cosmetics

bins.ensure() now stores the resolved interpreter path under
python_bin / lua_bin when the key is unset, so test scripts can use
$(python_bin) / $(lua_bin) in GUI mode (no -d override). Restores the
behaviour lost when bins.py centralised resolution. A user-provided
value is left untouched.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

DESIGN.md

@@ -97,6 +97,15 @@ All dialog items (`dialog_image`, `dialog_question`, `dialog_references`, `dialo
 - For the live stream (terminal in batch / GUI panel), prefixes every line emitted from a branch's thread with `[<branch_name>] ` so concurrent branches stay readable.
 - Exposes `write` / `writeln` / `flush` (Python 3.14's `unittest` calls `stream.writeln()` directly without `_WritelnDecorator`).
 
+### Subprocess RPC startup handshake (py_func / lua_func / eval_proc)
+
+The parent ↔ subprocess JSON-RPC link runs over a localhost TCP socket. The **subprocess** owns the port: it binds `port 0` (OS-assigned), `listen()`s, then prints `__TESTIUM_RPC_PORT__=<port>` on stdout (constant `RPC_PORT_SENTINEL` in `runtime/jrpc.py`). The parent reads that line (`proc_drain.drain_and_read_port` + `wait_for_port`, deadline `gd("proc_start_timeout", 30)`) and only *then* connects — the server is guaranteed to be listening, so the connect succeeds on the first attempt.
+
+This replaced the previous fragile scheme (parent reserved a port via `bind(0)`+close, child re-bound the same port, parent connected on a timing guess) which broke intermittently on Windows: cold-start/antivirus variance pushed the worker past the connect deadline, and `connect()` to a not-yet-listening localhost port *times out* (≈1 s) instead of refusing, exhausting the retry budget. Notes:
+- The server no longer sets `SO_REUSEADDR` (a fresh ephemeral port needs no TIME_WAIT override; on Windows it would enable port hijacking).
+- `JsonRpcBase.wait_ready()` always settles (event set on success **and** failure) and returns the actual connection outcome — a connect failure no longer hangs a `wait_ready()` caller.
+- Non-sentinel subprocess stdout/stderr is still forwarded to the parent log (early-startup errors stay visible).
+
 ### Subprocess API contract (py_func / lua_func)
 
 User test scripts running inside a `py_func` or `lua_func` subprocess **must** use the JSON-RPC bridge to interact with testium state:
@@ -115,7 +124,7 @@ To add a new API call usable from subprocesses:
 `src/testium/interpreter/utils/bins.py` — single source of truth for the paths to the external Python and Lua interpreters used by subprocesses.
 
 - `python_bin()` / `lua_bin()` : resolve and cache. The cache is keyed by `(name, override)` so that a later change to `gd[python_bin]` (typically when a `param.yaml` sets the key) triggers a re-resolution on the next lookup instead of returning the stale auto-discovered path. Falls back to discovery on PATH (candidates: `python3`/`python` and `lua`/`lua5.5`/`lua5.4`/`lua5.3`/`lua5.2`/`lua5.1`).
-- `ensure(*names)` : called by `TestSet._validate_runtime_deps()` at test load. Always requires `python` (the eval engine always runs); requires `lua` only if a `lua_func` item is in the tree. Fails fast with a clear error citing tried candidates and override key.
+- `ensure(*names)` : called by `TestSet._validate_runtime_deps()` at test load. Always requires `python` (the eval engine always runs); requires `lua` only if a `lua_func` item is in the tree. Fails fast with a clear error citing tried candidates and override key. Also **publishes** each resolved path into gd (`python_bin` / `lua_bin`) when the key is unset, so test scripts can reference `$(python_bin)` / `$(lua_bin)` regardless of launch mode (e.g. GUI, where no `-d` override is passed). A user-provided value is never overwritten.
 
 Engines (`PyProcessBase`, `LuaProcessBase`, `EvalExecEngine`) call `bins.python_bin()`/`bins.lua_bin()` themselves — call sites never pass an explicit binary path.
 
@@ -279,6 +288,7 @@ The `testium_assist` editor extension is a thin LSP client that spawns `testium
 Both Flatpak and AppImage export `TESTIUM_VERSION` from a launcher (Flatpak: launcher script in `org.testium.Testium.yaml`; AppImage: `runtime.env` in `AppImageBuilder.yml`). `get_testium_version()` checks `/.flatpak-info` / `APPIMAGE` and reads `TESTIUM_VERSION` rather than relying on package metadata or repo introspection.
 
 ## Recent fixes / notable changes
+- Subprocess RPC startup handshake: the `py_func`/`lua_func`/`eval_proc` worker now picks its own port (`bind 0`), announces it on stdout (`__TESTIUM_RPC_PORT__=`), and the parent connects only after reading it. Fixes intermittent Windows `failed to connect : timeout` and the matching `wait_ready()` hang; removes the reserve/close/rebind race and `SO_REUSEADDR`. See "Subprocess RPC startup handshake".
 - `build_all.sh`: builds the four heavy channels in parallel (serial prep for the shared venv + wheel), results in completion order, Ctrl+C kills the whole job tree; `--ram` puts the build scratch on tmpfs (`/dev/shm`) + skips UPX for fast builds on USB/SD storage (Flatpak excluded — rofiles-fuse can't mount tmpfs). See the "Building all channels" section.
 - LSP across packaging channels: `testium lsp` (and the `testium_assist` editor extension that spawns it) now works from source, wheel, PyInstaller, Flatpak and AppImage. Two enablers — (1) action items declare a class-level `ACTIONS = {key: class}` registry (like `PARAMS`), so `lsp/schema.py` builds the full schema from class attributes with no `inspect.getsource`/AST (which broke under frozen PyInstaller); (2) the `[lsp]` extra (pygls) is wired into every full-app channel. `test/validation/lsp_check.py`, run by `run.sh` before the suite, asserts per-channel that `schema` keeps its actions and `lsp` answers `initialize`. See the matching architecture sections.
 - Declarative test item parameters (v0.2): each `TestItem` subclass exposes a `PARAMS = ParamSet(...)` class attribute consumed by the base `__init__`. Catches unknown YAML keys (typo warnings listing the accepted names) and missing required params (load-time errors with `.tum` context). Lays the schema foundation for a future LSP server and auto-generated manual sections. See the matching architecture section.

release_note.txt

@@ -1,3 +1,14 @@
+version 0.2.3
+=============
+- Windows version now working reliably. Fix of a problem of jrpc ports 
+  handshakes between the py and lua processes and testium
+
+version 0.2.2
+==============
+- Flatpak sandbox issue fixed for term console. Now a term console is
+  exactly like a host console.
+- Persistence fix of dialogs in case of flatpak.
+
 version 0.2.1
 ==============
 - Faster test loading, especially for large tests built from jinja

src/VERSION

@@ -1 +1 @@
-0.2.1
+0.2.3

src/testium/__init__.py

@@ -11,6 +11,16 @@ sys.path.append(os.path.abspath(ourpath.parent))
 import interpreter.utils.constants as cst
 
 def main():
+    # Force UTF-8 on stdout/stderr so the runner's output survives a legacy
+    # console code page (Windows cp1252 can't encode box-drawing/accented
+    # chars). Only the stream encoders change; the locale default used for
+    # config files is untouched.
+    for _stream in (sys.stdout, sys.stderr):
+        try:
+            _stream.reconfigure(encoding="utf-8")
+        except (AttributeError, ValueError):
+            pass  # no stdout (frozen GUI) or non-reconfigurable stream
+
     # Subcommand dispatch (must run *before* argparse so neither 'schema' nor
     # 'lsp' has to share the GUI/batch flag surface). The subcommands also
     # skip the multiprocessing 'spawn' setup which is only meaningful for the

src/testium/api/termconsole.py

@@ -81,9 +81,13 @@ class TermConsole(Console):
                                    bufsize=0)
 
         else:
-            self.term = pexpect.spawn(  shell_cmd,
+            # In Flatpak this returns a `flatpak-spawn --host` wrapper so the
-                                        echo=False,
+            # console behaves like a host shell (matching py_func / lua_func /
-                                        cwd=self.ppath)
+            # run); elsewhere it's the chosen command unchanged.
+            from interpreter.utils import bins
+            argv = bins.host_console_command(shell_cmd, self.ppath)
+            self.term = pexpect.spawn(argv[0], args=argv[1:],
+                                      echo=False, cwd=self.ppath)
 
         self.q = BytesStore()
         self.t = threading.Thread(target=self.enqueue_output)

src/testium/interpreter/test_items/dialog_choices_files/choices_dialog.py

@@ -221,6 +221,11 @@ def main(args, conn=None):
 
     if conn:
         settings.setValue(SettingsLastChoices, result)
+        # Flush before sending: the parent terminates this subprocess as soon
+        # as it reads the result, so the QSettings destructor never runs and
+        # the write would race the kill (lost under Flatpak — see the
+        # tested-references dialog for the full rationale).
+        settings.sync()
         conn.send([result, success])
         conn.close()
     else:

src/testium/interpreter/test_items/test_item_py_func.py

@@ -90,7 +90,7 @@ class TestItemPyFunc(TestItem):
 
             if not engine.is_alive():
                 engine.start()
-                if not engine.wait_ready():
+                if not engine.wait_ready(10):
                     raise ETUMRuntimeError(
                         f"""Impossible to start the external python execution process.
 Is the python path correct ?

src/testium/interpreter/test_items/tested_references_files/tested_refs_dialog.py

@@ -76,6 +76,12 @@ def main(args, conn=None):
 
     if conn:
         settings.setValue(SettingsLastReference, result)
+        # Flush to disk *before* handing the result back: as soon as the parent
+        # receives it on the pipe it terminates this subprocess (SIGTERM, no
+        # handler), so the QSettings destructor never runs. Without sync() the
+        # write races the kill and is lost — reliably so under Flatpak, where
+        # the .conf is atomically renamed on the slower ~/.var/app overlay.
+        settings.sync()
         conn.send([result, success])
         conn.close()
     else:

src/testium/interpreter/test_report/report_export_html.py

@@ -14,7 +14,7 @@ class ReportExportHTML(rpe.ReportExport):
         self.prepareFile()
         self.create_base()
         self.process_tests()
-        with open(self._file_name, 'w') as f:
+        with open(self._file_name, 'w', encoding="utf-8") as f:
             f.write(lxml.html.tostring(self.root, pretty_print=True).decode())
 
     def testsIterate(self, row):

src/testium/interpreter/test_report/report_export_junit.py

@@ -20,7 +20,7 @@ class ReportExportJUnit(rpe.ReportExport):
 
         ts = TestSuite(repname, test_cases=self.test_cases,
                        hostname=tm.gd('host_ip'))
-        with open(self._file_name, 'w') as f:
+        with open(self._file_name, 'w', encoding="utf-8") as f:
             TestSuite.to_file(f, [ts])
 
     def testsIterate(self, row):

src/testium/interpreter/utils/bins.py

@@ -19,6 +19,7 @@ Public API
 
 import atexit
 import os
+import shlex
 import shutil
 import subprocess
 import tempfile
@@ -177,6 +178,27 @@ def flatpak_host_spawn(interp_bin, cmd_args, host_cwd, extra_env=None):
     return spawn
 
 
+def host_console_command(shell_cmd, cwd):
+    """Build the argv to start *shell_cmd* as an ordinary interactive console.
+
+    *shell_cmd* is the command the caller chose (a string — shell-split — or
+    an argv list); the choice is preserved verbatim.
+
+    Outside Flatpak the command is returned unchanged. Inside Flatpak a bare
+    spawn would run in the sandbox under the runtime python3, so a host venv
+    (``/path/venv/bin/python3 -m mod``) can't see its pip deps. We simply run
+    it on the host with ``flatpak-spawn --host`` so it behaves like any other
+    terminal: flatpak-spawn passes the current environment through unchanged
+    and the shell (sourced venv, profile, …) sets things up as the user wants.
+    No env forwarding or scrubbing — the launcher's leaked PYTHONPATH points at
+    /app paths absent on the host, so it's inert there.
+    """
+    argv = shlex.split(shell_cmd) if isinstance(shell_cmd, str) else list(shell_cmd)
+    if not _in_flatpak():
+        return argv
+    return ["flatpak-spawn", "--host", f"--directory={cwd}", *argv]
+
+
 def _which_host_flatpak(name):
     """Resolve a binary name (or absolute path) on the host via flatpak-spawn.
 
@@ -366,12 +388,16 @@ def ensure(*names):
     """
     missing = []
     for n in names:
-        if not _resolve(n):
+        path = _resolve(n)
         display, gd_key, candidates, _ = _SPECS[n]
+        if not path:
             missing.append(
                 f"  - {display}: tried {candidates} on PATH, none usable. "
                 f"Set '{gd_key}' in the YAML config to override."
             )
+        elif not tm.gd(gd_key):
+            # Publish resolved path so test scripts can use $(python_bin)/$(lua_bin).
+            tm.setgd(gd_key, path)
     if missing:
         raise ETUMRuntimeError(
             "Required external interpreter(s) not found:\n" + "\n".join(missing)

src/testium/interpreter/utils/lua_process.py

@@ -1,14 +1,13 @@
 import os
 import sys
 import subprocess
-import socket
 
 import api.testium as tm
 from runtime.jrpc import JsonRpcClient
 from interpreter.utils.paths import subproc_path
 from runtime.tum_except import ETUMRuntimeError
 from interpreter.utils import bins
-from interpreter.utils.proc_drain import drain_to_log
+from interpreter.utils.proc_drain import drain_and_read_port, wait_for_port
 
 
 class LuaProcessBase:
@@ -79,12 +78,7 @@ class LuaProcessBase:
                 else:
                     env[k] = e + ";" + env.get(k, "")
 
-        sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+        # POpen params (port 0 -> the Lua server picks a free port and reports it)
-        sock.bind(("localhost", 0))
-        self._port = sock.getsockname()[1]
-        sock.close()
-
-        # POpen params
         cmd_args = [
             "main.lua",
             "--timeout",
@@ -92,7 +86,7 @@ class LuaProcessBase:
             "--host",
             "127.0.0.1",
             "--port",
-            f"{self._port}",
+            "0",
         ]
 
         if tm.debug_enabled() and tm.gd("debug_rpc", False):
@@ -122,10 +116,16 @@ class LuaProcessBase:
             restore_signals=False,
             **popen_kwargs,
         )
-        # Route subprocess stdout/stderr (lua require failures, syntax
+        # Forward subprocess output to the log and read the startup port sentinel.
-        # errors, anything written to fd 1/2 before the in-script
+        holder = drain_and_read_port(self._process, prefix="[lua_func] ")
-        # remote_print is set up) into the parent's log.
+        self._port = wait_for_port(
-        drain_to_log(self._process, prefix="[lua_func] ")
+            self._process, holder, tm.gd("proc_start_timeout", 30)
+        )
+        if self._port is None:
+            # Worker died before announcing its port: reset so a later start() retries clean.
+            self.stop()
+            self.join()
+            return
 
         self._rpc = JsonRpcClient(
             "localhost", self._port, req_handler=self._req_handler

src/testium/interpreter/utils/proc_drain.py

@@ -8,6 +8,9 @@ exceptions before the in-process redirection kicks in, lua
 ``require`` failures, anything written to fd 1/2 directly).
 """
 import threading
+from time import monotonic
+
+from runtime.jrpc import RPC_PORT_SENTINEL
 
 
 def _drain_pipe(pipe, prefix):
@@ -46,3 +49,60 @@ def drain_to_log(process, prefix=""):
         t.start()
         threads.append(t)
     return threads
+
+
+def drain_and_read_port(process, prefix=""):
+    """Like :func:`drain_to_log`, but the stdout reader also watches for the
+    startup port sentinel. Returns a ``holder`` dict (passed to
+    :func:`wait_for_port`); all non-sentinel lines are still forwarded to the
+    log. stderr is drained as usual.
+    """
+    holder = {"port": None, "evt": threading.Event()}
+
+    def _read_stdout(pipe):
+        try:
+            for raw in iter(pipe.readline, b""):
+                line = raw.decode("utf-8", errors="replace").rstrip("\r\n")
+                if holder["port"] is None and line.startswith(RPC_PORT_SENTINEL):
+                    try:
+                        holder["port"] = int(line[len(RPC_PORT_SENTINEL):].strip())
+                    except ValueError:
+                        continue
+                    holder["evt"].set()
+                    continue
+                if line:
+                    print(f"{prefix}{line}" if prefix else line)
+        finally:
+            try:
+                pipe.close()
+            except Exception:
+                pass
+            holder["evt"].set()  # unblock waiter on EOF even without sentinel
+
+    if process.stdout is not None:
+        threading.Thread(
+            target=_read_stdout, args=(process.stdout,), daemon=True,
+        ).start()
+    if process.stderr is not None:
+        threading.Thread(
+            target=_drain_pipe, args=(process.stderr, prefix), daemon=True,
+        ).start()
+    return holder
+
+
+def wait_for_port(process, holder, deadline):
+    """Block until the port sentinel arrives, the process dies, or *deadline*
+    seconds elapse. Returns the port int or ``None``.
+    """
+    end = monotonic() + deadline
+    while holder["port"] is None:
+        remaining = end - monotonic()
+        if remaining <= 0:
+            break
+        holder["evt"].wait(min(remaining, 0.2))
+        if holder["port"] is not None:
+            break
+        if process.poll() is not None:
+            holder["evt"].wait(0.2)  # child exited; let the reader flush a trailing line
+            break
+    return holder["port"]

src/testium/interpreter/utils/py_process.py

@@ -1,13 +1,12 @@
 import os
 import sys
 import subprocess
-import socket
 from runtime.jrpc import JsonRpcClient
 import api.testium as tm
 from runtime.tum_except import ETUMRuntimeError
 from interpreter.utils.paths import testium_path, subproc_path
 from interpreter.utils import bins
-from interpreter.utils.proc_drain import drain_to_log
+from interpreter.utils.proc_drain import drain_and_read_port, wait_for_port
 
 
 class PyProcessBase:
@@ -54,13 +53,6 @@ class PyProcessBase:
                 else:
                     env[k] = e + os.pathsep + env.get(k, "")
 
-        sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
-        sock.bind(("localhost", 0))
-        self._port = sock.getsockname()[1]
-        # Port was reserved until the sub-process is started. Now released.
-        if sock is not None:
-            sock.close()
-
         # In Flatpak the host can't see /app/lib/testium, so use a staged copy
         # under /tmp (shared between sandbox and host) for both cwd and as the
         # root in PYTHONPATH. Outside Flatpak the original paths are used.
@@ -75,7 +67,7 @@ class PyProcessBase:
         cmd_args = [
             "py_func",
             "-p",
-            f"{self._port}",
+            "0",
             "-t",
             f"{self._timeout}",
         ]
@@ -107,11 +99,16 @@ class PyProcessBase:
             restore_signals=False,
             **popen_kwargs,
         )
-        # Route subprocess stdout/stderr (early-startup errors,
+        # Forward subprocess output to the log and read the startup port sentinel.
-        # unhandled exceptions, anything written to fd 1/2 before the
+        holder = drain_and_read_port(self._process, prefix="[py_func] ")
-        # in-process JSON-RPC stdio_redir kicks in) into the parent's
+        self._port = wait_for_port(
-        # log.
+            self._process, holder, tm.gd("proc_start_timeout", 30)
-        drain_to_log(self._process, prefix="[py_func] ")
+        )
+        if self._port is None:
+            # Worker died before announcing its port: reset so a later start() retries clean.
+            self.stop()
+            self.join()
+            return
 
         self._rpc = JsonRpcClient(
             "localhost", self._port, req_handler=self._req_handler

src/testium/lua_func/main.lua

@@ -3,7 +3,7 @@
 -- =========================
 local config = {
     host = "0.0.0.0",
-    port = 9000,
+    port = 0,          -- 0 = OS-assigned; actual port is reported on stdout
     timeout = 60,
     verbose = false,
 }
@@ -76,6 +76,10 @@ server_sock:listen(1)
 local ip, port = server_sock:getsockname()
 utils.log("listening on %s:%d for %.1f secs", ip, port, config.timeout)
 
+-- Announce the actual bound port so the parent connects only once we listen.
+io.stdout:write("__TESTIUM_RPC_PORT__=" .. port .. "\n")
+io.stdout:flush()
+
 server_sock:settimeout(config.timeout) -- Prevents hanging on dead connections
 
 -- Main Server Loop

src/testium/py_func/__init__.py

@@ -1,7 +1,9 @@
 #!/usr/bin/env python
+import sys
 import multiprocessing
 from py_func.tm import _init_api, _remote_print
 from runtime.stdout_redirect import stdio_redir
+from runtime.jrpc import RPC_PORT_SENTINEL
 
 
 class TcpStdOut:
@@ -24,21 +26,29 @@ def main():
     parser = argparse.ArgumentParser()
     parser.add_argument("-i", "--ip", type=str, help="Ip address or hostname to listen to",
                         default="localhost")
-    parser.add_argument("-p", "--port", type=int, help="port to listen to",
+    parser.add_argument("-p", "--port", type=int, help="port to listen to (0 = OS-assigned)",
-                        default=9000)
+                        default=0)
     parser.add_argument("-t", "--timeout", type=float, help="Timeout waiting for connection",
                         default=10)
     parser.add_argument("-v", "--verbose", action='store_true', help="port to listen to")
     args = parser.parse_args()
 
     thrd_api = _init_api(args.ip, args.port, args.timeout)
-    # redirect I/O
-    outstream = TcpStdOut()
-    stdio_redir.redirect(outstream)
     # debug the server
     if args.verbose:
         thrd_api.dbg_out = stdio_redir.ini_stdout
     thrd_api.start()
+
+    # Announce the bound port on real stdout (before redirection) so the parent connects.
+    port = thrd_api.wait_bound(args.timeout)
+    if port is None:
+        print("py_func: failed to bind a listening port", file=sys.stderr, flush=True)
+        return
+    print(f"{RPC_PORT_SENTINEL}{port}", flush=True)
+
+    # redirect I/O
+    outstream = TcpStdOut()
+    stdio_redir.redirect(outstream)
     try:
         while thrd_api.is_alive():
             thrd_api.join(1)

src/testium/runtime/jrpc.py

@@ -12,6 +12,9 @@ except:
 
 from runtime.tum_except import ETUMRuntimeError
 
+# Startup handshake: subprocess prints this + its bound port on stdout once listening.
+RPC_PORT_SENTINEL = "__TESTIUM_RPC_PORT__="
+
 """Lightweight JSON-RPC 2.0 helpers over TCP sockets.
 
 This module implements a minimal JSON-RPC 2.0 messaging layer using
@@ -279,6 +282,8 @@ class JsonRpcBase(threading.Thread):
         self._req_handler = req_handler
         self._dbg_out = dbg_out
         self._event_ready = threading.Event()
+        # Set on success AND failure so wait_ready() never hangs; outcome in _connected.
+        self._connected = False
 
     def handle_request(self, method, params):
         """Override to implement server-side request handling.
@@ -314,10 +319,12 @@ class JsonRpcBase(threading.Thread):
             self.name, sock, self.handle_request, dbg_out=self.dbg_out
         )
         self._rpc.wait_ready()
+        self._connected = True
         self._event_ready.set()
 
     def wait_ready(self, timeout=None):
-        return self._event_ready.wait(timeout)
+        self._event_ready.wait(timeout)
+        return self._connected
 
     @property
     def dbg_out(self):
@@ -348,20 +355,30 @@ class JsonRpcSrv(JsonRpcBase):
     def __init__(self, host, port, req_handler=None, timeout=10):
         super().__init__(host, port, req_handler, timeout)
         self.name = f"JsonRpcSvr_{port}"
+        self._bound_port = None
+        self._bound_evt = threading.Event()
+
+    @property
+    def bound_port(self):
+        return self._bound_port
+
+    def wait_bound(self, timeout=None):
+        self._bound_evt.wait(timeout)
+        return self._bound_port
 
     def run(self):
         # TCP/IP socket creation
         try:
             with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as sock:
 
-                sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+                # No SO_REUSEADDR: fresh ephemeral port; on Windows it enables hijacking.
-
-                # Link of the socket at the configured port
                 sock.bind((self._host, self._port))
 
                 # Listens incoming connections
                 sock.listen(1)
-                self.print_info(f"listening on {self._host}:{self._port}")
+                self._bound_port = sock.getsockname()[1]
+                self._bound_evt.set()
+                self.print_info(f"listening on {self._host}:{self._bound_port}")
 
                 self.print_info(f"awaiting connection for {self._timeout} secs")
                 sock.settimeout(self._timeout)
@@ -382,6 +399,7 @@ class JsonRpcSrv(JsonRpcBase):
                         sleep(0.1)
 
         finally:
+            self._bound_evt.set()  # unblock wait_bound() even on failure
             if self._rpc is not None:
                 self._rpc.stop()
                 self._rpc.join()
@@ -407,35 +425,34 @@ class JsonRpcClient(JsonRpcBase):
         self.name = f"JsonRpcClt_{port}"
 
     def run(self):
+        try:
             if tm.OS() == "Windows":
                 self.run_win()
             else:
                 self.run_lin()
+        except Exception as e:
+            self.print_info(f"connection failed: {e}")
+        finally:
+            self._event_ready.set()  # settle wait_ready() whatever the outcome
 
     def run_win(self):
-        # TCP/IP socket creation
+        # Server already listening (handshake); retry on refused/timeout until deadline.
-        tslice = 1
+        deadline = monotonic() + self._timeout
-        t = self._timeout
         sock = None
         try:
-            while t >= 0:
+            while True:
                 sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
-                sock.settimeout(tslice)
+                sock.settimeout(0.5)
-                # Link of the socket at the configured port
                 try:
                     sock.connect((self._host, self._port))
                     break
-                except socket.timeout:
+                except OSError as e:
                     sock.close()
-                    t -= tslice
+                    if monotonic() >= deadline:
-                    if t < 0:
                         raise ETUMRuntimeError(
-                            f"{self.name}: failed to connect : timeout"
+                            f"{self.name}: failed to connect : {e}"
                         )
-                    else:
+                    sleep(0.1)
-                        sleep(tslice)
-                except socket.error as e:
-                    raise ETUMRuntimeError(f"{self.name}: failed to connect : {e}")
 
             self.print_info("Connected to server")
             self.connect(sock)

test/validation/items/console/test.tum

@@ -94,6 +94,17 @@
 {% endif %}
         - read_until: {expected: endOfCmd, timeout: 1, process_result: "'Hello' in r'''$(result)''' and 'PASS' in r'''$(result)''' "}
 
+{% if os == "Linux" %}
+- console:
+    name: Console runs on host (not the Flatpak sandbox)
+    doc: Regression guard for the 0.2.1 Flatpak bug (term console spawned inside the sandbox instead of on the host). /.flatpak-info exists only inside the sandbox, so the host-only marker is emitted (and matched by read_until) ONLY when the shell really runs on the host. On a broken Flatpak the marker never appears, read_until times out and the item FAILS. The marker is built at runtime ($M) so it is never present in the command line itself. Passes on every other channel.
+    console_name: term
+    key: $(test)_PASS
+    steps:
+        - writeln: 'test -e /.flatpak-info && M=SANDBOX || M=HOST; echo "console_host_check_$M"'
+        - read_until: {expected: console_host_check_HOST, timeout: 5}
+{% endif %}
+
 - console:
     name: Console closure
     execute_on_stop: true

test/validation/items/jsonrpc/test.tum

@@ -20,7 +20,7 @@
             console_name: jrpces
             key: $(test)_PASS
             steps:
-                - writeln: python3 {{include_directory}}/jrpc_echo_server.py -c {{include_directory}}/jrpces.ini
+                - writeln: '"$(python_bin)" {{include_directory}}/jrpc_echo_server.py -c {{include_directory}}/jrpces.ini'
                 - read_until: {expected: ready, timeout: 5}
 
         - console:

test/validation/post_execution.py

@@ -89,7 +89,7 @@ def exec():
         junit_report = report.replace(".sqlite", f"-{test}.xml")
         print(junit_report)
         _prepare_file_to_save(junit_report)
-        with open(junit_report, "w") as f:
+        with open(junit_report, "w", encoding="utf-8") as f:
             f.write(TestSuite.to_xml_string([ts]))
 
     # cleanup

test/validation/run.bat

@@ -89,6 +89,13 @@ REM Reports are stamped with the mode so successive runs don't clobber each othe
 
 SET "TAIL=-b -d "python_bin=%VENV_PYTHON%" -d "validation_report_file=validation-%MODE%" -- "%SCRIPT_DIR%\main.tum"%EXTRA%"
 
+REM The report-exporter plugin (items\report_plugin) is a pip entry-point
+REM package. It must live in the *testium* environment, so it is installed into
+REM the source/wheel venvs below. A frozen PyInstaller binary cannot see
+REM externally-installed plugins, so report_plugin is expected to be skipped
+REM there (same as Linux pyinstaller mode).
+SET "FAKE_EXPORTER=%SCRIPT_DIR%\fake_exporter"
+
 REM ---------- per-mode launcher ----------------------------------------------
 
 echo -- validation mode: %MODE%
@@ -100,8 +107,25 @@ echo ERROR: unknown --mode '%MODE%'. Expected: source ^| wheel ^| pyinstaller.
 exit /b 1
 
 :MODE_SOURCE
-call "%PROJECT_DIR%\run.bat" %TAIL%
+REM Run testium from src\ in a dedicated venv set up here. We do NOT delegate to
-exit /b %ERRORLEVEL%
+REM the project's run.bat: that one launches the GUI and does not forward its
+REM arguments, so the suite would never run head-less.
+SET "TESTIUM_VENV=%PROJECT_DIR%\test\tmp\testium_venv"
+IF NOT EXIST "%TESTIUM_VENV%" (
+    echo Creating testium venv at %TESTIUM_VENV%
+    %PYTHON_EXE% -m venv "%TESTIUM_VENV%"
+    IF !ERRORLEVEL! NEQ 0 (
+        echo ERROR while creating the testium venv.
+        exit /b 1
+    )
+    call "%TESTIUM_VENV%\Scripts\pip" install --quiet --upgrade pip
+    call "%TESTIUM_VENV%\Scripts\pip" install --quiet -r "%PROJECT_DIR%\src\requirements.txt"
+    REM language-server extra so `testium lsp` works from source (lsp_check.py)
+    call "%TESTIUM_VENV%\Scripts\pip" install --quiet "pygls>=1.3"
+)
+call "%TESTIUM_VENV%\Scripts\pip" install --quiet -e "%FAKE_EXPORTER%"
+SET CMD="%TESTIUM_VENV%\Scripts\python.exe" "%PROJECT_DIR%\src\testium"
+GOTO LAUNCH
 
 :MODE_WHEEL
 SET "WHEEL=%PROJECT_DIR%\dist\testium-%VERSION%-py3-none-any.whl"
@@ -115,10 +139,13 @@ IF NOT EXIST "%WHEEL_VENV%" (
     echo Creating wheel venv at %WHEEL_VENV%
     %PYTHON_EXE% -m venv --system-site-packages "%WHEEL_VENV%"
     call "%WHEEL_VENV%\Scripts\pip" install --quiet --upgrade pip
-    call "%WHEEL_VENV%\Scripts\pip" install --quiet "%WHEEL%"
+    REM install with the [lsp] extra so the wheel channel is validated in its
+    REM language-server-capable form (pulls pygls), matching `pip install testium[lsp]`.
+    call "%WHEEL_VENV%\Scripts\pip" install --quiet "%WHEEL%[lsp]"
 )
-"%WHEEL_VENV%\Scripts\python.exe" -m testium %TAIL%
+call "%WHEEL_VENV%\Scripts\pip" install --quiet -e "%FAKE_EXPORTER%"
-exit /b %ERRORLEVEL%
+SET CMD="%WHEEL_VENV%\Scripts\python.exe" -m testium
+GOTO LAUNCH
 
 :MODE_PYI
 SET "PYI_BIN=%PROJECT_DIR%\dist\testium-%VERSION%.exe"
@@ -127,5 +154,22 @@ IF NOT EXIST "%PYI_BIN%" (
     echo ERROR: PyInstaller binary not found in %PROJECT_DIR%\dist -- run build_all.sh first.
     exit /b 1
 )
-"%PYI_BIN%" %TAIL%
+SET CMD="%PYI_BIN%"
+GOTO LAUNCH
+
+REM ---------- launch ----------------------------------------------------------
+
+:LAUNCH
+echo -- launch: %CMD%
+
+REM LSP check (this exact channel): `schema` must keep its nested actions and
+REM `lsp` must answer initialize. Mirrors run.sh; aborts the run on failure.
+echo -- LSP check (%MODE%)
+"%VENV_PYTHON%" "%SCRIPT_DIR%\lsp_check.py" %CMD%
+IF !ERRORLEVEL! NEQ 0 (
+    echo ERROR: LSP check failed for mode %MODE%.
+    exit /b 1
+)
+
+%CMD% %TAIL%
 exit /b %ERRORLEVEL%