_index in 0..games.max(1) {
233        let request = StartGameRequest::new(
234            format!("self-hosted-java-self-play-{game_index}"),
235            String::new(),
236            starting_life,
237            seed.wrapping_add(game_index as u64),
238            players.clone(),
239        );
240        let session_id = session.start_game(&request)?;
241        info!(
242            session_id,
243            game_index,
244            games,
245            players = seats.len(),
246            starting_life,
247            max_prompts,
248            "java-forge self-play game started"
249        );
250        let result = run_self_play_loop(&mut session, max_prompts);
251        let end_result = session.end_game();
252        result.and(end_result)?;
253    }
254    Ok(())
255}
256
257#[cfg(not(feature = "java-forge"))]
258pub fn run_self_play(
259    _seats: &[DeckSelection],
260    _starting_life: i32,
261    _seed: u64,
262    _max_prompts: usize,
263    _games: usize,
264) -> Result<(), String> {
265    Err(
266        "java-forge self-play requires building self-hosted-node with --features java-forge"
267            .to_string(),
268    )
269}
270
271#[cfg(feature = "java-forge")]
272type SharedBridge = Arc<Mutex<SubprocessBridge>>;
273
274#[cfg(feature = "java-forge")]
275struct PoolSlot {
276    bridge: SharedBridge,
277    active: usize,
278}
279
280#[cfg(feature = "java-forge")]
281pub struct JavaEnginePool {
282    config: JavaRuntimeConfig,
283    max_sessions: usize,
284    sessions_per_process: usize,
285    slots: Mutex<Vec<PoolSlot>>,
286    in_use: Mutex<HashMap<String, SharedBridge>>,
287}
288
289#[cfg(feature = "java-forge")]
290#[derive(Clone)]
291pub struct JavaEngineHandle {
292    pool: Arc<JavaEnginePool>,
293}
294
295#[cfg(feature = "java-forge")]
296impl JavaEnginePool {
297    pub fn start(
298        config: &JavaRuntimeConfig,
299        max_sessions: usize,
300        sessions_per_process: usize,
301    ) -> Result<Arc<Self>, String> {
302        let max_sessions = max_sessions.max(1);
303        let sessions_per_process = sessions_per_process.max(1);
304        let processes = max_sessions.div_ceil(sessions_per_process);
305        let mut slots = Vec::with_capacity(processes);
306        for slot in 0..processes {
307            info!(
308                slot,
309                processes, sessions_per_process, "pre-warming java subprocess"
310            );
311            let bridge = SubprocessBridge::spawn(config)?;
312            slots.push(PoolSlot {
313                bridge: Arc::new(Mutex::new(bridge)),
314                active: 0,
315            });
316        }
317        Ok(Arc::new(Self {
318            config: config.clone(),
319            max_sessions,
320            sessions_per_process,
321            slots: Mutex::new(slots),
322            in_use: Mutex::new(HashMap::new()),
323        }))
324    }
325
326    pub fn handle(self: &Arc<Self>) -> JavaEngineHandle {
327        JavaEngineHandle {
328            pool: Arc::clone(self),
329        }
330    }
331}
332
333#[cfg(feature = "java-forge")]
334impl Drop for JavaEnginePool {
335    fn drop(&mut self) {
336        let slots = self.slots.get_mut().map(std::mem::take).unwrap_or_default();
337        for slot in slots {
338            if let Ok(mutex) = Arc::try_unwrap(slot.bridge) {
339                if let Ok(inner) = mutex.into_inner() {
340                    inner.shutdown();
341                }
342            }
343        }
344    }
345}
346
347#[cfg(feature = "java-forge")]
348impl JavaEnginePool {
349    fn acquire(&self) -> Result<SharedBridge, String> {
350        let deadline = Instant::now() + Duration::from_secs(60);
351        loop {
352            let claimed = {
353                let mut slots = self
354                    .slots
355                    .lock()
356                    .map_err(|_| "java engine slots poisoned".to_string())?;
357                let mut claimed = None;
358                for slot in slots.iter_mut() {
359                    if slot.active < self.sessions_per_process {
360                        slot.active += 1;
361                        claimed = Some(Arc::clone(&slot.bridge));
362                        break;
363                    }
364                }
365                claimed
366            };
367            if let Some(bridge) = claimed {
368                let alive = bridge
369                    .lock()
370                    .ok()
371                    .map(|mut guard| guard.is_alive())
372                    .unwrap_or(false);
373                if alive {
374                    return Ok(bridge);
375                }
376                warn!("discarding dead java subprocess from pool");
377                self.replace_slot(&bridge);
378                continue;
379            }
380            if Instant::now() >= deadline {
381                return Err(format!(
382                    "java engine pool exhausted (max_sessions={}); no free session slot after 60s",
383                    self.max_sessions
384                ));
385            }
386            std::thread::sleep(Duration::from_millis(50));
387        }
388    }
389
390    fn release(&self, bridge: SharedBridge) {
391        let now_idle = {
392            let mut slots = match self.slots.lock() {
393                Ok(slots) => slots,
394                Err(_) => return,
395            };
396            match slots.iter_mut().find(|s| Arc::ptr_eq(&s.bridge, &bridge)) {
397                Some(slot) => {
398                    slot.active = slot.active.saturating_sub(1);
399                    slot.active == 0
400                }
401                None => return,
402            }
403        };
404        if !now_idle {
405            return;
406        }
407        let healthy = {
408            let mut guard = match bridge.lock() {
409                Ok(guard) => guard,
410                Err(_) => return,
411            };
412            guard.is_alive() && guard.reset().is_ok()
413        };
414        if !healthy {
415            warn!("java subprocess unhealthy at idle; respawning");
416            self.replace_slot(&bridge);
417        }
418    }
419
420    fn replace_slot(&self, dead: &SharedBridge) {
421        let Ok(mut slots) = self.slots.lock() else {
422            return;
423        };
424        let Some(index) = slots.iter().position(|s| Arc::ptr_eq(&s.bridge, dead)) else {
425            return;
426        };
427        match SubprocessBridge::spawn(&self.config) {
428            Ok(replacement) => {
429                slots[index] = PoolSlot {
430                    bridge: Arc::new(Mutex::new(replacement)),
431                    active: 0,
432                };
433            }
434            Err(error) => {
435                warn!(%error, "failed to respawn java subprocess; retiring pool slot");
436                slots.remove(index);
437            }
438        }
439    }
440}
441
442#[cfg(feature = "java-forge")]
443impl JavaEngineHandle {
444    fn bridge_for(&self, session_id: &str) -> Result<SharedBridge, String> {
445        let in_use = self
446            .pool
447            .in_use
448            .lock()
449            .map_err(|_| "java engine in_use map poisoned".to_string())?;
450        in_use
451            .get(session_id)
452            .cloned()
453            .ok_or_else(|| format!("unknown java session: {session_id}"))
454    }
455
456    pub fn start_game(&self, request_json: &str) -> Result<String, String> {
457        let bridge = self.pool.acquire()?;
458        let response = {
459            let mut guard = bridge
460                .lock()
461                .map_err(|_| "java subprocess mutex poisoned".to_string())?;
462            guard.start_game_json(request_json)
463        };
464        let response = match response {
465            Ok(response) => response,
466            Err(error) => {
467                self.pool.release(bridge);
468                return Err(error);
469            }
470        };
471        let parsed: StartGameResponse = match serde_json::from_str(&response) {
472            Ok(parsed) => parsed,
473            Err(error) => {
474                self.pool.release(bridge);
475                return Err(format!("malformed startGame response: {error}"));
476            }
477        };
478        let session_id = parsed.session_id.clone();
479        let displaced = {
480            let mut in_use = self
481                .pool
482                .in_use
483                .lock()
484                .map_err(|_| "java engine in_use map poisoned".to_string())?;
485            in_use.insert(session_id.clone(), bridge)
486        };
487        if let Some(displaced) = displaced {
488            warn!(
489                session_id,
490                "session_id collision; releasing displaced java subprocess"
491            );
492            self.pool.release(displaced);
493        }
494        Ok(session_id)
495    }
496
497    pub fn submit_action(&self, session_id: &str, action_json: &str) -> Result<String, String> {
498        let bridge = self.bridge_for(session_id)?;
499        let mutex_started = Instant::now();
500        let mut guard = bridge
501            .lock()
502            .map_err(|_| "java subprocess mutex poisoned".to_string())?;
503        crate::metrics::record_forge_decision_stage("bridge_mutex", mutex_started.elapsed());
504        let call_started = Instant::now();
505        let result = guard.submit_action(session_id, action_json);
506        crate::metrics::record_forge_decision_stage("submit_action", call_started.elapsed());
507        result
508    }
509
510    pub fn get_prompt(
511        &self,
512        session_id: &str,
513        player_index: usize,
514    ) -> Result<Option<String>, String> {
515        let bridge = self.bridge_for(session_id)?;
516        let mut guard = bridge
517            .lock()
518            .map_err(|_| "java subprocess mutex poisoned".to_string())?;
519        guard.get_prompt(session_id, player_index)
520    }
521
522    pub fn is_game_over(&self, session_id: &str) -> Result<bool, String> {
523        let bridge = self.bridge_for(session_id)?;
524        let mut guard = bridge
525            .lock()
526            .map_err(|_| "java subprocess mutex poisoned".to_string())?;
527        guard.is_game_over(session_id)
528    }
529
530    pub fn get_snapshot(&self, session_id: &str, viewer: Option<usize>) -> Result<String, String> {
531        let bridge = self.bridge_for(session_id)?;
532        let mut guard = bridge
533            .lock()
534            .map_err(|_| "java subprocess mutex poisoned".to_string())?;
535        guard.get_snapshot(session_id, viewer)
536    }
537
538    pub fn end_game(&self, session_id: &str) -> Result<(), String> {
539        let bridge = {
540            let mut in_use = self
541                .pool
542                .in_use
543                .lock()
544                .map_err(|_| "java engine in_use map poisoned".to_string())?;
545            in_use.remove(session_id)
546        };
547        let Some(bridge) = bridge else {
548            return Ok(());
549        };
550        let result = {
551            let mut guard = bridge
552                .lock()
553                .map_err(|_| "java subprocess mutex poisoned".to_string())?;
554            guard.end_game(session_id)
555        };
556        self.pool.release(bridge);
557        result
558    }
559
560    pub fn abort_game(&self, session_id: &str) -> Result<(), String> {
561        let bridge = {
562            let mut in_use = self
563                .pool
564                .in_use
565                .lock()
566                .map_err(|_| "java engine in_use map poisoned".to_string())?;
567            in_use.remove(session_id)
568        };
569        let Some(bridge) = bridge else {
570            return Ok(());
571        };
572        let result = {
573            let mut guard = bridge
574                .lock()
575                .map_err(|_| "java subprocess mutex poisoned".to_string())?;
576            guard.abort_game(session_id)
577        };
578        self.pool.release(bridge);
579        result
580    }
581}
582
583#[cfg(feature = "java-forge")]
584static JAVA_ENGINE: std::sync::OnceLock<Arc<JavaEnginePool>> = std::sync::OnceLock::new();
585
586#[cfg(feature = "java-forge")]
587pub fn init_engine() -> Result<(), String> {
588    if JAVA_ENGINE.get().is_some() {
589        return Ok(());
590    }
591    let config = JavaRuntimeConfig::from_env();
592    // SELF_HOSTED_NODE_MAX_GAMES is the concurrent-session ceiling for this node.
593    // SELF_HOSTED_NODE_GAMES_PER_JVM multiplexes that many sessions into each
594    // subprocess (the engine is concurrency-safe since the endstep patches), so
595    // the pool spawns ceil(max_games / games_per_jvm) processes. Default 1 keeps
596    // the historical one-subprocess-per-game shape.
597    let max_sessions = env::var("SELF_HOSTED_NODE_MAX_GAMES")
598        .ok()
599        .and_then(|value| value.parse::<usize>().ok())
600        .filter(|n| *n >= 1)
601        .unwrap_or(1);
602    let sessions_per_process = env::var("SELF_HOSTED_NODE_GAMES_PER_JVM")
603        .ok()
604        .and_then(|value| value.parse::<usize>().ok())
605        .filter(|n| *n >= 1)
606        .unwrap_or(1);
607    let pool = JavaEnginePool::start(&config, max_sessions, sessions_per_process)?;
608    JAVA_ENGINE
609        .set(pool)
610        .map_err(|_| "java engine already initialized".to_string())
611}
612
613#[cfg(all(feature = "graal-forge", not(feature = "java-forge")))]
614pub fn init_engine() -> Result<(), String> {
615    // Loading the card database into the isolate takes tens of seconds on a
616    // small box. Pay it here, before any room is advertised, rather than
617    // leaving it for whoever starts the first game.
618    if shared_isolate_enabled() {
619        drop(GraalEngineHandle::create(
620            &JavaRuntimeConfig::from_env().assets_dir,
621        )?);
622    }
623    Ok(())
624}
625
626#[cfg(not(forge_backend))]
627pub fn init_engine() -> Result<(), String> {
628    Err(
629        "forge engine requires building self-hosted-node with --features java-forge or graal-forge"
630            .to_string(),
631    )
632}
633
634#[cfg(feature = "java-forge")]
635fn engine_handle() -> Result<JavaEngineHandle, String> {
636    JAVA_ENGINE
637        .get()
638        .map(JavaEnginePool::handle)
639        .ok_or_else(|| "java engine is not initialized".to_string())
640}
641
642#[cfg(feature = "java-forge")]
643type ForgeEngine = JavaEngineHandle;
644
645#[cfg(feature = "java-forge")]
646fn obtain_engine() -> Result<ForgeEngine, String> {
647    engine_handle()
648}
649
650#[cfg(all(feature = "graal-forge", not(feature = "java-forge")))]
651type ForgeEngine = GraalEngineHandle;
652
653#[cfg(all(feature = "graal-forge", not(feature = "java-forge")))]
654fn obtain_engine() -> Result<ForgeEngine, String> {
655    GraalEngineHandle::create(&JavaRuntimeConfig::from_env().assets_dir)
656}
657
658#[cfg(feature = "graal-forge")]
659mod graal_ffi {
660    use std::os::raw::{c_char, c_int};
661
662    #[allow(non_camel_case_types)]
663    pub type graal_isolate_t = std::ffi::c_void;
664    #[allow(non_camel_case_types)]
665    pub type graal_isolatethread_t = std::ffi::c_void;
666
667    extern "C" {
668        pub fn graal_create_isolate(
669            params: *mut std::ffi::c_void,
670            isolate: *mut *mut graal_isolate_t,
671            thread: *mut *mut graal_isolatethread_t,
672        ) -> c_int;
673        pub fn graal_tear_down_isolate(thread: *mut graal_isolatethread_t) -> c_int;
674        pub fn graal_attach_thread(
675            isolate: *mut graal_isolate_t,
676            thread: *mut *mut graal_isolatethread_t,
677        ) -> c_int;
678        pub fn graal_detach_thread(thread: *mut graal_isolatethread_t) -> c_int;
679        pub fn forge_initialize(
680            thread: *mut graal_isolatethread_t,
681            assets_dir: *const c_char,
682        ) -> *mut c_char;
683        pub fn forge_start_game(
684            thread: *mut graal_isolatethread_t,
685            request_json: *const c_char,
686        ) -> *mut c_char;
687        pub fn forge_submit_action(
688            thread: *mut graal_isolatethread_t,
689            session_id: *const c_char,
690            action_json: *const c_char,
691        ) -> *mut c_char;
692        pub fn forge_get_prompt(
693            thread: *mut graal_isolatethread_t,
694            session_id: *const c_char,
695            player_index: c_int,
696        ) -> *mut c_char;
697        pub fn forge_get_snapshot(
698            thread: *mut graal_isolatethread_t,
699            session_id: *const c_char,
700            viewer: c_int,
701        ) -> *mut c_char;
702        pub fn forge_get_game_over(
703            thread: *mut graal_isolatethread_t,
704            session_id: *const c_char,
705        ) -> *mut c_char;
706        pub fn forge_end_game(
707            thread: *mut graal_isolatethread_t,
708            session_id: *const c_char,
709        ) -> *mut c_char;
710        pub fn forge_abort_game(
711            thread: *mut graal_isolatethread_t,
712            session_id: *const c_char,
713        ) -> *mut c_char;
714        pub fn forge_free_string(thread: *mut graal_isolatethread_t, ptr: *mut c_char);
715        pub fn forge_dump_heap(
716            thread: *mut graal_isolatethread_t,
717            path: *const c_char,
718            live_objects_only: c_int,
719        ) -> *mut c_char;
720        pub fn forge_gc_stats(thread: *mut graal_isolatethread_t) -> *mut c_char;
721    }
722}
723
724/// Isolate-wide GC and heap counters, as reported by the engine itself.
725/// `collections` and `pause_millis` are -1 when the native image exposes no
726/// collector beans; the heap figures are always present.
727#[cfg(feature = "graal-forge")]
728#[derive(serde::Deserialize)]
729#[serde(rename_all = "camelCase")]
730pub struct GcStats {
731    pub heap_used_bytes: u64,
732    pub heap_total_bytes: u64,
733    pub heap_max_bytes: u64,
734    pub collections: i64,
735    pub pause_millis: i64,
736    /// Keyed by collector name. An incremental collection is a few milliseconds;
737    /// a complete one traces the whole live set, which here is a permanently
738    /// reachable card database, so it costs the same whether it reclaims
739    /// anything or not. The totals above hide that difference.
740    #[serde(default)]
741    pub by_collector: std::collections::HashMap<String, GcCollector>,
742    /// Wall clock the engine was stopped, measured inside the isolate by sleep
743    /// overshoot rather than by the collector, because a G1 image registers no
744    /// collector beans and leaves every field above at zero. Counts scheduler
745    /// preemption as well as collection pauses.
746    #[serde(default)]
747    pub stalled_millis: u64,
748    #[serde(default)]
749    pub max_stall_millis: u64,
750    #[serde(default)]
751    pub long_stalls: u64,
752}
753
754#[cfg(feature = "graal-forge")]
755#[derive(serde::Deserialize)]
756#[serde(rename_all = "camelCase")]
757pub struct GcCollector {
758    pub collections: u64,
759    pub pause_millis: u64,
760}
761
762#[cfg(feature = "graal-forge")]
763#[derive(serde::Deserialize)]
764struct ForgeReply {
765    ok: bool,
766    #[serde(default)]
767    result: String,
768    #[serde(default)]
769    error: Option<String>,
770}
771
772// A GraalVM isolate hosts the in-process Forge engine. The `thread` handle is
773// bound to the hosted-engine thread that created or attached it (isolate
774// threads are not portable), so the handle is Rc/!Send by design. In shared
775// mode one isolate serves all rooms: the first creator initializes Forge, later
776// rooms attach their own thread to it and sessions coexist in the adapter.
777#[cfg(feature = "graal-forge")]
778struct GraalBridge {
779    thread: *mut graal_ffi::graal_isolatethread_t,
780    attached: bool,
781}
782
783#[cfg(feature = "graal-forge")]
784struct SharedIsolate(*mut graal_ffi::graal_isolate_t);
785#[cfg(feature = "graal-forge")]
786unsafe impl Send for SharedIsolate {}
787
788#[cfg(feature = "graal-forge")]
789static SHARED_GRAAL_ISOLATE: std::sync::Mutex<Option<SharedIsolate>> = std::sync::Mutex::new(None);
790
791// One isolate hosts every room's game (safe since the endstep concurrency
792// patches), so the card database loads once per node. Off keeps the historical
793// isolate-per-game shape.
794#[cfg(feature = "graal-forge")]
795fn shared_isolate_enabled() -> bool {
796    let asked = env::var("SELF_HOSTED_NODE_SHARED_ISOLATE")
797        .map(|value| value == "1" || value.eq_ignore_ascii_case("true"))
798        .unwrap_or(false);
799    // G1 permits one isolate per process and aborts the process on the second:
800    //   guarantee(SVMIsolateData::_heap_base == nullptr) failed:
801    //   G1 doesn't support multiple isolates at the moment.
802    // The published image is built with G1, and the collector is fixed at build
803    // time and not visible from here, so isolate-per-game cannot be honoured
804    // without risking that abort on the second room. One isolate is correct for
805    // a Serial build too — it is what production has run since the card
806    // database became shared — so this is forced rather than merely defaulted.
807    if !asked {
808        warn!(
809            "isolate-per-game is not available: the engine image is built with G1, \
810             which aborts on a second isolate. Hosting every room in one isolate."
811        );
812    }
813    true
814}
815
816#[cfg(feature = "graal-forge")]
817impl GraalBridge {
818    fn create() -> Result<Self, String> {
819        let mut isolate: *mut graal_ffi::graal_isolate_t = std::ptr::null_mut();
820        let mut thread: *mut graal_ffi::graal_isolatethread_t = std::ptr::null_mut();
821        let rc = unsafe {
822            graal_ffi::graal_create_isolate(std::ptr::null_mut(), &mut isolate, &mut thread)
823        };
824        if rc != 0 {
825            return Err(format!("graal_create_isolate failed with code {rc}"));
826        }
827        Ok(Self {
828            thread,
829            attached: false,
830        })
831    }
832
833    fn create_in_shared_isolate(assets_dir: &Path) -> Result<Self, String> {
834        let mut guard = SHARED_GRAAL_ISOLATE
835            .lock()
836            .map_err(|_| "shared graal isolate poisoned".to_string())?;
837        if let Some(shared) = guard.as_ref() {
838            let mut thread: *mut graal_ffi::graal_isolatethread_t = std::ptr::null_mut();
839            let rc = unsafe { graal_ffi::graal_attach_thread(shared.0, &mut thread) };
840            if rc != 0 {
841                return Err(format!("graal_attach_thread failed with code {rc}"));
842            }
843            return Ok(Self {
844                thread,
845                attached: true,
846            });
847        }
848        let mut isolate: *mut graal_ffi::graal_isolate_t = std::ptr::null_mut();
849        let mut thread: *mut graal_ffi::graal_isolatethread_t = std::ptr::null_mut();
850        let rc = unsafe {
851            graal_ffi::graal_create_isolate(std::ptr::null_mut(), &mut isolate, &mut thread)
852        };
853        if rc != 0 {
854            return Err(format!("graal_create_isolate failed with code {rc}"));
855        }
856        let mut bridge = Self {
857            thread,
858            attached: false,
859        };
860        let assets = cstring(&assets_dir.to_string_lossy())?;
861        bridge.decode(unsafe { graal_ffi::forge_initialize(bridge.thread, assets.as_ptr()) })?;
862        bridge.attached = true;
863        *guard = Some(SharedIsolate(isolate));
864        info!("shared graal isolate initialized");
865        Ok(bridge)
866    }
867
868    fn decode(&self, raw: *mut std::os::raw::c_char) -> Result<String, String> {
869        if raw.is_null() {
870            return Err("forge native lib returned null".to_string());
871        }
872        let envelope = unsafe { std::ffi::CStr::from_ptr(raw) }
873            .to_string_lossy()
874            .into_owned();
875        unsafe { graal_ffi::forge_free_string(self.thread, raw) };
876        let reply: ForgeReply = serde_json::from_str(&envelope)
877            .map_err(|err| format!("malformed forge envelope: {err}"))?;
878        if reply.ok {
879            Ok(reply.result)
880        } else {
881            Err(reply
882                .error
883                .unwrap_or_else(|| "unknown forge error".to_string()))
884        }
885    }
886}
887
888#[cfg(feature = "graal-forge")]
889impl Drop for GraalBridge {
890    fn drop(&mut self) {
891        if self.attached {
892            unsafe { graal_ffi::graal_detach_thread(self.thread) };
893        } else {
894            // Reachable only from the isolate-per-game path, which
895            // `shared_isolate_enabled` forces off. Teardown refuses while any
896            // thread is still attached, and the engine now keeps a permanent
897            // stall-probe thread inside the isolate, so restoring that path
898            // means giving the probe a way to be stopped first.
899            unsafe { graal_ffi::graal_tear_down_isolate(self.thread) };
900        }
901    }
902}
903
904#[cfg(feature = "graal-forge")]
905#[derive(Clone)]
906struct GraalEngineHandle {
907    bridge: std::rc::Rc<GraalBridge>,
908}
909
910#[cfg(feature = "graal-forge")]
911impl GraalEngineHandle {
912    fn create(assets_dir: &Path) -> Result<Self, String> {
913        let shared = shared_isolate_enabled();
914        let bridge = if shared {
915            GraalBridge::create_in_shared_isolate(assets_dir)?
916        } else {
917            let bridge = GraalBridge::create()?;
918            let assets = cstring(&assets_dir.to_string_lossy())?;
919            bridge
920                .decode(unsafe { graal_ffi::forge_initialize(bridge.thread, assets.as_ptr()) })?;
921            bridge
922        };
923        if shared {
924            start_engine_gc_sampler();
925        }
926        Ok(Self {
927            bridge: std::rc::Rc::new(bridge),
928        })
929    }
930
931    fn start_game(&self, request_json: &str) -> Result<String, String> {
932        let request = cstring(request_json)?;
933        let response = self
934            .bridge
935            .decode(unsafe { graal_ffi::forge_start_game(self.bridge.thread, request.as_ptr()) })?;
936        let parsed: StartGameResponse = serde_json::from_str(&response)
937            .map_err(|err| format!("malformed startGame response: {err}"))?;
938        Ok(parsed.session_id)
939    }
940
941    fn submit_action(&self, session_id: &str, action_json: &str) -> Result<String, String> {
942        let session = cstring(session_id)?;
943        let action = cstring(action_json)?;
944        let started = Instant::now();
945        let result = self.bridge.decode(unsafe {
946            graal_ffi::forge_submit_action(self.bridge.thread, session.as_ptr(), action.as_ptr())
947        });
948        crate::metrics::record_forge_decision_stage("submit_action", started.elapsed());
949        result
950    }
951
952    /// Cumulative collections, pause time and heap for the whole isolate. Every
953    /// room on this node shares it, so a collection here stops all of them.
954    fn gc_stats(&self) -> Result<GcStats, String> {
955        gc_stats_via(&self.bridge)
956    }
957
958    fn get_prompt(&self, session_id: &str, player_index: usize) -> Result<Option<String>, String> {
959        let session = cstring(session_id)?;
960        let prompt = self.bridge.decode(unsafe {
961            graal_ffi::forge_get_prompt(
962                self.bridge.thread,
963                session.as_ptr(),
964                player_index as std::os::raw::c_int,
965            )
966        })?;
967        Ok((!prompt.is_empty()).then_some(prompt))
968    }
969
970    fn is_game_over(&self, session_id: &str) -> Result<bool, String> {
971        let session = cstring(session_id)?;
972        let value = self.bridge.decode(unsafe {
973            graal_ffi::forge_get_game_over(self.bridge.thread, session.as_ptr())
974        })?;
975        Ok(value.trim() == "true")
976    }
977
978    fn get_snapshot(&self, session_id: &str, viewer: Option<usize>) -> Result<String, String> {
979        let session = cstring(session_id)?;
980        let viewer = viewer.map_or(-1, |v| v as std::os::raw::c_int);
981        self.bridge.decode(unsafe {
982            graal_ffi::forge_get_snapshot(self.bridge.thread, session.as_ptr(), viewer)
983        })
984    }
985
986    fn end_game(&self, session_id: &str) -> Result<(), String> {
987        let session = cstring(session_id)?;
988        self.bridge
989            .decode(unsafe { graal_ffi::forge_end_game(self.bridge.thread, session.as_ptr()) })
990            .map(|_| ())
991    }
992
993    fn abort_game(&self, session_id: &str) -> Result<(), String> {
994        let session = cstring(session_id)?;
995        self.bridge
996            .decode(unsafe { graal_ffi::forge_abort_game(self.bridge.thread, session.as_ptr()) })
997            .map(|_| ())
998    }
999}
1000
1001/// Write an hprof dump of the shared isolate's heap. Attaches its own isolate
1002/// thread, so it is callable from a signal handler task rather than the engine
1003/// thread that owns the room's bridge.
1004#[cfg(feature = "graal-forge")]
1005pub fn dump_shared_heap(path: &str) -> Result<(), String> {
1006    let isolate = SHARED_GRAAL_ISOLATE
1007        .lock()
1008        .map_err(|_| "shared graal isolate poisoned".to_string())?
1009        .as_ref()
1010        .map(|shared| shared.0)
1011        .ok_or_else(|| "no shared isolate to dump".to_string())?;
1012    let mut thread: *mut graal_ffi::graal_isolatethread_t = std::ptr::null_mut();
1013    let rc = unsafe { graal_ffi::graal_attach_thread(isolate, &mut thread) };
1014    if rc != 0 {
1015        return Err(format!("graal_attach_thread failed with code {rc}"));
1016    }
1017    let bridge = GraalBridge {
1018        thread,
1019        attached: true,
1020    };
1021    let target = cstring(path)?;
1022    bridge
1023        .decode(unsafe { graal_ffi::forge_dump_heap(bridge.thread, target.as_ptr(), 1) })
1024        .map(|_| ())
1025}
1026
1027#[cfg(feature = "graal-forge")]
1028fn cstring(value: &str) -> Result<std::ffi::CString, String> {
1029    std::ffi::CString::new(value).map_err(|_| "string contained interior NUL".to_string())
1030}
1031
1032#[cfg(feature = "java-forge")]
1033pub fn run_concurrent_self_play(
1034    seats: &[DeckSelection],
1035    starting_life: i32,
1036    seed: u64,
1037    max_prompts: usize,
1038    concurrency: usize,
1039) -> Result<(), String> {
1040    let config = JavaRuntimeConfig::from_env();
1041    let games_per_process = env::var("SELF_HOSTED_NODE_GAMES_PER_JVM")
1042        .ok()
1043        .and_then(|value| value.parse::<usize>().ok())
1044        .filter(|n| *n >= 1)
1045        .unwrap_or(1);
1046    let pool = JavaEnginePool::start(&config, concurrency.max(1), games_per_process)?;
1047    info!(
1048        concurrency,
1049        "java-engine started; launching concurrent games"
1050    );
1051
1052    let mut players = Vec::with_capacity(seats.len());
1053    for (i, seat) in seats.iter().enumerate() {
1054        let identities = deck_card_identities(&seat.deck);
1055        players.push(PlayerConfig::new(
1056            format!("Self-Play {}", i + 1),
1057            &identities,
1058            commander_names_for_java(&seat.deck, seat.commander_name.as_deref()),
1059        ));
1060    }
1061
1062    let mut joins = Vec::with_capacity(concurrency.max(1));
1063    for game_index in 0..concurrency.max(1) {
1064        let handle = pool.handle();
1065        let request = StartGameRequest::new(
1066            format!("self-hosted-java-concurrent-{game_index}"),
1067            String::new(),
1068            starting_life,
1069            seed.wrapping_add(game_index as u64),
1070            players.clone(),
1071        );
1072        joins.push(std::thread::spawn(move || -> Result<(), String> {
1073            let request_json = request.to_json().map_err(|error| error.to_string())?;
1074            let session_id = handle.start_game(&request_json)?;
1075            info!(session_id, game_index, "concurrent java game started");
1076            let result = drive_game_via_handle(&handle, &session_id, max_prompts);
1077            let _ = handle.end_game(&session_id);
1078            result
1079        }));
1080    }
1081
1082    let mut outcome = Ok(());
1083    for join in joins {
1084        match join.join() {
1085            Ok(Ok(())) => {}
1086            Ok(Err(error)) => outcome = Err(error),
1087            Err(_) => outcome = Err("concurrent game thread panicked".to_string()),
1088        }
1089    }
1090    outcome
1091}
1092
1093#[cfg(not(feature = "java-forge"))]
1094pub fn run_concurrent_self_play(
1095    _seats: &[DeckSelection],
1096    _starting_life: i32,
1097    _seed: u64,
1098    _max_prompts: usize,
1099    _concurrency: usize,
1100) -> Result<(), String> {
1101    Err(
1102        "java-forge concurrent self-play requires building self-hosted-node with --features java-forge"
1103            .to_string(),
1104    )
1105}
1106
1107#[cfg(all(feature = "graal-forge", not(feature = "java-forge")))]
1108fn report_engine_gc(engine: &ForgeEngine) {
1109    match engine.gc_stats() {
1110        Ok(stats) => publish_engine_gc(&stats),
1111        Err(err) => debug!(%err, "engine gc stats unavailable"),
1112    }
1113}
1114
1115#[cfg(feature = "graal-forge")]
1116fn publish_engine_gc(stats: &GcStats) {
1117    crate::metrics::record_engine_gc(
1118        stats.collections,
1119        stats.pause_millis,
1120        stats.heap_used_bytes,
1121        stats.heap_max_bytes,
1122    );
1123    crate::metrics::record_engine_stall(
1124        stats.stalled_millis,
1125        stats.max_stall_millis,
1126        stats.long_stalls,
1127    );
1128    for (name, per) in &stats.by_collector {
1129        crate::metrics::record_engine_gc_collector(name, per.collections, per.pause_millis);
1130    }
1131}
1132
1133#[cfg(feature = "graal-forge")]
1134fn gc_stats_via(bridge: &GraalBridge) -> Result<GcStats, String> {
1135    let raw = bridge.decode(unsafe { graal_ffi::forge_gc_stats(bridge.thread) })?;
1136    serde_json::from_str(&raw).map_err(|err| format!("malformed gcStats response: {err}"))
1137}
1138
1139#[cfg(feature = "graal-forge")]
1140const ENGINE_GC_SAMPLE_INTERVAL: Duration = Duration::from_secs(15);
1141
1142/// Sampling on the decision path made an idle node and a wedged node look the
1143/// same: both stop reporting. A node that is stalling is exactly the node that
1144/// stops completing decisions, so the sample has to come from a thread that is
1145/// not on that path. It attaches its own isolate thread because the handles the
1146/// rooms hold are bound to their creating thread.
1147#[cfg(feature = "graal-forge")]
1148fn start_engine_gc_sampler() {
1149    static STARTED: std::sync::Once = std::sync::Once::new();
1150    STARTED.call_once(|| {
1151        let spawned = std::thread::Builder::new()
1152            .name("engine-gc-sampler".to_string())
1153            .spawn(|| {
1154                let isolate = match SHARED_GRAAL_ISOLATE.lock() {
1155                    Ok(guard) => guard.as_ref().map(|shared| shared.0),
1156                    Err(_) => None,
1157                };
1158                let Some(isolate) = isolate else {
1159                    return;
1160                };
1161                let mut thread: *mut graal_ffi::graal_isolatethread_t = std::ptr::null_mut();
1162                let rc = unsafe { graal_ffi::graal_attach_thread(isolate, &mut thread) };
1163                if rc != 0 {
1164                    warn!(rc, "gc sampler could not attach to the engine isolate");
1165                    return;
1166                }
1167                let bridge = GraalBridge {
1168                    thread,
1169                    attached: true,
1170                };
1171                loop {
1172                    std::thread::sleep(ENGINE_GC_SAMPLE_INTERVAL);
1173                    match gc_stats_via(&bridge) {
1174                        Ok(stats) => publish_engine_gc(&stats),
1175                        Err(err) => debug!(%err, "engine gc stats unavailable"),
1176                    }
1177                }
1178            });
1179        if let Err(err) = spawned {
1180            warn!(%err, "could not start the engine gc sampler");
1181        }
1182    });
1183}
1184
1185/// The subprocess backend already reports this by parsing its own GC log.
1186#[cfg(feature = "java-forge")]
1187fn report_engine_gc(_engine: &ForgeEngine) {}
1188
1189#[cfg(feature = "java-forge")]
1190fn drive_game_via_handle(
1191    handle: &JavaEngineHandle,
1192    session_id: &str,
1193    max_prompts: usize,
1194) -> Result<(), String> {
1195    let mut bots: HashMap<usize, SimpleAi> = HashMap::new();
1196    let mut last_prompt: Option<String> = None;
1197    let mut acted = 0usize;
1198    let mut seen_prompt = false;
1199    let max_iterations = max_prompts.saturating_mul(200).max(2_000);
1200
1201    for _ in 0..max_iterations {
1202        if let Some(prompt_json) = handle.get_prompt(session_id, 0)? {
1203            seen_prompt = true;
1204            if last_prompt.as_deref() == Some(prompt_json.as_str()) {
1205                if handle.is_game_over(session_id)? {
1206                    return Ok(());
1207                }
1208                std::thread::sleep(Duration::from_millis(20));
1209                continue;
1210            }
1211            let prompt: AgentPrompt = serde_json::from_str(&prompt_json)
1212                .map_err(|error| format!("failed to parse concurrent prompt: {error}"))?;
1213            let player = player_index(&prompt.deciding_player_id);
1214            if let Some(action) = bots.entry(player).or_default().decide(prompt) {
1215                let action_json = serde_json::to_string(&action).map_err(|err| err.to_string())?;
1216                handle.submit_action(session_id, &action_json)?;
1217                acted += 1;
1218                if acted >= max_prompts {
1219                    return Err(format!(
1220                        "concurrent game {session_id} did not finish within {max_prompts} decisions"
1221                    ));
1222                }
1223            }
1224            last_prompt = Some(prompt_json);
1225            continue;
1226        }
1227        if seen_prompt && handle.is_game_over(session_id)? {
1228            return Ok(());
1229        }
1230        std::thread::sleep(Duration::from_millis(20));
1231    }
1232    Err(format!(
1233        "concurrent game {session_id} exceeded its iteration cap"
1234    ))
1235}
1236
1237/// G1 keeps pauses roughly bounded as the live set grows; Serial does not.
1238const DEFAULT_JAVA_COLLECTOR: &str = "G1";
1239const DEFAULT_JAVA_GC_LOG: &str = "stderr";
1240const DEFAULT_JAVA_HEAP_MB: u64 = 1024;
1241const DEFAULT_JAVA_ACTIVE_PROCESSORS: u64 = 2;
1242
1243#[derive(Debug, Clone)]
1244pub struct JavaRuntimeConfig {
1245    pub assets_dir: PathBuf,
1246    pub harness_jar: PathBuf,
1247    pub java_home: Option<PathBuf>,
1248    pub extra_classpath: Vec<PathBuf>,
1249    pub heap_mb: Option<u64>,
1250    pub active_processor_count: Option<u64>,
1251    pub gc_log: Option<String>,
1252    pub collector: Option<String>,
1253    pub extra_jvm_args: Vec<String>,
1254}
1255
1256impl JavaRuntimeConfig {
1257    pub fn from_env() -> Self {
1258        let root = workspace_root();
1259        Self {
1260            assets_dir: env_path("SELF_HOSTED_NODE_FORGE_ASSETS_DIR")
1261                .or_else(|| env_path("MANA_BREW_FORGE_ASSETS_DIR"))
1262                .unwrap_or_else(|| root.join("forge/forge-gui")),
1263            harness_jar: env_path("SELF_HOSTED_NODE_FORGE_HARNESS_JAR")
1264                .or_else(|| env_path("MANA_BREW_FORGE_HARNESS_JAR"))
1265                .unwrap_or_else(|| {
1266                    root.join("forge-harness/target/forge-harness-jar-with-dependencies.jar")
1267                }),
1268            java_home: env_path("SELF_HOSTED_NODE_JAVA_HOME")
1269                .or_else(|| env_path("MANA_BREW_JAVA_HOME"))
1270                .or_else(|| env_path("JAVA_HOME")),
1271            extra_classpath: env_classpath("SELF_HOSTED_NODE_FORGE_EXTRA_CLASSPATH")
1272                .into_iter()
1273                .chain(env_classpath("MANA_BREW_FORGE_EXTRA_CLASSPATH"))
1274                .collect(),
1275            heap_mb: env_sizing("SELF_HOSTED_NODE_JAVA_HEAP_MB", DEFAULT_JAVA_HEAP_MB),
1276            active_processor_count: env_sizing(
1277                "SELF_HOSTED_NODE_JAVA_ACTIVE_PROCESSORS",
1278                DEFAULT_JAVA_ACTIVE_PROCESSORS,
1279            ),
1280            collector: env::var("SELF_HOSTED_NODE_JAVA_COLLECTOR")
1281                .ok()
1282                .map(|value| value.trim().to_string())
1283                .filter(|value| !value.is_empty())
1284                .or_else(|| Some(DEFAULT_JAVA_COLLECTOR.to_string())),
1285            // Defaults on: the fleet ships no logs, so this log exists to be
1286            // turned into metrics by the stderr pump (see metrics.rs).
1287            gc_log: env::var("SELF_HOSTED_NODE_JAVA_GC_LOG")
1288                .ok()
1289                .map(|value| value.trim().to_string())
1290                .filter(|value| !value.is_empty())
1291                .or_else(|| Some(DEFAULT_JAVA_GC_LOG.to_string())),
1292            extra_jvm_args: env::var("SELF_HOSTED_NODE_JAVA_OPTS")
1293                .unwrap_or_default()
1294                .split_whitespace()
1295                .map(str::to_string)
1296                .collect(),
1297        }
1298    }
1299
1300    #[cfg(feature = "java-forge")]
1301    fn jvm_args(&self) -> Vec<String> {
1302        let mut args = vec![
1303            "-Dfile.encoding=UTF-8".to_string(),
1304            "-Dsun.stdout.encoding=UTF-8".to_string(),
1305            "-Dsun.stderr.encoding=UTF-8".to_string(),
1306            "-Djava.awt.headless=true".to_string(),
1307            // Forge's endstep concurrency patches assume a heap exhaustion kills the process
1308            // instead of thrashing; a supervised node is restarted, a thrashing one is not.
1309            "-XX:+ExitOnOutOfMemoryError".to_string(),
1310            // Forge calls System.gc() on match boundaries (Match.java,
1311            // HostedMatch.java) and so does the harness. Each one is a full
1312            // collection, and a full collection costs about 1.3ms per MB of
1313            // live set, so on a large board they are seconds of stop-the-world
1314            // for no benefit the collector would not have reached anyway.
1315            "-XX:+DisableExplicitGC".to_string(),
1316        ];
1317        // Pause time is linear in the live set and Serial collects on one
1318        // thread, so a bigger heap under Serial means longer freezes, not
1319        // shorter ones. Name the collector rather than inheriting whatever the
1320        // host's environment happens to set.
1321        if let Some(collector) = &self.collector {
1322            args.push(format!("-XX:+Use{collector}GC"));
1323        }
1324        // A fleet runs many of these side by side, and every JVM sizes its heap and its GC
1325        // thread count from the whole machine unless told otherwise.
1326        if let Some(heap_mb) = self.heap_mb {
1327            args.push(format!("-Xmx{heap_mb}m"));
1328        }
1329        if let Some(processors) = self.active_processor_count {
1330            args.push(format!("-XX:ActiveProcessorCount={processors}"));
1331        }
1332        // In a container the only log that reaches Loki is the node's own, and the
1333        // subprocess owns stdout for the protocol, so "stderr" routes GC through the
1334        // stderr pump instead of a file nothing ships.
1335        match self.gc_log.as_deref() {
1336            None => {}
1337            Some("stderr") => args.push("-Xlog:gc*:stderr:time,uptime,level,tags".to_string()),
1338            Some(dir) => args.push(format!(
1339                "-Xlog:gc*:file={}:time,uptime,level,tags:filecount=5,filesize=20M",
1340                Path::new(dir).join("engine-gc-%p.log").display()
1341            )),
1342        }
1343        args.extend(self.extra_jvm_args.iter().cloned());
1344        args
1345    }
1346
1347    pub fn validate(&self) -> Result<(), String> {
1348        require_dir(&self.assets_dir, "Forge assets directory")?;
1349        require_file(&self.harness_jar, "Forge harness jar")?;
1350        if let Some(java_home) = &self.java_home {
1351            require_dir(java_home, "Java home")?;
1352        }
1353        for entry in &self.extra_classpath {
1354            if !entry.exists() {
1355                return Err(format!(
1356                    "Classpath entry does not exist: {}",
1357                    entry.display()
1358                ));
1359            }
1360        }
1361        Ok(())
1362    }
1363
1364    pub fn classpath_entries(&self) -> Vec<PathBuf> {
1365        let mut entries = Vec::with_capacity(1 + self.extra_classpath.len());
1366        entries.push(self.harness_jar.clone());
1367        entries.extend(self.extra_classpath.iter().cloned());
1368        entries
1369    }
1370}
1371
1372#[cfg(forge_backend)]
1373#[allow(clippy::too_many_arguments)]
1374pub fn run_hosted_engine_game(
1375    game_id: String,
1376    player_names: Vec<String>,
1377    decks: Vec<Deck>,
1378    commander_names: Vec<Option<String>>,
1379    commander_variant: bool,
1380    game_variant: String,
1381    local_player_index: Option<usize>,
1382    ai_player_indices: Vec<usize>,
1383    starting_life: i32,
1384    remote_prompt_tx: std_mpsc::Sender<(usize, AgentMessage)>,
1385    remote_response_rxs: Vec<(usize, std_mpsc::Receiver<ClientToServerMessage>)>,
1386    game_over_tx: std_mpsc::Sender<HostedGameOver>,
1387    cancel: Arc<AtomicBool>,
1388) -> Result<(), String> {
1389    run_hosted_engine_game_inner(
1390        game_id,
1391        player_names,
1392        decks,
1393        commander_names,
1394        commander_variant,
1395        game_variant,
1396        local_player_index,
1397        ai_player_indices,
1398        starting_life,
1399        remote_prompt_tx,
1400        remote_response_rxs,
1401        game_over_tx,
1402        cancel,
1403    )
1404}
1405
1406#[cfg(not(forge_backend))]
1407#[allow(clippy::too_many_arguments)]
1408pub fn run_hosted_engine_game(
1409    _game_id: String,
1410    _player_names: Vec<String>,
1411    _decks: Vec<Deck>,
1412    _commander_names: Vec<Option<String>>,
1413    _commander_variant: bool,
1414    _game_variant: String,
1415    _local_player_index: Option<usize>,
1416    _ai_player_indices: Vec<usize>,
1417    _starting_life: i32,
1418    _remote_prompt_tx: std_mpsc::Sender<(usize, AgentMessage)>,
1419    _remote_response_rxs: Vec<(usize, std_mpsc::Receiver<ClientToServerMessage>)>,
1420    _game_over_tx: std_mpsc::Sender<HostedGameOver>,
1421    _cancel: Arc<AtomicBool>,
1422) -> Result<(), String> {
1423    Err(unsupported_message().to_string())
1424}
1425
1426#[cfg(forge_backend)]
1427#[allow(clippy::too_many_arguments)]
1428fn run_hosted_engine_game_inner(
1429    game_id: String,
1430    player_names: Vec<String>,
1431    decks: Vec<Deck>,
1432    commander_names: Vec<Option<String>>,
1433    commander_variant: bool,
1434    game_variant: String,
1435    local_player_index: Option<usize>,
1436    ai_player_indices: Vec<usize>,
1437    starting_life: i32,
1438    remote_prompt_tx: std_mpsc::Sender<(usize, AgentMessage)>,
1439    remote_response_rxs: Vec<(usize, std_mpsc::Receiver<ClientToServerMessage>)>,
1440    game_over_tx: std_mpsc::Sender<HostedGameOver>,
1441    cancel: Arc<AtomicBool>,
1442) -> Result<(), String> {
1443    let engine = obtain_engine()?;
1444
1445    let mut players = Vec::with_capacity(player_names.len());
1446    for (index, name) in player_names.iter().enumerate() {
1447        let identities = deck_card_identities(&decks[index]);
1448        let seat_commander_names = if commander_variant {
1449            commander_names_for_java(&decks[index], commander_names[index].as_deref())
1450        } else {
1451            Vec::new()
1452        };
1453        players.push(PlayerConfig::new(
1454            name.clone(),
1455            &identities,
1456            seat_commander_names,
1457        ));
1458    }
1459    for &idx in &ai_player_indices {
1460        if let Some(player) = players.get_mut(idx) {
1461            player.ai = true;
1462        }
1463    }
1464    let request = StartGameRequest::new(
1465        game_id.clone(),
1466        game_variant,
1467        starting_life,
1468        rand::random(),
1469        players,
1470    );
1471    let session_id = engine.start_game(&request.to_json().map_err(|err| err.to_string())?)?;
1472    info!(game_id, session_id, "hosted java-forge session started");
1473
1474    struct SessionGuard {
1475        engine: ForgeEngine,
1476        session_id: String,
1477        armed: std::cell::Cell<bool>,
1478    }
1479    impl Drop for SessionGuard {
1480        fn drop(&mut self) {
1481            if self.armed.get() {
1482                if let Err(error) = self.engine.abort_game(&self.session_id) {
1483                    warn!(session_id = %self.session_id, %error, "failed to abort java session; context may leak");
1484                }
1485            }
1486        }
1487    }
1488    let guard = SessionGuard {
1489        engine: engine.clone(),
1490        session_id: session_id.clone(),
1491        armed: std::cell::Cell::new(true),
1492    };
1493
1494    let mut remote_response_rxs: HashMap<usize, std_mpsc::Receiver<ClientToServerMessage>> =
1495        remote_response_rxs.into_iter().collect();
1496    let mut last_prompt: Option<AgentPrompt> = None;
1497    let mut pending_roll_acks: usize = 0;
1498    let mut decision_received: Option<Instant> = None;
1499    let mut decision_submitted: Option<Instant> = None;
1500
1501    loop {
1502        if cancel.load(std::sync::atomic::Ordering::Relaxed) {
1503            info!(
1504                session_id,
1505                "hosted java-forge session cancelled; player left the game"
1506            );
1507            return Ok(());
1508        }
1509        for (player_index, rx) in &mut remote_response_rxs {
1510            loop {
1511                match rx.try_recv() {
1512                    Ok(ClientToServerMessage::Response {
1513                        action: PromptOutput::DiceRolled(DiceRolledOutput::DiceRolledAcknowledged),
1514                        ..
1515                    }) => {
1516                        if pending_roll_acks > 0 {
1517                            pending_roll_acks -= 1;
1518                            if pending_roll_acks == 0 {
1519                                let ack = serde_json::to_string(&PromptOutput::DiceRolled(
1520                                    DiceRolledOutput::DiceRolledAcknowledged,
1521                                ))
1522                                .map_err(|err| format!("failed to serialize roll ack: {err}"))?;
1523                                engine.submit_action(&session_id, &ack)?;
1524                            }
1525                        }
1526                    }
1527                    Ok(ClientToServerMessage::Response { prompt_id, action }) => {
1528                        // prompt_id 0 is transport-synthesized: the
1529                        // absent-player default, exempt from validation.
1530                        if prompt_id != 0 {
1531                            let Some(prompt) =
1532                                last_prompt.as_ref().filter(|p| p.prompt_id == prompt_id)
1533                            else {
1534                                reject_response(
1535                                    &remote_prompt_tx,
1536                                    *player_index,
1537                                    last_prompt.as_ref().filter(|p| {
1538                                        self::player_index(&p.deciding_player_id) == *player_index
1539                                    }),
1540                                    ProtocolErrorCode::StalePrompt,
1541                                    format!("response for prompt {prompt_id} is not open"),
1542                                );
1543                                continue;
1544                            };
1545                            if self::player_index(&prompt.deciding_player_id) != *player_index {
1546                                reject_response(
1547                                    &remote_prompt_tx,
1548                                    *player_index,
1549                                    None,
1550                                    ProtocolErrorCode::WrongPlayer,
1551                                    format!(
1552                                        "prompt {prompt_id} is for {}",
1553                                        prompt.deciding_player_id
1554                                    ),
1555                                );
1556                                continue;
1557                            }
1558                            match prompt.input.validate_response(&action) {
1559                                Ok(()) => {}
1560                                Err(ResponseViolation::WrongPromptType) => {
1561                                    reject_response(
1562                                        &remote_prompt_tx,
1563                                        *player_index,
1564                                        Some(prompt),
1565                                        ProtocolErrorCode::WrongPromptType,
1566                                        "response output does not match the prompt type"
1567                                            .to_string(),
1568                                    );
1569                                    continue;
1570                                }
1571                                Err(ResponseViolation::UnknownActionId(id)) => {
1572                                    reject_response(
1573                                        &remote_prompt_tx,
1574                                        *player_index,
1575                                        Some(prompt),
1576                                        ProtocolErrorCode::UnknownActionId,
1577                                        format!(
1578                                            "action id {id:?} was not advertised by the prompt"
1579                                        ),
1580                                    );
1581                                    continue;
1582                                }
1583                                Err(ResponseViolation::CancelNotAllowed) => {
1584                                    reject_response(
1585                                        &remote_prompt_tx,
1586                                        *player_index,
1587                                        Some(prompt),
1588                                        ProtocolErrorCode::CancelNotAllowed,
1589                                        "this prompt is not cancellable".to_string(),
1590                                    );
1591                                    continue;
1592                                }
1593                            }
1594                        }
1595                        decision_received = Some(Instant::now());
1596                        let action_json = serde_json::to_string(&action).map_err(|err| {
1597                            format!(
1598                                "failed to serialize prompt output for player {player_index}: {err}"
1599                            )
1600                        })?;
1601                        debug!(player_index, %action_json, "submitting remote response to java");
1602                        engine.submit_action(&session_id, &action_json)?;
1603                        decision_submitted = Some(Instant::now());
1604                    }
1605                    Ok(ClientToServerMessage::Directive {
1606                        directive: DirectiveInput::Concede,
1607                    }) => {
1608                        // A directive can arrive while another player's prompt
1609                        // is open (this loop drains every seat), so it names
1610                        // its seat.
1611                        let directive_json = directive_concede_json(*player_index);
1612                        debug!(player_index, %directive_json, "submitting concede directive to java");
1613                        engine.submit_action(&session_id, &directive_json)?;
1614                    }
1615                    Err(TryRecvError::Empty) => break,
1616                    Err(TryRecvError::Disconnected) => {
1617                        debug!(player_index, "java-forge response channel disconnected");
1618                        break;
1619                    }
1620                }
1621            }
1622        }
1623
1624        if let Some(prompt_json) = engine.get_prompt(&session_id, 0)? {
1625            let prompt: AgentPrompt = serde_json::from_str(&prompt_json)
1626                .map_err(|err| format!("failed to parse java prompt: {err}"))?;
1627            if last_prompt.as_ref().map(|p| p.prompt_id) != Some(prompt.prompt_id) {
1628                if let Some(started) = decision_submitted.take() {
1629                    crate::metrics::record_forge_decision_stage("next_prompt", started.elapsed());
1630                }
1631                if let Some(started) = decision_received.take() {
1632                    crate::metrics::record_forge_decision_stage(
1633                        "decision_total",
1634                        started.elapsed(),
1635                    );
1636                    report_engine_gc(&engine);
1637                }
1638                last_prompt = Some(prompt.clone());
1639                let player = player_index(&prompt.deciding_player_id);
1640                debug!(player, "forwarding java prompt to remote");
1641                if matches!(prompt.input, PromptInput::DiceRolled(_)) {
1642                    let snapshots_started = Instant::now();
1643                    let prompt_msg = AgentMessage::Prompt(prompt);
1644                    for &agent_index in remote_response_rxs.keys() {
1645                        let state = AgentMessage::State(state_via_handle(
1646                            &engine,
1647                            &session_id,
1648                            Some(agent_index),
1649                        )?);
1650                        let _ = remote_prompt_tx.send((agent_index, state));
1651                        let _ = remote_prompt_tx.send((agent_index, prompt_msg.clone()));
1652                    }
1653                    send_observer_state(&engine, &session_id, &remote_prompt_tx);
1654                    pending_roll_acks = remote_response_rxs.len();
1655                    if pending_roll_acks == 0 {
1656                        let ack = serde_json::to_string(&PromptOutput::DiceRolled(
1657                            DiceRolledOutput::DiceRolledAcknowledged,
1658                        ))
1659                        .map_err(|err| format!("failed to serialize roll ack: {err}"))?;
1660                        engine.submit_action(&session_id, &ack)?;
1661                    }
1662                    crate::metrics::record_forge_decision_stage(
1663                        "snapshots",
1664                        snapshots_started.elapsed(),
1665                    );
1666                } else if Some(player) == local_player_index {
1667                    if let Some(output) = auto_action(&prompt) {
1668                        let action_json = serde_json::to_string(&output)
1669                            .map_err(|err| format!("failed to serialize auto action: {err}"))?;
1670                        engine.submit_action(&session_id, &action_json)?;
1671                    }
1672                } else {
1673                    let snapshots_started = Instant::now();
1674                    for &agent_index in remote_response_rxs.keys() {
1675                        let state = AgentMessage::State(state_via_handle(
1676                            &engine,
1677                            &session_id,
1678                            Some(agent_index),
1679                        )?);
1680                        if remote_prompt_tx.send((agent_index, state)).is_err() {
1681                            return Ok(());
1682                        }
1683                    }
1684                    send_observer_state(&engine, &session_id, &remote_prompt_tx);
1685                    let prompt_msg = AgentMessage::Prompt(prompt);
1686                    if remote_prompt_tx.send((player, prompt_msg)).is_err() {
1687                        return Ok(());
1688                    }
1689                    crate::metrics::record_forge_decision_stage(
1690                        "snapshots",
1691                        snapshots_started.elapsed(),
1692                    );
1693                }
1694            }
1695        }
1696
1697        if engine.is_game_over(&session_id)? {
1698            info!("hosted java-forge session reached game over");
1699            let mut final_messages = Vec::new();
1700            for &agent_index in remote_response_rxs.keys() {
1701                match state_via_handle(&engine, &session_id, Some(agent_index)) {
1702                    Ok(state_update) => {
1703                        final_messages.push((agent_index, AgentMessage::State(state_update)));
1704                    }
1705                    Err(error) => {
1706                        warn!(%error, agent_index, "game over: final snapshot unavailable; sending game-over prompt only");
1707                    }
1708                }
1709            }
1710            if let Ok(state_update) = state_via_handle(&engine, &session_id, None) {
1711                final_messages.push((
1712                    crate::host::OBSERVER_SEAT,
1713                    AgentMessage::State(state_update),
1714                ));
1715            }
1716            let game_over = AgentMessage::Prompt(game_over_prompt());
1717            for &agent_index in remote_response_rxs.keys() {
1718                final_messages.push((agent_index, game_over.clone()));
1719            }
1720            let _ = game_over_tx.send(HostedGameOver {
1721                game_id: game_id.clone(),
1722                messages: final_messages,
1723            });
1724            engine.end_game(&session_id)?;
1725            guard.armed.set(false);
1726            return Ok(());
1727        }
1728
1729        std::thread::sleep(Duration::from_millis(50));
1730    }
1731}
1732
1733#[cfg(feature = "java-forge")]
1734fn wait_for_prompt<B: JavaBridge>(
1735    session: &mut JavaForgeSession<B>,
1736    max_polls: usize,
1737) -> Result<Option<String>, String> {
1738    for _ in 0..max_polls {
1739        if let Some(prompt) = session.get_prompt(0)? {
1740            return Ok(Some(prompt));
1741        }
1742        std::thread::sleep(Duration::from_millis(50));
1743    }
1744    Ok(None)
1745}
1746
1747#[cfg(forge_backend)]
1748fn send_observer_state(
1749    engine: &ForgeEngine,
1750    session_id: &str,
1751    remote_prompt_tx: &std_mpsc::Sender<(usize, AgentMessage)>,
1752) {
1753    match state_via_handle(engine, session_id, None) {
1754        Ok(state_update) => {
1755            let _ = remote_prompt_tx.send((
1756                crate::host::OBSERVER_SEAT,
1757                AgentMessage::State(state_update),
1758            ));
1759        }
1760        Err(error) => warn!(%error, "observer snapshot unavailable"),
1761    }
1762}
1763
1764#[cfg(forge_backend)]
1765fn player_index(deciding_player_id: &str) -> usize {
1766    deciding_player_id
1767        .strip_prefix("player-")
1768        .and_then(|n| n.parse().ok())
1769        .unwrap_or(0)
1770}
1771
1772#[cfg(forge_backend)]
1773fn reject_response(
1774    remote_prompt_tx: &std_mpsc::Sender<(usize, AgentMessage)>,
1775    seat: usize,
1776    reopen_prompt: Option<&AgentPrompt>,
1777    code: ProtocolErrorCode,
1778    message: String,
1779) {
1780    let _ = remote_prompt_tx.send((
1781        seat,
1782        AgentMessage::Error(ProtocolError {
1783            code,
1784            message,
1785            prompt_id: reopen_prompt.map(|p| p.prompt_id),
1786        }),
1787    ));
1788    if let Some(prompt) = reopen_prompt {
1789        let _ = remote_prompt_tx.send((seat, AgentMessage::Prompt(prompt.clone())));
1790    }
1791}
1792
1793#[cfg(forge_backend)]
1794fn auto_action(prompt: &AgentPrompt) -> Option<PromptOutput> {
1795    match prompt.input {
1796        PromptInput::ChooseAction(_) => {
1797            Some(PromptOutput::ChooseAction(ChooseActionOutput::Pass {
1798                until: None,
1799                exhaust_stack: false,
1800            }))
1801        }
1802        _ => None,
1803    }
1804}
1805
1806#[cfg(forge_backend)]
1807fn game_over_prompt() -> AgentPrompt {
1808    AgentPrompt {
1809        prompt_id: u32::MAX,
1810        deciding_player_id: "player-0".to_string(),
1811        source_card: None,
1812        input: PromptInput::GameOver(GameOverInput {}),
1813    }
1814}
1815
1816#[cfg(forge_backend)]
1817fn state_via_handle(
1818    engine: &ForgeEngine,
1819    session_id: &str,
1820    viewer: Option<usize>,
1821) -> Result<StateUpdate, String> {
1822    let game_view: GameViewDto = serde_json::from_str(&engine.get_snapshot(session_id, viewer)?)
1823        .map_err(|err| format!("failed to parse java snapshot: {err}"))?;
1824    Ok(StateUpdate { game_view })
1825}
1826
1827#[cfg(forge_backend)]
1828fn deck_card_identities(deck: &Deck) -> Vec<DeckCardIdentity> {
1829    deck.cards
1830        .iter()
1831        .chain(deck.commanders.iter().flatten())
1832        .map(|card| card.identity.clone())
1833        .collect()
1834}
1835
1836#[cfg(forge_backend)]
1837fn commander_names_for_java(deck: &Deck, fallback: Option<&str>) -> Vec<String> {
1838    let names: Vec<String> = deck
1839        .commanders
1840        .iter()
1841        .flatten()
1842        .map(|card| java_card_name(&card.identity.name))
1843        .collect();
1844    if !names.is_empty() {
1845        return names;
1846    }
1847    fallback
1848        .filter(|name| !name.is_empty())
1849        .map(|name| vec![java_card_name(name)])
1850        .unwrap_or_default()
1851}
1852
1853#[cfg(forge_backend)]
1854fn smoke_deck(land_name: &str, spell_name: &str) -> Vec<DeckCardIdentity> {
1855    (0..24)
1856        .map(|_| DeckCardIdentity {
1857            name: land_name.to_string(),
1858            ..Default::default()
1859        })
1860        .chain((0..36).map(|_| DeckCardIdentity {
1861            name: spell_name.to_string(),
1862            ..Default::default()
1863        }))
1864        .collect()
1865}
1866
1867#[cfg(feature = "java-forge")]
1868fn scenario_deck(land_name: &str) -> Vec<DeckCardIdentity> {
1869    (0..60)
1870        .map(|_| DeckCardIdentity {
1871            name: land_name.to_string(),
1872            ..Default::default()
1873        })
1874        .collect()
1875}
1876
1877#[cfg(feature = "java-forge")]
1878enum JavaScenario {
1879    KeepAndPlayLand {
1880        played_land: bool,
1881    },
1882    MulliganOncePlayLand {
1883        mulliganed: bool,
1884        kept_second_hand: bool,
1885        put_back_done: bool,
1886        played_land: bool,
1887    },
1888}
1889
1890#[cfg(feature = "java-forge")]
1891impl JavaScenario {
1892    fn from_name(name: &str) -> Result<Self, String> {
1893        match name {
1894            "keep-and-play-land" => Ok(Self::KeepAndPlayLand { played_land: false }),
1895            "mulligan-once-play-land" => Ok(Self::MulliganOncePlayLand {
1896                mulliganed: false,
1897                kept_second_hand: false,
1898                put_back_done: false,
1899                played_land: false,
1900            }),
1901            _ => Err(format!(
1902                "unknown java-forge scenario '{name}'. Supported scenarios: keep-and-play-land, mulligan-once-play-land"
1903            )),
1904        }
1905    }
1906
1907    fn name(&self) -> &'static str {
1908        match self {
1909            Self::KeepAndPlayLand { .. } => "keep-and-play-land",
1910            Self::MulliganOncePlayLand { .. } => "mulligan-once-play-land",
1911        }
1912    }
1913
1914    fn next_action(
1915        &mut self,
1916        prompt: &Value,
1917        game_view: &GameViewDto,
1918    ) -> Result<Option<PromptOutput>, String> {
1919        match self {
1920            Self::KeepAndPlayLand { played_land } => {
1921                if *played_land && battlefield_contains(game_view, "Swamp") {
1922                    return Ok(None);
1923                }
1924                match prompt_type(prompt) {
1925                    Some("mulligan") => Ok(Some(PromptOutput::Mulligan(
1926                        MulliganOutput::MulliganDecision { keep: true },
1927                    ))),
1928                    Some("chooseAction") => {
1929                        if let Some(action) = play_first_card_action(prompt, "Swamp")? {
1930                            *played_land = true;
1931                            Ok(Some(action))
1932                        } else {
1933                            Ok(Some(PromptOutput::ChooseAction(ChooseActionOutput::Pass {
1934                                until: None,
1935                                exhaust_stack: false,
1936                            })))
1937                        }
1938                    }
1939                    other => Err(format!(
1940                        "scenario '{}' expected mulligan or chooseAction, got {:?}",
1941                        self.name(),
1942                        other
1943                    )),
1944                }
1945            }
1946            Self::MulliganOncePlayLand {
1947                mulliganed,
1948                kept_second_hand,
1949                put_back_done,
1950                played_land,
1951            } => {
1952                if *played_land && battlefield_contains(game_view, "Swamp") {
1953                    return Ok(None);
1954                }
1955                match prompt_type(prompt) {
1956                    Some("mulligan") if !*mulliganed => {
1957                        *mulliganed = true;
1958                        Ok(Some(PromptOutput::Mulligan(MulliganOutput::MulliganDecision { keep: false })))
1959                    }
1960                    Some("mulligan") if !*kept_second_hand => {
1961                        *kept_second_hand = true;
1962                        Ok(Some(PromptOutput::Mulligan(MulliganOutput::MulliganDecision { keep: true })))
1963                    }
1964                    Some("mulliganPutBack") if !*put_back_done => {
1965                        let count = prompt
1966                            .get("input")
1967                            .and_then(|input| input.get("count"))
1968                            .and_then(Value::as_u64)
1969                            .unwrap_or(1) as usize;
1970                        let card_ids = prompt_card_ids(prompt, "handCardIds", count)?;
1971                        *put_back_done = true;
1972                        Ok(Some(PromptOutput::MulliganPutBack(MulliganPutBackOutput::MulliganPutBackDecision { card_ids })))
1973                    }
1974                    Some("chooseAction") => {
1975                        if let Some(action) = play_first_card_action(prompt, "Swamp")? {
1976                            *played_land = true;
1977                            Ok(Some(action))
1978                        } else {
1979                            Ok(Some(PromptOutput::ChooseAction(ChooseActionOutput::Pass { until: None, exhaust_stack: false })))
1980                        }
1981                    }
1982                    other => Err(format!(
1983                        "scenario '{}' expected mulligan, mulliganPutBack, or chooseAction, got {:?}",
1984                        self.name(),
1985                        other
1986                    )),
1987                }
1988            }
1989        }
1990    }
1991}
1992
1993#[cfg(feature = "java-forge")]
1994fn run_scenario_loop<B: JavaBridge>(
1995    session: &mut JavaForgeSession<B>,
1996    mut scenario: JavaScenario,
1997    max_prompts: usize,
1998) -> Result<(), String> {
1999    let mut prompts_seen = 0usize;
2000    let mut last_prompt_json: Option<String> = None;
2001    while prompts_seen < max_prompts {
2002        let Some(prompt_json) = wait_for_prompt(session, 600)? else {
2003            return Err(format!(
2004                "timed out waiting for java-forge scenario '{}' prompt",
2005                scenario.name()
2006            ));
2007        };
2008        if last_prompt_json.as_deref() == Some(prompt_json.as_str()) {
2009            std::thread::sleep(Duration::from_millis(50));
2010            continue;
2011        }
2012        last_prompt_json = Some(prompt_json.clone());
2013        prompts_seen += 1;
2014
2015        let prompt: AgentPrompt = serde_json::from_str(&prompt_json)
2016            .map_err(|err| format!("failed to parse java scenario prompt: {err}"))?;
2017        let player = player_index(&prompt.deciding_player_id);
2018        if player != 0 {
2019            if let Some(output) = auto_action(&prompt) {
2020                session.submit_action(
2021                    &serde_json::to_string(&output).map_err(|err| err.to_string())?,
2022                )?;
2023            }
2024            continue;
2025        }
2026
2027        let game_view: GameViewDto = serde_json::from_str(&session.get_snapshot(Some(0))?)
2028            .map_err(|err| format!("failed to parse java scenario snapshot: {err}"))?;
2029        let normalized_prompt = serde_json::to_value(&prompt).map_err(|err| err.to_string())?;
2030        info!(
2031            scenario = scenario.name(),
2032            prompts_seen,
2033            prompt_type = prompt_type(&normalized_prompt).unwrap_or("<missing>"),
2034            "java-forge scenario prompt"
2035        );
2036        let Some(action) = scenario.next_action(&normalized_prompt, &game_view)? else {
2037            info!(
2038                scenario = scenario.name(),
2039                prompts_seen, "java-forge scenario assertions satisfied"
2040            );
2041            return Ok(());
2042        };
2043        submit_player_action(session, &action)?;
2044    }
2045    Err(format!(
2046        "java-forge scenario '{}' did not complete within {max_prompts} prompts",
2047        scenario.name()
2048    ))
2049}
2050
2051#[cfg(feature = "java-forge")]
2052pub fn run_concede_smoke() -> Result<(), String> {
2053    let config = JavaRuntimeConfig::from_env();
2054    let assets_dir = config.assets_dir.to_string_lossy().to_string();
2055    let bridge = SubprocessBridge::spawn(&config)?;
2056    let mut session = JavaForgeSession::new(bridge);
2057    session.initialize(&assets_dir)?;
2058    run_concede_game(&mut session, 3, 2, 12)?;
2059    run_concede_game(&mut session, 2, 1, 8)?;
2060    Ok(())
2061}
2062
2063#[cfg(not(feature = "java-forge"))]
2064pub fn run_concede_smoke() -> Result<(), String> {
2065    Err(unsupported_message().to_string())
2066}
2067
2068#[cfg(forge_backend)]
2069fn directive_concede_json(player: usize) -> String {
2070    format!(r#"{{"type":"directive","directive":{{"type":"concede"}},"player":{player}}}"#)
2071}
2072
2073#[cfg(feature = "java-forge")]
2074fn run_concede_game<B: JavaBridge>(
2075    session: &mut JavaForgeSession<B>,
2076    seats: usize,
2077    conceder: usize,
2078    concede_after: usize,
2079) -> Result<(), String> {
2080    const POST_CONCEDE_DECISIONS: usize = 12;
2081    const STALL_REPEATS: usize = 300;
2082
2083    let cards: Vec<manabrew_protocol::deck_dto::DeckCard> = (0..60)
2084        .map(|_| manabrew_protocol::deck_dto::DeckCard {
2085            identity: DeckCardIdentity {
2086                name: "Mountain".to_string(),
2087                set_code: "M20".to_string(),
2088                ..Default::default()
2089            },
2090            ..Default::default()
2091        })
2092        .collect();
2093    let deck = Deck {
2094        name: "concede-smoke".to_string(),
2095        cards,
2096        ..Default::default()
2097    };
2098    let identities = deck_card_identities(&deck);
2099    let players: Vec<PlayerConfig> = (0..seats)
2100        .map(|i| PlayerConfig::new(format!("Concede {}", i + 1), &identities, Vec::new()))
2101        .collect();
2102    let request = StartGameRequest::new(
2103        format!("concede-smoke-{seats}p"),
2104        String::new(),
2105        20,
2106        7,
2107        players,
2108    );
2109    let session_id = session.start_game(&request)?;
2110    info!(
2111        session_id,
2112        seats, conceder, concede_after, "concede smoke game started"
2113    );
2114
2115    let mut bots: HashMap<usize, SimpleAi> = HashMap::new();
2116    let mut last_prompt_json: Option<String> = None;
2117    let mut acted = 0usize;
2118    let mut acted_after_concede = 0usize;
2119    let mut conceded = false;
2120    let mut repeat_count = 0usize;
2121
2122    for _ in 0..40_000 {
2123        if session.is_game_over()? {
2124            if !conceded {
2125                return Err("concede smoke: game ended before the concede fired".to_string());
2126            }
2127            break;
2128        }
2129        let Some(prompt_json) = session.get_prompt(0)? else {
2130            std::thread::sleep(Duration::from_millis(10));
2131            continue;
2132        };
2133        if last_prompt_json.as_deref() == Some(prompt_json.as_str()) {
2134            repeat_count += 1;
2135            if repeat_count > STALL_REPEATS {
2136                return Err(format!(
2137                    "concede smoke stalled on the same prompt (seats={seats} acted={acted} conceded={conceded} after={acted_after_concede}): {prompt_json}"
2138                ));
2139            }
2140            std::thread::sleep(Duration::from_millis(10));
2141            continue;
2142        }
2143        repeat_count = 0;
2144
2145        let prompt: AgentPrompt = serde_json::from_str(&prompt_json)
2146            .map_err(|err| format!("concede smoke: bad prompt: {err}"))?;
2147        let player = player_index(&prompt.deciding_player_id);
2148
2149        if !conceded && acted >= concede_after {
2150            info!(conceder, acted, "concede smoke: injecting concede");
2151            session.submit_action(&directive_concede_json(conceder))?;
2152            conceded = true;
2153            last_prompt_json = None;
2154            continue;
2155        }
2156        if conceded && player == conceder {
2157            session.submit_action(&directive_concede_json(conceder))?;
2158            last_prompt_json = Some(prompt_json);
2159            continue;
2160        }
2161
2162        if let Some(action) = bots.entry(player).or_default().decide(prompt) {
2163            submit_player_action(session, &action)?;
2164            acted += 1;
2165            if conceded {
2166                acted_after_concede += 1;
2167            }
2168        }
2169        last_prompt_json = Some(prompt_json);
2170        if conceded && seats > 2 && acted_after_concede >= POST_CONCEDE_DECISIONS {
2171            break;
2172        }
2173    }
2174
2175    if !conceded {
2176        return Err("concede smoke: never reached the injection point".to_string());
2177    }
2178    if seats == 2 {
2179        if !session.is_game_over()? {
2180            return Err("concede smoke: 2p concession did not end the game".to_string());
2181        }
2182        info!(acted, "concede smoke: 2p concession ended the game");
2183    } else {
2184        if acted_after_concede < POST_CONCEDE_DECISIONS && !session.is_game_over()? {
2185            return Err(format!(
2186                "concede smoke: game did not progress after the concession (only {acted_after_concede} decisions)"
2187            ));
2188        }
2189        let snapshot = parse_snapshot(session)?;
2190        let status = snapshot
2191            .pointer(&format!("/players/{conceder}/status"))
2192            .and_then(Value::as_str)
2193            .unwrap_or("");
2194        if status != "conceded" {
2195            return Err(format!(
2196                "concede smoke: seat {conceder} has status '{status}', expected 'conceded'"
2197            ));
2198        }
2199        info!(
2200            acted,
2201            acted_after_concede, "concede smoke: game continued past the concession"
2202        );
2203    }
2204    session.end_game()?;
2205    Ok(())
2206}
2207
2208#[cfg(feature = "java-forge")]
2209fn run_self_play_loop<B: JavaBridge>(
2210    session: &mut JavaForgeSession<B>,
2211    max_prompts: usize,
2212) -> Result<(), String> {
2213    const STALL_REPEATS: usize = 100;
2214
2215    let mut bots: HashMap<usize, SimpleAi> = HashMap::new();
2216    let mut last_prompt_json: Option<String> = None;
2217    let mut acted = 0usize;
2218    let mut repeat_count = 0usize;
2219    let mut seen_prompt = false;
2220    let max_iterations = max_prompts.saturating_mul(200).max(2_000);
2221
2222    for _ in 0..max_iterations {
2223        if let Some(prompt_json) = session.get_prompt(0)? {
2224            seen_prompt = true;
2225            if last_prompt_json.as_deref() == Some(prompt_json.as_str()) {
2226                if session.is_game_over()? {
2227                    info!(acted, "java-forge self-play reached game over");
2228                    return Ok(());
2229                }
2230                repeat_count += 1;
2231                if repeat_count > STALL_REPEATS {
2232                    let raw_value: Value =
2233                        serde_json::from_str(&prompt_json).unwrap_or(Value::Null);
2234                    dump_stuck(
2235                        "java re-emitted the same prompt after the bot acted (stall)",
2236                        &raw_value,
2237                        None,
2238                        session,
2239                    );
2240                    return Err(
2241                        "self-play stalled: java re-emitted the same prompt after the bot's action"
2242                            .to_string(),
2243                    );
2244                }
2245                std::thread::sleep(Duration::from_millis(20));
2246                continue;
2247            }
2248            repeat_count = 0;
2249
2250            let prompt: AgentPrompt = serde_json::from_str(&prompt_json)
2251                .map_err(|err| format!("failed to parse java self-play prompt: {err}"))?;
2252            let player = player_index(&prompt.deciding_player_id);
2253            let raw_value: Value = serde_json::from_str(&prompt_json).unwrap_or(Value::Null);
2254            let normalized = raw_value.clone();
2255
2256            match bots.entry(player).or_default().decide(prompt) {
2257                Some(action) => {
2258                    if let Err(err) = submit_player_action(session, &action) {
2259                        dump_stuck(
2260                            "java rejected the bot action",
2261                            &raw_value,
2262                            Some(&normalized),
2263                            session,
2264                        );
2265                        return Err(format!(
2266                            "self-play: java rejected action for player {player}: {err}"
2267                        ));
2268                    }
2269                    acted += 1;
2270                    if acted >= max_prompts {
2271                        dump_stuck(
2272                            "did not reach game over within max prompts",
2273                            &raw_value,
2274                            Some(&normalized),
2275                            session,
2276                        );
2277                        return Err(format!(
2278                            "self-play did not reach game over within {max_prompts} decisions"
2279                        ));
2280                    }
2281                }
2282                None => debug!(
2283                    player,
2284                    prompt_type = prompt_type(&normalized).unwrap_or("<missing>"),
2285                    "self-play: no action for prompt (display-only)"
2286                ),
2287            }
2288            last_prompt_json = Some(prompt_json);
2289            continue;
2290        }
2291
2292        if seen_prompt && session.is_game_over()? {
2293            info!(acted, "java-forge self-play reached game over");
2294            return Ok(());
2295        }
2296        std::thread::sleep(Duration::from_millis(20));
2297    }
2298
2299    dump_stuck(
2300        "self-play exceeded its iteration cap without game over",
2301        &Value::Null,
2302        None,
2303        session,
2304    );
2305    Err("self-play exceeded its iteration cap without reaching game over".to_string())
2306}
2307
2308#[cfg(feature = "java-forge")]
2309fn parse_snapshot<B: JavaBridge>(session: &mut JavaForgeSession<B>) -> Result<Value, String> {
2310    let snapshot_json = session.get_snapshot(Some(0))?;
2311    serde_json::from_str(&snapshot_json)
2312        .map_err(|err| format!("failed to parse java self-play snapshot: {err}"))
2313}
2314
2315#[cfg(feature = "java-forge")]
2316fn dump_stuck<B: JavaBridge>(
2317    reason: &str,
2318    prompt: &Value,
2319    normalized: Option<&Value>,
2320    session: &mut JavaForgeSession<B>,
2321) {
2322    let snapshot = parse_snapshot(session).unwrap_or(Value::Null);
2323    let artifact = json!({
2324        "reason": reason,
2325        "rawPrompt": prompt,
2326        "normalizedPrompt": normalized,
2327        "snapshot": snapshot,
2328    });
2329    let ts = std::time::SystemTime::now()
2330        .duration_since(std::time::UNIX_EPOCH)
2331        .map(|d| d.as_millis())
2332        .unwrap_or(0);
2333    let path = workspace_root().join(format!("target/self-play-stuck-{ts}.json"));
2334    match serde_json::to_string_pretty(&artifact) {
2335        Ok(body) => {
2336            if let Err(error) = std::fs::write(&path, body) {
2337                warn!(%error, reason, "self-play stuck; failed to write artifact");
2338            } else {
2339                warn!(path = %path.display(), reason, "self-play stuck; wrote artifact");
2340            }
2341        }
2342        Err(error) => warn!(%error, reason, "self-play stuck; failed to serialize artifact"),
2343    }
2344}
2345
2346#[cfg(feature = "java-forge")]
2347fn submit_player_action<B: JavaBridge>(
2348    session: &mut JavaForgeSession<B>,
2349    action: &PromptOutput,
2350) -> Result<(), String> {
2351    let action_json = serde_json::to_string(action)
2352        .map_err(|err| format!("failed to serialize scenario action: {err}"))?;
2353    session.submit_action(&action_json)?;
2354    Ok(())
2355}
2356
2357#[cfg(feature = "java-forge")]
2358fn prompt_type(prompt: &Value) -> Option<&str> {
2359    prompt
2360        .get("input")
2361        .and_then(|input| input.get("type"))
2362        .and_then(Value::as_str)
2363}
2364
2365#[cfg(feature = "java-forge")]
2366fn play_first_card_action(prompt: &Value, card_name: &str) -> Result<Option<PromptOutput>, String> {
2367    let Some(action) = prompt
2368        .get("input")
2369        .and_then(|input| input.get("actions"))
2370        .and_then(Value::as_array)
2371        .and_then(|actions| {
2372            actions.iter().find(|action| {
2373                action
2374                    .get("modeLabel")
2375                    .and_then(Value::as_str)
2376                    .is_some_and(|label| label.contains(card_name))
2377            })
2378        })
2379    else {
2380        return Ok(None);
2381    };
2382    let action_id = action
2383        .get("id")
2384        .and_then(Value::as_str)
2385        .ok_or_else(|| format!("playable action for '{card_name}' is missing id"))?;
2386    Ok(Some(PromptOutput::ChooseAction(ChooseActionOutput::Act {
2387        action_id: action_id.to_string(),
2388    })))
2389}
2390
2391#[cfg(feature = "java-forge")]
2392fn prompt_card_ids(prompt: &Value, field: &str, count: usize) -> Result<Vec<String>, String> {
2393    let card_ids = prompt
2394        .get("input")
2395        .and_then(|input| input.get(field))
2396        .and_then(Value::as_array)
2397        .ok_or_else(|| format!("prompt is missing {field}"))?;
2398    if card_ids.len() < count {
2399        return Err(format!(
2400            "prompt field {field} has {} cards, need {count}",
2401            card_ids.len()
2402        ));
2403    }
2404    Ok(card_ids
2405        .iter()
2406        .take(count)
2407        .filter_map(Value::as_str)
2408        .map(str::to_string)
2409        .collect())
2410}
2411
2412#[cfg(feature = "java-forge")]
2413fn battlefield_contains(game_view: &GameViewDto, card_name: &str) -> bool {
2414    use manabrew_agent_interface::game_view_dto::{CardView, ZoneKind};
2415    game_view.zones.iter().any(|zone| {
2416        zone.zone == ZoneKind::Battlefield
2417            && zone.owner_id == "player-0"
2418            && zone.cards.iter().any(|card| match card {
2419                CardView::Visible(dto) => dto.identity.name == card_name,
2420                CardView::Hidden { .. } => false,
2421            })
2422    })
2423}
2424
2425pub trait JavaBridge {
2426    fn initialize(&mut self, assets_dir: &str) -> Result<(), String>;
2427    fn start_game_json(&mut self, request_json: &str) -> Result<String, String>;
2428    fn submit_action(&mut self, session_id: &str, action_json: &str) -> Result<String, String>;
2429    fn get_prompt(
2430        &mut self,
2431        session_id: &str,
2432        player_index: usize,
2433    ) -> Result<Option<String>, String>;
2434    fn get_snapshot(&mut self, session_id: &str, viewer: Option<usize>) -> Result<String, String>;
2435    fn is_game_over(&mut self, session_id: &str) -> Result<bool, String>;
2436    fn end_game(&mut self, session_id: &str) -> Result<(), String>;
2437    fn abort_game(&mut self, session_id: &str) -> Result<(), String>;
2438}
2439
2440pub struct JavaForgeSession<B> {
2441    bridge: B,
2442    session_id: Option<String>,
2443}
2444
2445impl<B: JavaBridge> JavaForgeSession<B> {
2446    pub fn new(bridge: B) -> Self {
2447        Self {
2448            bridge,
2449            session_id: None,
2450        }
2451    }
2452
2453    pub fn initialize(&mut self, assets_dir: &str) -> Result<(), String> {
2454        self.bridge.initialize(assets_dir)
2455    }
2456
2457    pub fn start_game(&mut self, request: &StartGameRequest) -> Result<String, String> {
2458        let request_json = request.to_json().map_err(|err| err.to_string())?;
2459        let response_json = self.bridge.start_game_json(&request_json)?;
2460        let response: StartGameResponse =
2461            serde_json::from_str(&response_json).map_err(|err| err.to_string())?;
2462        self.session_id = Some(response.session_id.clone());
2463        Ok(response.session_id)
2464    }
2465
2466    pub fn submit_action(&mut self, action_json: &str) -> Result<String, String> {
2467        let session_id = self.require_session_id()?.to_string();
2468        self.bridge.submit_action(&session_id, action_json)
2469    }
2470
2471    pub fn get_prompt(&mut self, player_index: usize) -> Result<Option<String>, String> {
2472        let session_id = self.require_session_id()?.to_string();
2473        self.bridge.get_prompt(&session_id, player_index)
2474    }
2475
2476    pub fn get_snapshot(&mut self, viewer: Option<usize>) -> Result<String, String> {
2477        let session_id = self.require_session_id()?.to_string();
2478        self.bridge.get_snapshot(&session_id, viewer)
2479    }
2480
2481    pub fn is_game_over(&mut self) -> Result<bool, String> {
2482        let session_id = self.require_session_id()?.to_string();
2483        self.bridge.is_game_over(&session_id)
2484    }
2485
2486    pub fn end_game(&mut self) -> Result<(), String> {
2487        let Some(session_id) = self.session_id.take() else {
2488            return Ok(());
2489        };
2490        self.bridge.end_game(&session_id)
2491    }
2492
2493    fn require_session_id(&self) -> Result<&str, String> {
2494        self.session_id
2495            .as_deref()
2496            .ok_or_else(|| "java-forge session has not started".to_string())
2497    }
2498}
2499
2500pub struct UnavailableJavaBridge;
2501
2502impl JavaBridge for UnavailableJavaBridge {
2503    fn initialize(&mut self, _assets_dir: &str) -> Result<(), String> {
2504        Err(unsupported_message().to_string())
2505    }
2506
2507    fn start_game_json(&mut self, _request_json: &str) -> Result<String, String> {
2508        Err(unsupported_message().to_string())
2509    }
2510
2511    fn submit_action(&mut self, _session_id: &str, _action_json: &str) -> Result<String, String> {
2512        Err(unsupported_message().to_string())
2513    }
2514
2515    fn get_prompt(
2516        &mut self,
2517        _session_id: &str,
2518        _player_index: usize,
2519    ) -> Result<Option<String>, String> {
2520        Err(unsupported_message().to_string())
2521    }
2522
2523    fn get_snapshot(
2524        &mut self,
2525        _session_id: &str,
2526        _viewer: Option<usize>,
2527    ) -> Result<String, String> {
2528        Err(unsupported_message().to_string())
2529    }
2530
2531    fn is_game_over(&mut self, _session_id: &str) -> Result<bool, String> {
2532        Err(unsupported_message().to_string())
2533    }
2534
2535    fn end_game(&mut self, _session_id: &str) -> Result<(), String> {
2536        Err(unsupported_message().to_string())
2537    }
2538
2539    fn abort_game(&mut self, _session_id: &str) -> Result<(), String> {
2540        Err(unsupported_message().to_string())
2541    }
2542}
2543
2544#[cfg(feature = "java-forge")]
2545#[derive(serde::Deserialize)]
2546struct SubprocessReply {
2547    ok: bool,
2548    #[serde(default)]
2549    result: String,
2550    #[serde(default)]
2551    error: Option<String>,
2552}
2553
2554#[cfg(feature = "java-forge")]
2555const CALL_TIMEOUT: Duration = Duration::from_secs(60);
2556#[cfg(feature = "java-forge")]
2557const SHUTDOWN_GRACE: Duration = Duration::from_secs(5);
2558
2559#[cfg(feature = "java-forge")]
2560pub struct SubprocessBridge {
2561    child: Child,
2562    stdin: BufWriter<ChildStdin>,
2563    stdout_rx: std_mpsc::Receiver<String>,
2564    stdout_handle: Option<std::thread::JoinHandle<()>>,
2565    stderr_handle: Option<std::thread::JoinHandle<()>>,
2566}
2567
2568#[cfg(feature = "java-forge")]
2569impl SubprocessBridge {
2570    fn spawn(config: &JavaRuntimeConfig) -> Result<Self, String> {
2571        config.validate()?;
2572
2573        let java_bin = resolve_java_bin(config);
2574        let jvm_args = config.jvm_args();
2575        info!(target: "self_hosted_node::java", args = %jvm_args.join(" "), "spawning java engine");
2576        let mut cmd = Command::new(&java_bin);
2577        // The JVM applies JAVA_TOOL_OPTIONS before anything on the command line,
2578        // so whatever is set on the host silently governs every flag we do not
2579        // name. Production ran -Xmx512m -XX:+UseSerialGC that way for months
2580        // with nothing in the repo to show it. Use SELF_HOSTED_NODE_JAVA_OPTS
2581        // instead: it is passed here, logged above, and lives in config.
2582        cmd.env_remove("JAVA_TOOL_OPTIONS");
2583        cmd.args(&jvm_args);
2584        cmd.arg("-jar").arg(&config.harness_jar);
2585        cmd.arg("--interactive-server");
2586        cmd.arg("--forge-home")
2587            .arg(format!("{}/", config.assets_dir.display()));
2588
2589        let mut child = cmd
2590            .stdin(Stdio::piped())
2591            .stdout(Stdio::piped())
2592            .stderr(Stdio::piped())
2593            .spawn()
2594            .map_err(|err| format!("failed to spawn java subprocess: {err}"))?;
2595
2596        let stdin = child
2597            .stdin
2598            .take()
2599            .ok_or_else(|| "java subprocess has no stdin".to_string())?;
2600        let stdout = child
2601            .stdout
2602            .take()
2603            .ok_or_else(|| "java subprocess has no stdout".to_string())?;
2604        let stderr = child.stderr.take();
2605
2606        // Bounded so a chatty Java side can't grow the queue without bound.
2607        // Cap is generous — protocol replies are one line per request.
2608        let (stdout_tx, stdout_rx) = std_mpsc::sync_channel::<String>(1024);
2609        let stdout_handle = std::thread::spawn(move || {
2610            let reader = BufReader::new(stdout);
2611            for line in reader.lines().map_while(Result::ok) {
2612                if stdout_tx.send(line).is_err() {
2613                    break;
2614                }
2615            }
2616        });
2617
2618        let stderr_handle = std::thread::spawn(move || {
2619            if let Some(stderr) = stderr {
2620                let reader = BufReader::new(stderr);
2621                for line in reader.lines().map_while(Result::ok) {
2622                    if let Some(pause) = parse_gc_pause(&line) {
2623                        crate::metrics::record_jvm_gc(
2624                            pause.kind,
2625                            Duration::from_secs_f64(pause.millis / 1000.0),
2626                            pause.heap_after_mb,
2627                        );
2628                    }
2629                    if line.contains("Exception") || line.contains("ERROR") {
2630                        warn!(target: "self_hosted_node::java", "[java] {line}");
2631                    } else if line.contains("][gc") {
2632                        // -Xlog:gc*:stderr is opt-in, and debug would drop it before Loki.
2633                        info!(target: "self_hosted_node::java", "[java] {line}");
2634                    } else {
2635                        debug!(target: "self_hosted_node::java", "[java] {line}");
2636                    }
2637                }
2638            }
2639        });
2640
2641        Ok(Self {
2642            child,
2643            stdin: BufWriter::new(stdin),
2644            stdout_rx,
2645            stdout_handle: Some(stdout_handle),
2646            stderr_handle: Some(stderr_handle),
2647        })
2648    }
2649
2650    fn call(&mut self, request_json: &str) -> Result<String, String> {
2651        // Drain anything still queued from a prior request — a previous call()
2652        // that timed out may have left its reply in the channel, and consuming
2653        // it now would shift every subsequent call off-by-one.
2654        loop {
2655            match self.stdout_rx.try_recv() {
2656                Ok(stale) => {
2657                    debug!(target: "self_hosted_node::java", line = %stale, "discarding stale stdout line");
2658                }
2659                Err(TryRecvError::Empty) => break,
2660                Err(TryRecvError::Disconnected) => {
2661                    return Err("java subprocess closed stdout (crashed?)".to_string());
2662                }
2663            }
2664        }
2665
2666        self.stdin
2667            .write_all(request_json.as_bytes())
2668            .map_err(|err| format!("failed to write subprocess stdin: {err}"))?;
2669        self.stdin
2670            .write_all(b"\n")
2671            .map_err(|err| format!("failed to write subprocess newline: {err}"))?;
2672        self.stdin
2673            .flush()
2674            .map_err(|err| format!("failed to flush subprocess stdin: {err}"))?;
2675
2676        let deadline = Instant::now() + CALL_TIMEOUT;
2677        loop {
2678            let remaining = deadline.saturating_duration_since(Instant::now());
2679            if remaining.is_zero() {
2680                return Err(format!(
2681                    "java subprocess timed out after {}s",
2682                    CALL_TIMEOUT.as_secs()
2683                ));
2684            }
2685            match self.stdout_rx.recv_timeout(remaining) {
2686                Ok(line) => {
2687                    let trimmed = line.trim();
2688                    if trimmed.is_empty() {
2689                        continue;
2690                    }
2691                    match serde_json::from_str::<SubprocessReply>(trimmed) {
2692                        Ok(reply) if reply.ok => return Ok(reply.result),
2693                        Ok(reply) => {
2694                            return Err(reply.error.unwrap_or_else(|| "unknown java error".into()));
2695                        }
2696                        Err(_) => {
2697                            debug!(target: "self_hosted_node::java", line = trimmed, "non-protocol stdout line");
2698                        }
2699                    }
2700                }
2701                Err(RecvTimeoutError::Timeout) => {
2702                    return Err(format!(
2703                        "java subprocess timed out after {}s",
2704                        CALL_TIMEOUT.as_secs()
2705                    ));
2706                }
2707                Err(RecvTimeoutError::Disconnected) => {
2708                    return Err("java subprocess closed stdout (crashed?)".to_string());
2709                }
2710            }
2711        }
2712    }
2713
2714    fn reset(&mut self) -> Result<(), String> {
2715        self.call("{\"command\":\"reset\"}").map(|_| ())
2716    }
2717
2718    fn is_alive(&mut self) -> bool {
2719        matches!(self.child.try_wait(), Ok(None))
2720    }
2721
2722    fn shutdown(mut self) {
2723        let _ = self.stdin.write_all(b"{\"command\":\"quit\"}\n");
2724        let _ = self.stdin.flush();
2725        let deadline = Instant::now() + SHUTDOWN_GRACE;
2726        loop {
2727            match self.child.try_wait() {
2728                Ok(Some(_)) => break,
2729                Ok(None) if Instant::now() >= deadline => {
2730                    let _ = self.child.kill();
2731                    let _ = self.child.wait();
2732                    break;
2733                }
2734                Ok(None) => std::thread::sleep(Duration::from_millis(100)),
2735                Err(_) => break,
2736            }
2737        }
2738        if let Some(handle) = self.stdout_handle.take() {
2739            let _ = handle.join();
2740        }
2741        if let Some(handle) = self.stderr_handle.take() {
2742            let _ = handle.join();
2743        }
2744    }
2745}
2746
2747#[cfg(feature = "java-forge")]
2748impl JavaBridge for SubprocessBridge {
2749    fn initialize(&mut self, _assets_dir: &str) -> Result<(), String> {
2750        Ok(())
2751    }
2752
2753    fn start_game_json(&mut self, request_json: &str) -> Result<String, String> {
2754        let body = json!({ "command": "startGame", "payload": request_json });
2755        self.call(&body.to_string())
2756    }
2757
2758    fn submit_action(&mut self, session_id: &str, action_json: &str) -> Result<String, String> {
2759        let body = json!({
2760            "command": "submitAction",
2761            "sessionId": session_id,
2762            "payload": action_json,
2763        });
2764        self.call(&body.to_string())
2765    }
2766
2767    fn get_prompt(
2768        &mut self,
2769        session_id: &str,
2770        player_index: usize,
2771    ) -> Result<Option<String>, String> {
2772        let body = json!({
2773            "command": "getPrompt",
2774            "sessionId": session_id,
2775            "playerIndex": player_index,
2776        });
2777        let prompt = self.call(&body.to_string())?;
2778        Ok((!prompt.is_empty()).then_some(prompt))
2779    }
2780
2781    fn get_snapshot(&mut self, session_id: &str, viewer: Option<usize>) -> Result<String, String> {
2782        let mut body = json!({ "command": "getSnapshot", "sessionId": session_id });
2783        if let Some(viewer) = viewer {
2784            body["viewer"] = json!(viewer);
2785        }
2786        self.call(&body.to_string())
2787    }
2788
2789    fn is_game_over(&mut self, session_id: &str) -> Result<bool, String> {
2790        let body = json!({ "command": "getGameOver", "sessionId": session_id });
2791        let value = self.call(&body.to_string())?;
2792        Ok(value.trim() == "true")
2793    }
2794
2795    fn end_game(&mut self, session_id: &str) -> Result<(), String> {
2796        let body = json!({ "command": "endGame", "sessionId": session_id });
2797        self.call(&body.to_string()).map(|_| ())
2798    }
2799
2800    fn abort_game(&mut self, session_id: &str) -> Result<(), String> {
2801        let body = json!({ "command": "abortGame", "sessionId": session_id });
2802        self.call(&body.to_string()).map(|_| ())
2803    }
2804}
2805
2806#[cfg(feature = "java-forge")]
2807impl Drop for SubprocessBridge {
2808    fn drop(&mut self) {
2809        let _ = self.child.kill();
2810        let _ = self.child.wait();
2811        if let Some(handle) = self.stdout_handle.take() {
2812            let _ = handle.join();
2813        }
2814        if let Some(handle) = self.stderr_handle.take() {
2815            let _ = handle.join();
2816        }
2817    }
2818}
2819
2820#[cfg(feature = "java-forge")]
2821fn resolve_java_bin(config: &JavaRuntimeConfig) -> String {
2822    if let Some(home) = &config.java_home {
2823        let bin = home.join("bin").join("java");
2824        if bin.is_file() {
2825            return bin.to_string_lossy().to_string();
2826        }
2827    }
2828    if let Ok(home) = env::var("JAVA_HOME") {
2829        let bin = PathBuf::from(home).join("bin").join("java");
2830        if bin.is_file() {
2831            return bin.to_string_lossy().to_string();
2832        }
2833    }
2834    "java".to_string()
2835}
2836
2837#[derive(Debug, Clone, Serialize)]
2838#[serde(rename_all = "camelCase")]
2839pub struct StartGameRequest {
2840    game_id: String,
2841    variant: String,
2842    starting_life: i32,
2843    seed: u64,
2844    players: Vec<PlayerConfig>,
2845}
2846
2847#[derive(Debug, Clone, Serialize)]
2848#[serde(rename_all = "camelCase")]
2849pub struct PlayerConfig {
2850    name: String,
2851    deck: Vec<CardIdentityForJava>,
2852    commander_names: Vec<String>,
2853    ai: bool,
2854}
2855
2856#[derive(Debug, Clone, Serialize)]
2857#[serde(rename_all = "camelCase")]
2858pub struct CardIdentityForJava {
2859    name: String,
2860    set_code: Option<String>,
2861    collector_number: Option<String>,
2862    foil: bool,
2863}
2864
2865#[derive(Debug, serde::Deserialize)]
2866#[serde(rename_all = "camelCase")]
2867struct StartGameResponse {
2868    session_id: String,
2869    #[allow(dead_code)]
2870    player_indexes: Vec<usize>,
2871}
2872
2873impl StartGameRequest {
2874    pub fn new(
2875        game_id: String,
2876        variant: String,
2877        starting_life: i32,
2878        seed: u64,
2879        players: Vec<PlayerConfig>,
2880    ) -> Self {
2881        Self {
2882            game_id,
2883            variant,
2884            starting_life,
2885            seed,
2886            players,
2887        }
2888    }
2889
2890    pub fn to_json(&self) -> Result<String, serde_json::Error> {
2891        serde_json::to_string(self)
2892    }
2893}
2894
2895impl PlayerConfig {
2896    pub fn new(name: String, deck: &[DeckCardIdentity], commander_names: Vec<String>) -> Self {
2897        Self {
2898            name,
2899            deck: deck.iter().map(CardIdentityForJava::from).collect(),
2900            commander_names,
2901            ai: false,
2902        }
2903    }
2904}
2905
2906impl From<&DeckCardIdentity> for CardIdentityForJava {
2907    fn from(identity: &DeckCardIdentity) -> Self {
2908        Self {
2909            name: java_card_name(&identity.name),
2910            set_code: (!identity.set_code.is_empty()).then(|| identity.set_code.clone()),
2911            collector_number: (!identity.card_number.is_empty())
2912                .then(|| identity.card_number.clone()),
2913            foil: identity.foil.unwrap_or(false),
2914        }
2915    }
2916}
2917
2918fn java_card_name(name: &str) -> String {
2919    name.split_once(" // ")
2920        .map(|(front, _)| front.to_string())
2921        .unwrap_or_else(|| name.to_string())
2922}
2923
2924fn env_path(key: &str) -> Option<PathBuf> {
2925    env::var_os(key)
2926        .filter(|value| !value.is_empty())
2927        .map(PathBuf::from)
2928}
2929
2930fn env_sizing(key: &str, default: u64) -> Option<u64> {
2931    let value = match env::var(key) {
2932        Ok(raw) if !raw.trim().is_empty() => match raw.trim().parse::<u64>() {
2933            Ok(parsed) => parsed,
2934            Err(_) => {
2935                tracing::warn!(target: "self_hosted_node::java", key, raw, "ignoring unparseable jvm sizing");
2936                default
2937            }
2938        },
2939        _ => default,
2940    };
2941    (value > 0).then_some(value)
2942}
2943
2944fn env_classpath(key: &str) -> Vec<PathBuf> {
2945    let Some(value) = env::var_os(key) else {
2946        return Vec::new();
2947    };
2948    env::split_paths(&value).collect()
2949}
2950
2951fn require_dir(path: &Path, label: &str) -> Result<(), String> {
2952    if path.is_dir() {
2953        Ok(())
2954    } else {
2955        Err(format!("{label} does not exist: {}", path.display()))
2956    }
2957}
2958
2959fn require_file(path: &Path, label: &str) -> Result<(), String> {
2960    if path.is_file() {
2961        Ok(())
2962    } else {
2963        Err(format!("{label} does not exist: {}", path.display()))
2964    }
2965}
2966
2967/// One `Pause` line from the JVM's unified GC log.
2968struct GcPause {
2969    kind: &'static str,
2970    millis: f64,
2971    heap_after_mb: Option<u64>,
2972}
2973
2974/// Parse a unified-log GC pause, e.g.
2975/// `[12.3s][info][gc] GC(7) Pause Full (Allocation Failure) 240M->171M(247M) 226.856ms`
2976fn parse_gc_pause(line: &str) -> Option<GcPause> {
2977    let kind = if line.contains("Pause Full") {
2978        "full"
2979    } else if line.contains("Pause Young") {
2980        "young"
2981    } else {
2982        return None;
2983    };
2984    let millis = line
2985        .rsplit(' ')
2986        .find_map(|token| token.strip_suffix("ms")?.parse::<f64>().ok())?;
2987    // "240M->171M(247M)": the figure after the arrow is what survived.
2988    let heap_after_mb = line.split_once("->").and_then(|(_, rest)| {
2989        let digits: String = rest.chars().take_while(char::is_ascii_digit).collect();
2990        digits.parse::<u64>().ok()
2991    });
2992    Some(GcPause {
2993        kind,
2994        millis,
2995        heap_after_mb,
2996    })
2997}
2998
2999#[cfg(test)]
3000mod gc_log_tests {
3001    use super::{parse_gc_pause, PlayerConfig, StartGameRequest};
3002    use manabrew_protocol::deck_dto::DeckCardIdentity;
3003
3004    #[test]
3005    fn reads_a_full_collection() {
3006        let pause = parse_gc_pause(
3007            "[117.6s][info][gc] GC(21) Pause Full (Allocation Failure) 240M->171M(247M) 226.856ms",
3008        )
3009        .expect("parsed");
3010        assert_eq!(pause.kind, "full");
3011        assert_eq!(pause.heap_after_mb, Some(171));
3012        assert!((pause.millis - 226.856).abs() < f64::EPSILON);
3013    }
3014
3015    #[test]
3016    fn reads_a_young_collection() {
3017        let pause = parse_gc_pause(
3018            "[3.1s][info][gc] GC(2) Pause Young (Allocation Failure) 216M->91M(494M) 17.254ms",
3019        )
3020        .expect("parsed");
3021        assert_eq!(pause.kind, "young");
3022        assert_eq!(pause.heap_after_mb, Some(91));
3023    }
3024
3025    /// The node asks for `-Xlog:gc*:stderr:time,uptime,level,tags`, which is a
3026    /// richer line than plain `-Xlog:gc`. Real production output.
3027    #[test]
3028    fn reads_the_format_the_node_actually_requests() {
3029        let pause = parse_gc_pause(
3030            "[2026-08-18T20:41:42.907+0000][120025.819s][info][gc] GC(1202) Pause Full (Allocation Failure) 494M->487M(494M) 545.344ms",
3031        )
3032        .expect("parsed");
3033        assert_eq!(pause.kind, "full");
3034        assert_eq!(pause.heap_after_mb, Some(487));
3035        assert!((pause.millis - 545.344).abs() < f64::EPSILON);
3036    }
3037
3038    /// `gc*` also emits region and metaspace lines that carry `->`; they must
3039    /// not be mistaken for pauses.
3040    #[test]
3041    fn ignores_the_extra_lines_gc_star_adds() {
3042        assert!(
3043            parse_gc_pause("[120025.819s][info][gc,heap] GC(1202) Eden regions: 10->0(12)")
3044                .is_none()
3045        );
3046        assert!(
3047            parse_gc_pause("[120025.819s][info][gc,metaspace] Metaspace: 48M->48M(1088M)")
3048                .is_none()
3049        );
3050    }
3051
3052    #[test]
3053    fn ignores_other_output() {
3054        assert!(parse_gc_pause("[java] LOGGER ERROR: something").is_none());
3055        assert!(parse_gc_pause("[1.0s][info][gc,init] CardTable entry size: 512").is_none());
3056    }
3057
3058    #[test]
3059    fn serializes_exact_printing_identity_for_java() {
3060        let card = DeckCardIdentity {
3061            name: "Forest".to_string(),
3062            set_code: "EOE".to_string(),
3063            card_number: "266".to_string(),
3064            foil: Some(true),
3065            ..Default::default()
3066        };
3067        let request = StartGameRequest::new(
3068            "printing-test".to_string(),
3069            "Constructed".to_string(),
3070            20,
3071            42,
3072            vec![PlayerConfig::new("Player".to_string(), &[card], Vec::new())],
3073        );
3074        let json = request.to_json().expect("serialized");
3075
3076        assert!(json.contains(
3077            r#""deck":[{"name":"Forest","setCode":"EOE","collectorNumber":"266","foil":true}]"#
3078        ));
3079    }
3080}