`gjA UddlZddlZddlZejddlZddlZddlZddlZddlZddl m Z ddl m Z ddl mZmZmZmZddlmZej(eZdaddZdZdefd Zd eddfd Zeed r'ej ej>ej@eed rej ejBeeedrej ejDen&eedrej ejFeeedr'ej ejHej@de%ddfdZ&dddZ'de(fdZ)ddedzde(fdZ*gZ+e,ee-d<dZ.edk(re.yy)N)handle_spurious_eof)server) _CRASH_LOGdispatch resolve_skin write_json) TeeTransportreturnctjjd}|syddlm}t t j||t _y)aMirror every dispatcher emit to the dashboard sidebar via WS. Activated by `HERMES_TUI_SIDECAR_URL`, set by the dashboard's ``/api/pty`` endpoint when a chat tab passes a ``channel`` query param. Best-effort: connect failure or runtime drop falls back to stdio-only. HERMES_TUI_SIDECAR_URLNr)WsPublisherTransport)osenvirongettui_gateway.event_publisherr r r_stdio_transport)urlr s D/root/.hermes/venv/lib/python3.12/site-packages/tui_gateway/entry.py_install_sidecar_publisherr!s> **..1 2C @*!5c!:Fg?ctjjdxsdj}|stS t |}|dkDr|StS#t $r tcYSwxYw)N#HERMES_TUI_GATEWAY_SHUTDOWN_GRACE_Sr)rrrstrip_DEFAULT_SHUTDOWN_GRACE_Sfloat ValueError)rawvalues r_shutdown_grace_secondsr ?se ::>>? @ FB M M OC (()c AI5<#<< )(()s AA%$A%signumc xi}dD]$}tt|d}|||t|<&|j|d|} t j tj jtdttdd5}|jd |d tjd d |(|jd tj||ddl}|j j#D]{\}} |jd| j$d|d|jdj'tj(t+j,j|} dddt1d|t*j2dddl}dd} |j5t7| } d| _| j; ddlm} |  t+j@dy#1swYxYw#t.$rYwxYw#t.$rY=wxYw#tB$rwxYw)uCapture WHICH thread and WHERE a termination signal hit us. SIG_DFL for SIGPIPE kills the process silently the instant any background thread (TTS playback, beep, voice status emitter, etc.) writes to a stdout the TUI has stopped reading. Without this handler the gateway-exited banner in the TUI has no trace — the crash log never sees a Python exception because the kernel reaps the process before the interpreter runs anything. Termination semantics: ``sys.exit(0)`` here used to race the worker pool — a thread holding ``_stdout_lock`` mid-flush would block the interpreter shutdown indefinitely. We now log the stack, give the process the configured shutdown grace (``HERMES_TUI_GATEWAY_SHUTDOWN_GRACE_S``, default ``_DEFAULT_SHUTDOWN_GRACE_S``) to drain naturally on a background thread, and fall back to ``os._exit(0)`` so a wedged write/flush can never strand the process. )SIGPIPESIGTERMSIGHUPSIGINTSIGBREAKNzsignal Texist_okautf-8encodingz === u received · %Y-%m-%d %H:%M:%S === z&main-thread stack at signal delivery: )filerz --- thread z (id=z) --- rz[gateway-signal] r0flushc.tjdy)Nr)r_exitrr _hard_exitz_log_signal.._hard_exitzs  r)_shutdown_sessionsr N)"getattrsignalintrrmakedirspathdirnameropenwritetimestrftime traceback print_stack threading_activeitemsnamejoin format_stacksys_current_frames ExceptionprintstderrTimerr daemonstarttui_gateway.serverr7exit SystemExit) r!frame _signal_names_attr_sigrHf _threadingtidthr6timerr7s r _log_signalr_Js *%'MG-vud+  ',M#d) $-   Vwvh%7 8D  BGGOOJ/$? *cG 4 Y GGmDMM:M,N+OvV  AB%%e!4 +%--335 YR-ycU'BC 6 6s7J7J7L7P7PQT7U VWX Y Y dV $3::TB"   46 CEEL KKM 9   [ Y Y    8     sJA H C2H?H H.H.H  H HH H+*H+. H9r#r$r%r'r&reasoncv tjtjjtdt tdd5}|j dtjdd|d d d d td |tjd y #1swY)xYw#t$rY7wxYw) uRecord why the gateway subprocess is shutting down. Three exit paths (startup write fail, parse-error-response write fail, dispatch-response write fail, stdin EOF) all collapse into a silent sys.exit(0) here. Without this trail the TUI shows "gateway exited" with no actionable clue about WHICH broken pipe or WHICH message triggered it — the main reason voice-mode turns look like phantom crashes when the real story is "TUI read pipe closed on this event". Tr(r*r+r,u === gateway exit · r.u · reason=r/Nz[gateway-exit] r1) rr<r=r>rr?r@rArBrMrNrKrO)r`rZs r _log_exitrbs  BGGOOJ/$? *cG 4  GG(7J)K(LM#HF,   OF8 $3::TB      s*A B, ,B 8B, B)%B,, B87B8timeoutct}||jsy ddlm}||}|j |y#t$r ||nd}Y$wxYw)aBlock until background MCP discovery finishes, up to the resolved bound. MCP discovery runs in a daemon thread spawned at startup (see main()) so a slow/dead server can't freeze ``gateway.ready``. But the agent snapshots its tool list ONCE at build time and never re-reads it, so a reachable-but- slow server that finishes connecting *after* the first prompt would be invisible for the whole session. Joining with a bounded timeout before the first agent build lets already-spawning servers land without re-introducing the startup hang: ``thread.join(timeout)`` returns the instant discovery completes (so fast/no-MCP startups pay ~0s), and a dead server is simply not waited on beyond the bound. No-op when no discovery thread was started. The bound comes from ``mcp_discovery_timeout`` in config (shared with the CLI path via ``hermes_cli.mcp_startup``); ``timeout`` overrides it. Nr)_resolve_discovery_timeoutg?rc)_mcp_discovery_threadis_alivehermes_cli.mcp_startuprerMrI)rcthreadrebounds rwait_for_mcp_discoveryrlsY #F ~V__.9E*73 KKK 9".D9s<A Acnt}||jry ddlm}|S#t$rYywxYw)uReturn True if ANY background MCP discovery thread is still running. Used by the agent-build path to decide whether to schedule a late tool snapshot refresh: if discovery didn't land within the bounded ``wait_for_mcp_discovery`` join, the agent was built without those tools and the banner/tool count will be stale until they arrive. There are two independent discovery-thread owners by surface: the stdio ``hermes --tui`` path spawns ITS thread here (``_mcp_discovery_thread``), while the desktop app + dashboard WebSocket sidecar (``tui_gateway/ws.py``) and ``hermes dashboard`` spawn theirs via ``hermes_cli.mcp_startup.start_background_mcp_discovery``. The late-refresh scheduler imports this function regardless of surface, so it MUST consult both — checking only the entry thread left the desktop/dashboard surfaces with no late refresh, so a slow MCP server's tools never surfaced for the whole session (#51587). Tr)mcp_discovery_in_flightF)rgrhrirnrM)rj_startup_in_flights rrnrnsB$#F foo/ "## s ( 44cd}t}|#|j||j } ddlm}||}|xr|S#t $rd}YwxYw)aBlock until background MCP discovery finishes, up to ``timeout`` seconds. Returns True if discovery has completed (both thread owners absent or no longer alive), False if either is still running after the timeout. Unlike ``wait_for_mcp_discovery`` this accepts an unbounded/long wait and reports the outcome, for the off-critical-path late-refresh waiter. Joins both discovery-thread owners (see ``mcp_discovery_in_flight``): the entry thread first, then the ``hermes_cli.mcp_startup`` thread used by the desktop/dashboard surfaces. ``timeout`` bounds EACH join, mirroring the pre-#51587 single-owner behavior for the entry thread. Trfr)join_mcp_discovery)rgrIrhrirqrM)rc entry_donerj _startup_join startup_dones rrqrq sfJ "F  G $** N$W5   &,&  sA AA_recovery_timesct ddlm}|xsijd}t |t xrt |dkD}|r.dd}ddl}|j|dd}|j|a tdd d d tid d s tdtj d tj"j%}|st't(tsy7|j+}|sJ t-j.|}t |t r|jdnd} t3|} | /t| s$td| dtj d#t$rd}Y:wxYw#t,j0$r6tdddddds tdtj dYwxYw)Nr)read_raw_config mcp_serversTcl ddlm}|y#t$rtj ddYywxYw)Nr-_discover_mcp_tools_without_interactive_oauthz$Background MCP tool discovery failedT)exc_info)rir{rMloggerwarningrzs r_discover_mcp_backgroundz&main.._discover_mcp_backgroundKs9 >? :T s  33ztui-mcp-discovery)targetrHrQz2.0eventz gateway.readyskin)typepayload)jsonrpcmethodparamszORS>S   +$+++$, !, * 7OP  PQ  ii  "' B yy{  **T"C'1d&;"}  d#=fZG\]^ 5 K    h## %6Vc:dlpqrRS   s%AE8F 8 FF AGG__main__r8)N)rcz float | Noner N)/rrKhermes_bootstrapharden_import_pathrloggingr:rArCtui_gateway._stdin_recoveryr tui_gatewayrrSrrrrtui_gateway.transportr getLogger__name__r}rgrrrr r;r_hasattrr#SIG_IGNr$r%r'r&strrbrlboolrnrqrulist__annotations__rr5rrrs ###%  ;MM.   8 $ 6 ==RRtRF 69FMM&..&..1 69FMM&..+. 68FMM&--- VZ FMM&//;/ 68FMM&--0CcCdC,8>' ''8 "e!Zz zFr